| |
Last updated on July 24, 2024. This conference program is tentative and subject to change
Technical Program for Wednesday July 17, 2024
|
WeM01 Regular Session, ISEC 136: Classroom |
Add to My Program |
Bayesian Methods |
|
|
Chair: Weyer, Erik | University of Melbourne |
|
10:00-10:20, Paper WeM01.1 | Add to My Program |
Kernel-Based Particle Filtering for Scalable Inference in Partially Observed Boolean Dynamical Systems |
|
Alali, Mohammad | Northeastern University |
Imani, Mahdi | Northeastern University |
Keywords: Parameter Estimation, Maximum Likelihood Methods, Biological Systems
Abstract: This paper addresses the inference challenges associated with a class of hidden Markov models with binary state variables, known as partially observed Boolean dynamical systems (POBDS). POBDS have demonstrated remarkable success in modeling the ON and OFF dynamics of genes, microbes, and bacteria in systems biology, as well as in network security to represent the propagation of attacks among interconnected elements. Despite existing optimal and approximate inference solutions for POBDS, scalability remains a significant issue due to the computational cost associated with likelihood evaluations and the exploration of extensive parameter spaces. To overcome these challenges, this paper proposes a kernel-based particle filtering approach for large-scale inference of POBDS. Our method employs a Gaussian process (GP) to efficiently represent the expensive-to-evaluate likelihood function across the parameter space. The likelihood evaluation is approximated using a particle filtering technique, enabling the GP to account for various sources of uncertainty, including limited likelihood evaluations. Leveraging the GP's predictive behavior, a Bayesian optimization strategy is derived for effectively seeking parameters yielding the highest likelihood, minimizing the overall computational burden while balancing exploration and exploitation. The proposed method's performance is demonstrated using two biological networks: the mammalian cell-cycle network and the T-cell large granular lymphocyte leukemia network.
|
|
10:20-10:40, Paper WeM01.2 | Add to My Program |
Bayesian System Identification of a River |
|
Jibran, Muhammad | The University of Melbourne |
Weyer, Erik | University of Melbourne |
Wu, Wenyan | The University of Melbourne |
Keywords: Bayesian Methods, Parameter Estimation, Identification for Control
Abstract: Hydrological systems like rivers and lakes are a vital part of any community. Mathematical models of such systems underpin management, decision making and control of rivers. There is always uncertainty associated with the models, and in this paper, we consider Bayesian system identification of a river. Bayesian system identification delivers an a-posteriori probability distribution of the unknown parameters. This uncertainty description is useful for solving chance-constrained problems encountered in the management of rivers. The Bayesian system identification approach is demonstrated by applying it to the upper part of the Murray River in Australia using real water level and flow measurements. The obtained models show good simulation performance capturing the observed water levels well. Moreover, posterior distributions of the parameters are delivered which are useful for control of rivers.
|
|
10:40-11:00, Paper WeM01.3 | Add to My Program |
Recursive Identification with Regularization and Online Hyperparameters Estimation |
|
Vau, Bernard | Ixblue |
Airimitoaie, Tudor-Bogdan | Univ. Bordeaux |
Keywords: Bayesian Methods, Recursive Identification, Regularization and Kernel Methods
Abstract: This paper presents a regularized recursive identification algorithm with simultane- ous on-line estimation of both the model parameters and the algorithms hyperparameters. A new kernel is proposed to facilitate the algorithm development. The performance of this novel scheme is compared with that of the recursive least-squares algorithm in simulation.
|
|
11:00-11:20, Paper WeM01.4 | Add to My Program |
Tensor Train Discrete Grid-Based Filters: Breaking the Curse of Dimensionality |
|
Matousek, Jakub | University of West Bohemia |
Brandner, Marek | University of West Bohemia |
Dunik, Jindrich | University of West Bohemia |
Puncochar, Ivo | University of West Bohemia |
Keywords: Bayesian Methods, Filtering and Smoothing
Abstract: This paper deals with the state estimation of stochastic systems and examines the possible employment of tensor decompositions in grid-based filtering routines, in particular, the tensor-train decomposition. The aim is to show that these techniques can lead to a massive reduction in both the computational and storage complexity of grid-based filtering algorithms without considerable tradeoffs in accuracy. This claim is supported by an algorithm descriptions and numerical illustrations.
|
|
11:20-11:40, Paper WeM01.5 | Add to My Program |
Gaussian Sum Filtering for Wiener State-Space Models with a Class of Non-Monotonic Piecewise Nonlinearities |
|
Cedeño, Angel L. | Universidad Técnica Federico Santa María |
González, Rodrigo A. | Eindhoven University of Technology |
Aguero, Juan C | Universidad Santa Maria |
Keywords: Filtering and Smoothing, Bayesian Methods, Particle Filtering/Monte Carlo Methods
Abstract: State estimation of nonlinear dynamical systems has gained significant attention due to its countless applications in control, signal processing, fault diagnosis, and power networks. The complexity posed by challenging nonlinearities like dead-zones, saturations, and linear rectification requires advanced state estimation. This paper presents a novel filtering technique designed for state-space Wiener systems encompassing these specific nonlinear behaviors. The filtering approach developed in this work introduces an explicit model for the probability function of the nonlinear output conditioned to the system state, which is derived from a Gaussian quadrature-based approximation. A Gaussian sum filtering algorithm is then used to obtain the filtering distributions and state estimates of systems with the aforementioned nonlinearities. Extensive numerical simulations are conducted to assess the accuracy of the proposed method compared to conventional techniques.
|
|
11:40-12:00, Paper WeM01.6 | Add to My Program |
A Kernel-Based PEM Estimator for Forward Models |
|
Fattore, Giulio | University of Padova |
Peruzzo, Marco | University of Padua |
Sartori, Giacomo | University of Padova, NTNU Trondheim |
Zorzi, Mattia | Università Degli Studi Di Padova |
Keywords: Regularization and Kernel Methods, Nonparametric Methods, Bayesian Methods
Abstract: This paper addresses the problem of learning the impulse responses characterizing forward models by means of a regularized kernel-based Prediction Error Method (PEM). The common approach to accomplish that is to approximate the system with a high-order stable ARX model. However, such choice induces a certain undesired prior information in the system that we want to estimate. To overcome this issue, we propose a new kernel-based paradigm which is formulated directly in terms of the impulse responses of the forward model and leading to the identification of a high-order ARMAX model. The most challenging step is the estimation of the kernel hyperparameters optimizing the marginal likelihood. The latter, indeed, does not admit a closed form expression. We propose a method for evaluating the marginal likelihood which makes possible the hyperparameters estimation. Finally, some numerical results showing the effectiveness of the method are presented.
|
|
WeM02 Regular Session, ISEC 138: Classroom |
Add to My Program |
Grey Box Modelling |
|
|
Chair: Mercère, Guillaume | Poitiers University |
|
10:00-10:20, Paper WeM02.1 | Add to My Program |
Regularized Iterative Weighted Total Least Squares for Vehicle Mass Estimation |
|
Koide, Hugo | University of Poitiers |
Mercère, Guillaume | Poitiers University |
Vayssettes, Jérémy | ISAE |
Keywords: Parameter Estimation, Errors in Variables Identification, Uncertainty Quantification
Abstract: This work addresses the challenge of vehicle mass estimation using a longitudinal vehicle dynamics model, which adopts an errors-in-variables formulation due to the presence of noise-contaminated measurements in both input and output variables. The reduced vehicle dynamics model is ill-conditioned by nature of correlated input variables and a lack of persistent excitation in the measured data. A regularized iterative weighted total least squares (RIWTLS) method is therefore developed and has the advantage of producing parameter uncertainty quantification and measurement bias estimation alongside the estimated system parameters. A complementary adaptive regularization scheme is developed and serves to control the numerical stability of the RIWTLS algorithm based on the conditioning of incoming data. Experimental tests using electric vehicle data and a batch estimation scheme highlight the performance of the proposed RIWTLS algorithm, estimating vehicle mass to within ±1% accuracy.
|
|
10:20-10:40, Paper WeM02.2 | Add to My Program |
Dynamics Modeling of Robot Joints with Asymmetric Load-Dependent Friction |
|
Graabæk, Søren | University of Southern Denmark |
Poulsen, Niels Kjølstad | Technical University of Denmark |
Sloth, Christoffer | Aalborg University |
Keywords: Grey Box Modelling, Nonlinear System Identification, Identification for Control
Abstract: This paper presents a strain wave gear model intended for industrial robots that are operated in both forward drive and backward drive under different load conditions. We show that a simple joint model captures the joint dynamics well at one load, but fails to capture the joint dynamics when the load is changed. Similarly, the simple model cannot capture the efficiency of the joint in both forward drive and backward drive. We propose a tooth model with sliding friction to resolve these issues. This model is simple to simulate and can thus be used for real-time control; the proposed model captures the load dependent friction and difference in efficiency in backward drive and forward drive. We validate the simulation performance of the tooth models on data from a Universal Robots UR10e robot with different payloads which shows an improvement compared to the standard rigid joint model.
|
|
10:40-11:00, Paper WeM02.3 | Add to My Program |
SINDy vs Hard Nonlinearities and Hidden Dynamics: A Benchmarking Study |
|
Raffa Ugolini, Aurelio | Politecnico Di Milano |
Breschi, Valentina | Eindhoven University of Technology |
Manzoni, Andrea | Politecnico Di Milano |
Tanelli, Mara | Politecnico Di Milano |
Keywords: Grey Box Modelling, Nonlinear System Identification, Machine Learning and Data Mining
Abstract: In this work we analyze the effectiveness of the Sparse Identification of Nonlinear Dynamics (SINDy) technique on three benchmark datasets for nonlinear identification, to provide a better understanding of its suitability as a modeling tool for real dynamical systems. While SINDy is often portrayed as an appealing strategy for pursuing physics-based learning, our analysis highlights two weaknesses, i.e., the difficulties in applying this technique when dealing with unobserved states and non-smooth dynamics. Due to the ubiquity of these features in real systems in general, and control applications in particular, we complement our analysis with hands-on approaches to tackle these issues.
|
|
11:00-11:20, Paper WeM02.4 | Add to My Program |
Grey-Box Modelling and Identification of the Industrial Oven of a Shrink Tunnel |
|
Previtali, Davide | University of Bergamo |
Pitturelli, Leandro | University of Bergamo |
Ferramosca, Antonio | Univeristy of Bergamo |
Previdi, Fabio | Universita' Degli Studi Di Bergamo |
Keywords: Grey Box Modelling, Parameter Estimation, Multivariable System Identification
Abstract: This paper presents a lumped-parameter grey-box sampled-data state-space model for the industrial oven of a shrink tunnel. The model is derived following the thermal-electrical analogy. A novel discretization strategy is developed to take into account that the sampling time of the system is equal to the lowest period of the pulse-width-modulated voltage signals which drive the heat resistors of the industrial oven. The model parameters are estimated by means of an extensive experimental campaign. Experimental results show that the derived model outperforms state-of-the-art transfer-function models while depending on fewer parameters.
|
|
11:20-11:40, Paper WeM02.5 | Add to My Program |
A Global Approach to Estimate Continuous-Time LPV Models for Wastewater Nitrification |
|
Boutourda, Fatima Zahra | Université De Poitiers |
Ouvrard, Régis | Université De Poitiers |
Poinot, Thierry | Université De Poitiers |
Mehdi, Driss | LIAS-ENSIP |
Mesquine, Fouad | Cadi Ayyad Univ |
De Tredern, Eloïse | Syndicat Interdépartemental Pour L'assainissement De L'aggloméra |
Jauzein, Vincent | Syndicat Interdépartemental Pour L'assainissement De L'aggloméra |
Keywords: Continuous Time System Estimation, Parameter Estimation, Other Applications
Abstract: In wastewater treatment, understanding and modeling the nitrification process is crucial to implement control. However, the complexity of this process makes it challenging to create simplified models. This study introduces an innovative method for estimating linear parameter-varying (LPV) models in the context of biological nitrification processes. The research focuses on the development of a continuous-time LPV model for a submerged aerated biofiltration system, considering the conversion of ammonium to nitrate in wastewater treatment. The methodology adopts the reinitialized partial moment approach within a global identification framework. The resultant LPV model is structured to capture the dynamics of the biological nitrification process, considering various factors like flow rates, feed concentrations and environmental regulations. Application of this approach to measured data from a wastewater treatment plant, demonstrates its effectiveness in accurately estimating the LPV model parameters. The results not only offer valuable insights into the dynamics and the nonlinear behaviour of the nitrification process but also contribute to the design and optimization of wastewater treatment plants, particularly those employing submerged aerated nitrifying biofilters.
|
|
11:40-12:00, Paper WeM02.6 | Add to My Program |
Regularized Finite Impulse Response Models versus Laguerre Models: A Comparison |
|
Illg, Christopher | University of Siegen |
Nelles, Oliver | University of Siegen |
Keywords: Regularization and Kernel Methods, Grey Box Modelling
Abstract: While in recent years, the estimation of finite impulse response (FIR) models has been improved by introducing new regularization schemes, also other orthonormal basis function (OBF) models are now becoming more prominent again. Although Laguerre models show very similar properties to the regularized FIR models, they are only rarely used for system identification. Therefore, in this paper, regularized FIR models and Laguerre models will be compared. This work focuses on the model structure and its similarities and differences, as well as the hyperparameter optimization utilizing the generalized cross-validation (GCV) error. Finally, the two model types are investigated using three different processes and the model performance is evaluated. Both model types show significant improvements compared to standard (unregularized) FIR models.
|
|
WeM03 Invited Session, ISEC 140: Classroom |
Add to My Program |
To Model or Not to Model? the Thin Line between Model-Based and Model-Free
Data-Driven Predictive Control |
|
|
Chair: Chiuso, Alessandro | University of Padova |
Co-Chair: Breschi, Valentina | Eindhoven University of Technology |
Organizer: Breschi, Valentina | Eindhoven University of Technology |
Organizer: Chiuso, Alessandro | University of Padova |
Organizer: Formentin, Simone | Politecnico Di Milano |
|
10:00-10:20, Paper WeM03.1 | Add to My Program |
Data-Driven Predictive Control and MPC: Do We Achieve Optimality? (I) |
|
Anand, Akhil | Norwegian University of Science and Technology (NTNU) |
Sawant, Shambhuraj | NTNU Trondheim |
Reinhardt, Dirk Peter | Norwegian University of Science and Technology |
Gros, Sebastien | NTNU |
Keywords: Data-driven Control, Identification for Control, Process Control
Abstract: In this paper, we explore the interplay between Predictive Control and closed-loop optimality, spanning from Model Predictive Control to Data-Driven Predictive Control. Predictive Control in general relies on some form of prediction scheme on the real system trajectories. However, these predictions may not accurately capture the real system dynamics, for e.g., due to stochasticity, resulting in sub-optimal control policies. This lack of optimality is a critical issue in case of problems with economic objectives. We address this by providing sufficient conditions on the underlying prediction scheme such that a Predictive Controller can achieve closed-loop optimality. However, these conditions do not readily extend to Data-Driven Predictive Control. In this context of closed-loop optimality, we conclude that the factor distinguishing the approaches within Data-Driven Predictive Control is if they can be cast as a sequential decision-making process or not, rather than the dichotomy of model-based vs. model-free. Furthermore, we show that the conventional approach of improving the prediction accuracy from data may not guarantee optimality.
|
|
10:20-10:40, Paper WeM03.2 | Add to My Program |
Harnessing the Final Control Error forOptimal Data-Driven Predictive Control (I) |
|
Chiuso, Alessandro | University of Padova |
Fabris, Marco | University of Padova |
Breschi, Valentina | Eindhoven University of Technology |
Formentin, Simone | Politecnico Di Milano |
Keywords: Data-driven Control, Regularization and Kernel Methods, Uncertainty Quantification
Abstract: Model Predictive Control (MPC) is a powerful method for complex system regulation, but its reliance on accurate models poses many limitations in real-world applications. Data-driven predictive control (DDPC) offers a valid alternative, eliminating the need for model identification. In this work, we present a unified stochastic framework for direct DDPC where control actions are obtained by optimizing the Final Control Error, directly computed from available data only, that automatically weighs the impact of uncertainty on the control objective. Our approach generalizes existing DDPC methods, like regularized Data-enabled Predictive Control (DeePC) and γ-DDPC, and thus provides a path toward noise-tolerant data-based control, with rigorous optimality guarantees. The theoretical investigation is complemented by a numerical case study, revealing that the proposed method consistently outperforms or, at worst, matches existing techniques without requiring any tuning effort
|
|
10:40-11:00, Paper WeM03.3 | Add to My Program |
Stochastic Data-Driven Predictive Control: Regularization, Estimation, and Constraint Tightening (I) |
|
Yin, Mingzhou | ETH Zurich |
Iannelli, Andrea | University of Stuttgart |
Smith, Roy S. | Swiss Federal Institute of Technology (ETH) |
Keywords: Data-driven Control, Nonparametric Methods, Filtering and Smoothing
Abstract: Data-driven predictive control methods based on the Willems' fundamental lemma have shown great success in recent years. These approaches use receding horizon predictive control with nonparametric data-driven predictors instead of model-based predictors. This study addresses three problems of applying such algorithms under unbounded stochastic uncertainties: 1) tuning-free regularizer design, 2) initial condition estimation, and 3) reliable constraint satisfaction, by using stochastic prediction error quantification. The regularizer is designed by leveraging the expected output cost. An initial condition estimator is proposed by filtering the measurements with the one-step-ahead stochastic data-driven prediction. A novel constraint-tightening method, using second-order cone constraints, is presented to ensure high-probability chance constraint satisfaction. Numerical results demonstrate that the proposed methods lead to satisfactory control performance in terms of both control cost and constraint satisfaction, with significantly improved initial condition estimation.
|
|
11:00-11:20, Paper WeM03.4 | Add to My Program |
A Bias-Variance Perspective of Data-Driven Control (I) |
|
Colin, Kévin | KTH Royal Institute of Technology |
Ju, Yue | The Chinese University of Hong Kong, Shenzhen, China |
Bombois, Xavier | Ecole Centrale De Lyon |
Rojas, Cristian R. | KTH Royal Institute of Technology |
Hjalmarsson, Håkan | KTH |
Keywords: Data-driven Control, Regularization and Kernel Methods
Abstract: Data-driven control, the task of designing a controller based on process data, finds application in a wide range of disciplines and the topic has been intensively studied over more than half a century. The main purpose of this contribution is to elucidate on the commonalities between data-driven control and parameter estimation. In particular, we discuss the bias-variance trade-off, i.e. rather than aiming for the optimal controller one should aim for a constrained version, that may be characterized by tunable parameters, corresponding to hyperparameters in parameter estimation. As a result we shift attention from indirect vs direct data driven control by highlighting the important role played by (complete) minimal sufficient statistics. To keep technicalities at a minimum, still capturing the essential features of the problem, we consider the problem of minimizing the expected control cost for a quadratic open loop control problem applied to a finite impulse response system. In a Gaussian white noise setting, the maximum-likelihood parameter estimate constitutes a complete minimal sufficient statistic which allows us to focus on controllers that are functions of this model estimate without loss of statistical accuracy. We make a systematic study of three different controller structures and two different tuning techniques and illustrate their behaviours numerically
|
|
11:20-11:40, Paper WeM03.5 | Add to My Program |
Mind the Gap: The Role of Metrics in Data-Driven Modelling (I) |
|
Padoan, Alberto | ETH Zurich |
Coulson, Jeremy | University of Wisconsin-Madison |
Keywords: Subspace Methods, Data-driven Control
Abstract: Distances are naturally used to quantify misfits in modelling problems, errors in approximation problems, and objectives in optimal control problems. The formulation of such problems usually relies on distance functions between vector quantities, which depend on the particular application. This, however, generates issues if the quantities involved are intrinsically subspaces of possibly different dimensions, which motivates the need for dedicated distances. We propose to resolve these issues by adopting a geometric approach and regarding subspaces as points on the manifold of all subspaces of all dimensions. This approach not only provides a solid basis for optimization over finite-horizon linear time-invariant behaviors, but also opens the door to solving new interpolation and approximation problems.
|
|
11:40-12:00, Paper WeM03.6 | Add to My Program |
Neural Data-Enabled Predictive Control (I) |
|
Lazar, Mircea | Eindhoven Univ. of Technology |
Keywords: Data-driven Control, Neural Networks, Basis Functions
Abstract: Data-enabled predictive control (DeePC) for linear systems utilizes data matrices of recorded trajectories to directly predict new system trajectories, which is very appealing for real-life applications. In this paper we leverage the universal approximation properties of neural networks (NNs) to develop neural DeePC algorithms for nonlinear systems. Firstly, we point out that the outputs of the last hidden layer of a deep NN implicitly construct a basis in a so-called neural (feature) space, while the output linear layer performs affine interpolation in the neural space. As such, we can train off-line a deep NN using large data sets of trajectories to learn the neural basis and compute on-line a suitable affine interpolation using DeePC. Secondly, methods for guaranteeing consistency of neural DeePC and for reducing computational complexity are developed. Several neural DeePC formulations are illustrated on a nonlinear pendulum example.
|
|
WeM04 Regular Session, ISEC 142: Classroom |
Add to My Program |
Frequency Domain Identification |
|
|
Chair: Abdalmoaty, Mohamed Rasheed Hilmy | ETH Zurich |
Co-Chair: Schoukens, Maarten | Eindhoven University of Technology |
|
10:00-10:20, Paper WeM04.1 | Add to My Program |
Parametric Estimation of Arbitrary Fractional Order Models for Battery Impedances |
|
Vandeputte, Freja | Vrije Universiteit Brussel |
Hallemans, Noël | University of Oxford |
Lataire, John | Vrije Universiteit Brussel |
Keywords: Frequency Domain Identification
Abstract: Electrochemical impedance spectroscopy (EIS) is a widely-used non-invasive technique for estimating the impedance of a battery from current and voltage measurements. While EIS is commonly used as a nonparametric, purely data-driven estimation method, this article proposes a parametric, physics-informed alternative. As an underlying parametric model, we use an equivalent circuit model for the battery impedance with a Warburg element to model the low-frequency diffusion. This fractional order impedance model is linear in all the parameters except one, namely the fractional order itself. Hence, we present a separable total least squares estimator, which first eliminates the linear parameters using their total least squares solution, and then minimises the resulting nonlinear least squares problem over the fractional order. Measuring multiple periods of the signals allows to weigh the problem with the noise variances, thus making the estimation consistent. The parametric estimation method is validated on simulations and applied to measurement data of commercial Samsung 48X cells.
|
|
10:20-10:40, Paper WeM04.2 | Add to My Program |
Nonparametric Frequency-Domain Identification of Magnetic Bearings: An Experimental Study |
|
Schietecat, Mathias | KU Leuven |
Jacobs, Laurens | KU Leuven |
Swevers, Jan | KU Leuven R&D |
Keywords: Frequency Domain Identification, Multivariable System Identification, Nonparametric Methods
Abstract: This paper reports on an experimental study on the key factors affecting the quality and time-efficiency of MIMO frequency response function measurements for an industrial test rig equipped with active magnetic bearings. The influence of several design choices are explored, including the input power and the choice of excitation signal. In particular, different methodologies relying on multisines and pseudo random binary sequences are compared. The accuracy of the resulting FRFs is analyzed through formal measures such as the covariances of the estimated FRFs. Additionally, the suitability of the different approaches for analyzing the nonlinearity of the plant is assessed.
|
|
10:40-11:00, Paper WeM04.3 | Add to My Program |
Time-Domain versus Frequency-Domain System Identification of Lithium-Ion Batteries Using Fractional Models |
|
Adel, Abderrahmane | University of Bordeaux |
Malti, Rachid | ICFDA 2024 |
Vinassa, Jean-Michel | Univ. Bordeaux, CNRS, Bordeaux INP, IMS UMR 5218 |
Briat, Olivier | University of Bordeaux |
Keywords: Frequency Domain Identification, Other Applications
Abstract: This study presents a comparison between time-domain and frequency-domain approaches for system identification of Lithium-ion batteries using fractional models. As reported in the literature, time-domain approaches have the advantage of collecting data on smaller time-intervals (few seconds) as compared to frequency-domain data (several minutes). The model is derived using a fractional-order equivalent circuit model (FO-ECM) which exhibit long memory diffusion phenomena, and its impedance parameters are determined through the application of Levenberg-Marquardt optimization algorithm both in time and frequency domains. Comparisons are conducted on experimental data generated from three commercial batterie cells with different chemistries and cross-validation results are shown between the obtained time-domain and frequency-domain models. Similar models found with both methods which affirm the reliability and applicability of the proposed identification methods both in time and frequency domains.
|
|
11:00-11:20, Paper WeM04.4 | Add to My Program |
A Local Gaussian Process Regression Approach to Frequency Response Function Estimation |
|
Fang, Xiaozhu | The Chinese University of Hong Kong, Shenzhen |
Xu, Yu | The Chinese University of Hong Kong, Shenzhen |
Chen, Tianshi | The Chinese University of Hong Kong, Shenzhen, 518172, China |
Keywords: Frequency Domain Identification, Regularization and Kernel Methods, Bayesian Methods
Abstract: Frequency response function (FRF) estimation is a classical subject in system identification. In the past two decades, there have been remarkable advances in developing local methods for this subject, e.g., the local polynomial method, local rational method, and iterative local rational method. The recent concentrations for local methods are two issues: the model order selection and the identification of lightly damped systems. To address these two issues, we propose a new local method called local Gaussian process regression (LGPR). We show that the frequency response function locally is either analytic or resonant, and this prior knowledge can be embedded into a kernel-based regularized estimate through a dot-product kernel plus a resonance kernel induced by a second-order resonant system. The LGPR provides a new route to tackle the aforementioned issues. In the numerical simulations, the LGPR shows the best FRF estimation accuracy compared with the existing local methods, and moreover, the LGPR is more robust with respect to sample size and noise level.
|
|
11:20-11:40, Paper WeM04.5 | Add to My Program |
Frequency-Domain Identification of Discrete-Time Systems Using Sum-Of-Rational Optimization |
|
Abdalmoaty, Mohamed Rasheed Hilmy | ETH Zurich |
Miller, Jared | ETH Zurich |
Yin, Mingzhou | ETH Zurich |
Smith, Roy S. | Swiss Federal Institute of Technology (ETH) |
Keywords: Parameter Estimation, Frequency Domain Identification
Abstract: This paper proposes a new computationally tractable method to fit coefficients of a fixed-order discrete-time transfer function to a measured frequency response, with stability guaranteed. The problem is formulated as a non-convex global sum-of-rational optimization problem whose objective function is the sum of weighted squared residuals at each observed frequency datapoint. Stability is enforced using a polynomial matrix inequality constraint. The problem is solved by a moment-sum-of-squares hierarchy of semidefinite programs through a framework for sum-of-rational-functions optimization. Convergence of the moment-sum-of-squares program is guaranteed as the bound on the degree of the sum-of-squares polynomials approaches infinity. The performance of the proposed method is demonstrated using numerical simulation examples.
|
|
11:40-12:00, Paper WeM04.6 | Add to My Program |
An Efficient Implementation for Regularized Frequency Response Function and Transient Estimation |
|
Xu, Yu | The Chinese University of Hong Kong, Shenzhen |
Fang, Xiaozhu | The Chinese University of Hong Kong, Shenzhen |
Mu, Biqiang | AMSS, CAS |
Chen, Tianshi | The Chinese University of Hong Kong, Shenzhen, 518172, China |
Keywords: Regularization and Kernel Methods, Frequency Domain Identification, Bayesian Methods
Abstract: In this paper, we propose an efficient implementation for the kernel-based regularized frequency response function (FRF) and transient estimation. In particular, we show that when the diagonal correlated (DC) kernel is used, the output kernel matrix in the frequency domain has a hierarchically off-diagonal low-rank (HODLR) structure, and by exploring this structure, the FRF and transient estimates can be computed very efficiently. The overall computational complexity is O(γ2Nb log2 Nb), where Nb is the number of data in the interested frequency band and γ is the HODLR rank of the output kernel matrix that is often much smaller than Nb. The efficacy of the proposed implementation is demonstrated by numerical simulations.
|
|
WeA101 Regular Session, ISEC 136: Classroom |
Add to My Program |
Estimation |
|
|
Chair: Garulli, Andrea | Universita' Di Siena |
Co-Chair: Forgione, Marco | SUPSI-USI |
|
14:00-14:20, Paper WeA101.1 | Add to My Program |
A Comparison of Indirect and Direct Filter Designs from Data for LTI Systems: The Effect of Unknown Noise Covariance Matrices |
|
Mazzoleni, Mirko | University of Bergamo |
Maurelli, Luca | University of Bergamo |
Formentin, Simone | Politecnico Di Milano |
Previdi, Fabio | Universita' Degli Studi Di Bergamo |
Keywords: Filtering and Smoothing, Parameter Estimation
Abstract: Existing literature on model-based filter design for stochastic LTI systems assumes complete correspondence between the system and its model. When the system is not completely known, the standard indirect model-based (two-steps) filtering solution consists of: (i) identify a model of the system from measured input/output data; (ii) design a Kalman filter based on the estimated model. The performance of this indirect approach are limited by the model and noise covariance matrices accuracy. To overcome such limitations, this paper investigates a direct (one-step) solution to the filtering problem for SISO LTI systems in the Prediction Error Method (PEM) identification framework. Simulation results indicate the effectiveness of the direct filtering approach, especially when the noise covariance matrices are misspecified.
|
|
14:20-14:40, Paper WeA101.2 | Add to My Program |
Data-Augmented Numerical Integration in State Prediction: Rule Selection |
|
Dunik, Jindrich | University of West Bohemia |
Kral, Ladislav | University of West Bohemia |
Matousek, Jakub | University of West Bohemia |
Straka, Ondrej | University of West Bohemia |
Brandner, Marek | University of West Bohemia |
Keywords: Filtering and Smoothing, Particle Filtering/Monte Carlo Methods, Neural Networks
Abstract: This paper deals with the state prediction of nonlinear stochastic dynamic systems. The emphasis is laid on a solution to the integral Chapman-Kolmogorov equation by a deterministic-integration-rule-based point-mass method. A novel concept of reliable data- augmented, i.e., mathematics- and data-informed, integration rule is developed to enhance the point-mass state predictor, where the trained neural network (representing data contribution) is used for the selection of the best integration rule from a set of available rules (representing mathematics contribution). The proposed approach combining the best properties of the standard mathematics-informed and novel data-informed rules is thoroughly discussed.
|
|
14:40-15:00, Paper WeA101.3 | Add to My Program |
In-Context Learning of State Estimators |
|
Busetto, Riccardo | Politecnico Di Milano |
Breschi, Valentina | Eindhoven University of Technology |
Forgione, Marco | SUPSI-USI |
Piga, Dario | SUPSI-USI |
Formentin, Simone | Politecnico Di Milano |
Keywords: Machine Learning and Data Mining, Filtering and Smoothing
Abstract: State estimation has a pivotal role in several applications, including but not limited to advanced control design. Especially when dealing with nonlinear systems state estimation is a nontrivial task, often entailing approximations and challenging fine-tuning phases. In this work, we propose to overcome these challenges by formulating an in-context state-estimation problem, enabling us to learn a state estimator for a class of (nonlinear) systems abstracting from particular instances of the state seen during training. To this end, we extend an in-context earning framework recently proposed for system identification, showing via a benchmark numerical example that this approach allows us to (i) use training data directly for the design of the state estimator, (ii) not requiring extensive fine-tuning procedures, while (iii) achieving superior performance compared to state-of-the-art benchmarks.
|
|
15:00-15:20, Paper WeA101.4 | Add to My Program |
Li-Ion Cell Impedance Identification in the Time Domain As an Alternative to Identification in the Frequency Domain |
|
Arahbi, Omar | Université De Poitiers |
Huard, Benoît | Université De Poitiers |
Gabano, Jean-Denis | Université De Poitiers |
Poinot, Thierry | Université De Poitiers |
Keywords: Nonlinear System Identification, Parameter Estimation, Frequency Domain Identification
Abstract: Electrochemical Impedance Spectroscopy (EIS) is a widely used tool for selecting a pertinent Equivalent Circuit Model (ECM) of Li-ion cells which is characterized by non integer order operators. The main drawback of EIS is the long time required to scan a whole spectrum down to low frequencies. Thus, chronopotentiometry (CP) is an alternative method which consists in identifying impedance parameters using an excitation current sequence lasting a few seconds and the corresponding induced voltage variations. An impedance fractional model needs to be synthesized in order to allow simulation also in the time domain. The pertinence of CP is demonstrated using experimental results obtained with a Samsung 3.4 Ah Li-ion cell. Due to its inability to take into account the inductive behavior, fractional model used in the time domain exhibit limitations. Nevertherless, parameters identified in the time domain present similar results to the ones identified in the frequency domain under the condition of a restricted frequency range.
|
|
15:20-15:40, Paper WeA101.5 | Add to My Program |
Set Membership State Estimation with Quantized Measurements and Optimal Threshold Selection |
|
Casini, Marco | Universita' Di Siena |
Garulli, Andrea | Universita' Di Siena |
Vicino, Antonio | Universita' Di Siena |
Keywords: Uncertainty Quantification, Bounded Error Identification, Filtering and Smoothing
Abstract: The problem of state estimation with quantized measurements is addressed in the set membership estimation setting. The main contribution concerns the optimal selection of the quantizer thresholds in order to minimize the worst-case radius of the feasible state set. This allows one to design adaptive quantizers reducing the uncertainty associated to the state estimates. The proposed solution is applied to several outer approximations of the feasible sets, based on parallelotopes, zonotopes and constrained zonotopes. The benefits of the threshold selection mechanism are assessed on a numerical example, highlighting the trade off between computational burden and uncertainty reduction.
|
|
15:40-16:00, Paper WeA101.6 | Add to My Program |
Lasso-Based State Estimation for Cyber-Physical Systems under Sensor Attacks |
|
Cerone, Vito | Politecnico Di Torino |
Fosson, Sophie M. | Politecnico Di Torino |
Regruto, Diego | Politecnico Di Torino |
Ripa, Francesco | Politecnico Di Torino |
Keywords: Parameter Estimation, Recursive Identification, Algorithms
Abstract: The development of algorithms for secure state estimation in vulnerable cyber-physical systems has been gaining attention in the last years. A consolidated assumption is that an adversary can tamper a relatively small number of sensors. In the literature, block-sparsity methods exploit this prior information to recover the attack locations and the state of the system. In this paper, we propose an alternative, Lasso-based approach and we analyse its effectiveness. In particular, we theoretically derive conditions that guarantee successful attack/state recovery, independently of established time sparsity patterns. Furthermore, we develop a sparse state observer, by starting from the iterative soft thresholding algorithm for Lasso, to perform online estimation. Through several numerical experiments, we compare the proposed methods to the state-of-the-art algorithms.
|
|
WeA102 Regular Session, ISEC 138: Classroom |
Add to My Program |
Recent Advances in System Identification Theory and Implementation - Part I |
|
|
Chair: Weyer, Erik | University of Melbourne |
Co-Chair: Mazzoleni, Mirko | University of Bergamo |
|
14:00-14:20, Paper WeA102.1 | Add to My Program |
A Weighted Least-Squares Method for Non-Asymptotic Identification of Markov Parameters from Multiple Trajectories |
|
He, Jiabao | KTH Royal Institute of Technology |
Rojas, Cristian R. | KTH Royal Institute of Technology |
Hjalmarsson, Håkan | KTH |
Keywords: Bounded Error Identification
Abstract: Markov parameters play a key role in system identification. There exists many algorithms where these parameters are estimated using least-squares in a first, pre-processing, step, including subspace identification and multi-step least-squares algorithms, such as Weighted Null-Space Fitting. Recently, there has been an increasing interest in non-asymptotic analysis of estimation algorithms. In this contribution we identify the Markov parameters using weighted least-squares and present non-asymptotic analysis for such estimator. To cover both stable and unstable systems, multiple trajectories are collected. We show that with the optimal weighting matrix, weighted least-squares gives a tighter error bound than ordinary least-squares for the case of non-uniformly distributed measurement errors. Moreover, as the optimal weighting matrix depends on the system's true parameters, we introduce two methods to consistently estimate the optimal weighting matrix, where the convergence rate of these estimates is also provided. Numerical experiments demonstrate improvements of weighted least-squares over ordinary least-squares in finite sample settings.
|
|
14:20-14:40, Paper WeA102.2 | Add to My Program |
Continuous-Time Identification of Grey-Box and Black-Box Models of an Industrial Oven |
|
Previtali, Davide | University of Bergamo |
Scandella, Matteo | University of Bergamo |
Pitturelli, Leandro | University of Bergamo |
Mazzoleni, Mirko | University of Bergamo |
Ferramosca, Antonio | Univeristy of Bergamo |
Previdi, Fabio | Universita' Degli Studi Di Bergamo |
Keywords: Continuous Time System Estimation, Multivariable System Identification, Nonparametric Methods
Abstract: The design of temperature controllers is impaired by the limited accuracy of the models employed for thermal systems, which are commonly estimated from uninformative data, such as step responses, due to the restrictive experimental design connected to the long duration of the experiments. This paper focuses on modelling an industrial convection oven following different rationales. Three continuous-time models are proposed and compared: a grey-box parametric thermal network model, a black-box parametric first order lag plus time delay model, and a black-box non-parametric model based on reproducing kernel Hilbert spaces. These are all estimated and validated on step response experimental data. Lastly, the pros and cons of each model are highlighted.
|
|
14:40-15:00, Paper WeA102.3 | Add to My Program |
Generalized Performance Criteria for Identified Models |
|
Vau, Bernard | Ixblue |
Bourles, Henri | ENS De Cachan |
Keywords: Model Validation
Abstract: It is shown that some usual criteria evaluating the performances of an identified model with respect to experimental data, like the FIT criterion, can be not well-suited to fast sampled systems. This leads us to propose some generalized criteria where the signals are filtered by transfer functions belonging to an orthonormal basis. An interpretation of this filtering in the frequency domain is proposed. The basis poles selection is equivalent to making a specification about the criteria in function of the expected use of the identified model.
|
|
15:00-15:20, Paper WeA102.4 | Add to My Program |
Differentiable Multi-Ridge Regression for System Identification |
|
Maroni, Gabriele | IDSIA USI-SUPSI |
Cannelli, Loris | SUPSI University |
Piga, Dario | SUPSI-USI |
Keywords: Model Validation, Parameter Estimation
Abstract: Regularization aims to shrink model parameters, reducing complexity and overfitting risk. Traditional methods like LASSO and Ridge regression, limited by a single regularization hyperparameter, can restrict bias-variance trade-off adaptability. This paper addresses system identification in a multi-ridge regression framework, where an l2-penalty on the model coefficients is introduced, and a different regularization hyperparameter is assigned to each model parameter. To compute the optimal hyperparameters, a cross-validation-based criterion is optimized through gradient descent. Autoregressive and Output Error models are considered. The former requires formulating a regularized least-squares problem. The identification of the latter class is more challenging and is addressed by adopting regularized instrumental variable methods to ensure a consistent parameter estimation.
|
|
15:20-15:40, Paper WeA102.5 | Add to My Program |
A Sample Based Algorithm for Constructing Guaranteed Confidence Ellipsoids for Linear Regression Models with Deterministic Regressor |
|
Wang, Xiaopuwen | Chengdu Technological University |
Weyer, Erik | University of Melbourne |
Keywords: Parameter Estimation, Uncertainty Quantification
Abstract: This paper presents a novel algorithm for constructing non-asymptotic confidence ellipsoids for linear regression models with deterministic regressors. The confidence ellipsoids are centered at the least-squares estimate, and the true parameter lies within the ellipsoid with a probability specified by the user. The confidence ellipsoid is constructed by drawing independent samples of the noise sequence and exploiting an ordering property for independent and identically distributed random variables. The main assumption on the noise is that it has a known joint density so that independent samples can be drawn. The efficacy of the approach is demonstrated through a simulation example of an FIR system. The constructed confidence ellipsoid is compared with the confidence regions obtained using the Sign-Perturbed Sum method and asymptotic system identification theory.
|
|
15:40-16:00, Paper WeA102.6 | Add to My Program |
Identification of Low Order Systems in a Loewner Framework |
|
Honarpisheh, Arya | Northeastern University |
Singh, Rajiv | Northeastern University |
Miller, Jared | ETH Zurich |
Sznaier, Mario | Northeastern University |
Keywords: Reduced-order Modeling, Subspace Methods, Bounded Error Identification
Abstract: This paper considers the problem of non-parametric identification of low order models from time-domain experimental data using a combination of Caratheodory Fejer and Loewner based interpolation, followed by a Loewner matrix Balanced Reduction (LBR) step. As we show in the paper, the Loewner matrix is an estimator for the trace norm of a system, playing a role similar to the one played by the Hankel matrix. However, utilizing Zolotarev numbers to establish decay rate bounds for singular values reveals that the decay of singular values in the Loewner matrix is considerably faster than that in the Hankel matrix. Thus, Loewner based methods yield lower order systems, with the same error bound, than comparable ones based on Hankel matrices. The effectiveness of our method is demonstrated through a numerical example.
|
|
WeA103 Regular Session, ISEC 140: Classroom |
Add to My Program |
Data-Driven Control |
|
|
Chair: Formentin, Simone | Politecnico Di Milano |
Co-Chair: Novara, Carlo | Politecnico Di Torino |
|
14:00-14:20, Paper WeA103.1 | Add to My Program |
Data-Driven Control of Input Saturated Systems: A LMI-Based Approach |
|
Porcari, Federico | Politecnico Di Milano |
Breschi, Valentina | Eindhoven University of Technology |
Zaccarian, Luca | LAAS-CNRS and University of Trento |
Formentin, Simone | Politecnico Di Milano |
Keywords: Data-driven Control
Abstract: This paper addresses three complex control challenges related to input-saturated systems from a data-driven perspective. Unlike the traditional two-stage process involving system identification and model-based control, the proposed approach eliminates the need for an explicit model description. The method combines data-based closed-loop representations, Lyapunov theory, instrumental variables, and a generalized sector condition to formulate data-driven linear matrix inequalities (LMIs). These LMIs are applied to maximize the origin’s basin of attraction, minimize the closed-loop reachable set with bounded disturbances, and introduce a new data-driven l2-gain minimization problem. Demonstrations on benchmark examples highlight the advantages and limitations of the proposed approach compared to an explicit identification of the system, emphasizing notable benefits in handling nonlinear dynamics.
|
|
14:20-14:40, Paper WeA103.2 | Add to My Program |
Iterative Feedback Tuning with Automated Reference Model Selection |
|
Ickenroth, Tjeerd | Eindhoven University of Technology |
Breschi, Valentina | Eindhoven University of Technology |
Oomen, Tom | Eindhoven University of Technology |
Formentin, Simone | Politecnico Di Milano |
Keywords: Data-driven Control
Abstract: Iterative Feedback Tuning (IFT) is a direct, data-driven control technique, that relies on a reference model to capture the desired behavior of the unknown system. The choice of this hyper-parameter is particularly critical, as it potentially jeopardizes performance and even closed-loop stability. This paper aims to explore the suitability of three search methods (grid search, random search, and successive halving) to automatically tune the reference model from data based on a set of user-defined, soft specifications on the desired closed-loop behavior. To compare the three methods and demonstrate their effectiveness, we consider a benchmark simulation case study on the control of a mass-spring-damper system. From our results, successive halving turns out to be the most efficient way to run IFT with automatic reference model selection on a finite budget of time for data collection.
|
|
14:40-15:00, Paper WeA103.3 | Add to My Program |
Set Membership Identification for NMPC Complexity Reduction |
|
Boggio, Mattia | Politecnico Di Torino |
Novara, Carlo | Politecnico Di Torino |
Taragna, Michele | Politecnico Di Torino |
Keywords: Data-driven Control, Nonlinear System Identification, Automotive Systems
Abstract: A Set Membership (SM) approach is proposed to reduce the computational burden of Nonlinear Model Predictive Control (NMPC) algorithms. In particular, a SM identification method is applied to derive an approximation and tight bounds of the NMPC control law, using a set of its values computed offline. These quantities are used online to reduce the dimension and the volume of the search domain of the NMPC optimization algorithm, and to perform a warm start, allowing a significant shortening of the computational time. The developed NMPC methodology is tested in simulation, considering an obstacle avoidance application in a realistic autonomous vehicle scenario. The obtained results demonstrate the effectiveness of the proposed approach in terms of computation time, without affecting the solution quality.
|
|
15:00-15:20, Paper WeA103.4 | Add to My Program |
Online Learning and Control for Data-Augmented Quadrotor Model |
|
Šmíd, Matěj | CTU |
Dunik, Jindrich | University of West Bohemia |
Keywords: Data-driven Control, Identification for Control, Algorithms
Abstract: The ability to adapt to changing conditions is a key feature of a successful autonomous system. In this work, we use the Recursive Gaussian Processes (RGP) for identification of the quadrotor air drag model online, without the need to precollect training data. The identified drag model then augments a physics-based model of the quadrotor dynamics, which allows more accurate quadrotor state prediction with increased ability to adapt to changing conditions. This data-augmented physics-based model is utilized for precise quadrotor trajectory tracking using the suitably modified Model Predictive Control (MPC) algorithm. The proposed modelling and control approach is evaluated using the Gazebo simulator and it is shown that the proposed approach tracks a desired trajectory with a higher accuracy compared to the MPC with the non-augmented (purely physics-based) model.
|
|
15:20-15:40, Paper WeA103.5 | Add to My Program |
Data-Driven Identification of Quadrotor Dynamics: A Tutorial |
|
Wi, Yejin | University of Houston |
Cescon, Marzia | University of Houston |
Keywords: Identification for Control, Mechanical and Aerospace, Closed Loop Identification
Abstract: In this work, we provide a tutorial-style exposition of quadrotor dynamics model identification from experimental data collected in closed-loop. Our objective is to provide guidelines for the scientist who is approaching model-based flight control design for multirotors UAVs for which a model is not readily available, but also to provide instructors with pedagogical material which can be utilized to develop experiential learning opportunities complementing face-to-face lectures in the system identification and control engineering curricula.
|
|
15:40-16:00, Paper WeA103.6 | Add to My Program |
Virtual Reference Feedback Tuning Revisited |
|
Scampicchio, Anna | ETH Zürich |
Breschi, Valentina | Eindhoven University of Technology |
Keywords: Data-driven Control, Bayesian Methods, Parameter Estimation
Abstract: We focus on the Virtual Reference Feedback Tuning (VRFT) approach, in which control design is rephrased as a system identification problem. We propose a new formulation of the VRFT optimization program and derive two algorithms to find its solution. The first consists of a deterministic, alternating minimization scheme; the second leverages the Bayesian interpretation of the novel formulation and relies on Markov Chain Monte Carlo. We carry out a theoretical analysis of the proposed strategy, using its deterministic view to derive consistency results, and the Bayesian one to improve performance in the non-asymptotic regime. The effectiveness of the proposed approaches is then tested on a benchmark system in the literature.
|
|
WeA104 Regular Session, ISEC 142: Classroom |
Add to My Program |
Neural Networks |
|
|
Chair: Schoukens, Maarten | Eindhoven University of Technology |
Co-Chair: Cescon, Marzia | University of Houston |
|
14:00-14:20, Paper WeA104.1 | Add to My Program |
Physiology-Informed Deep Learning Modeling of Type 1 Diabetes Dynamics: Mapping Data to Virtual Subjects |
|
Crespo-Santiago, Alvaro | University of Houston |
Cescon, Marzia | University of Houston |
Keywords: Biomedical Systems, Machine Learning and Data Mining, Neural Networks
Abstract: We consider the problem of creating Digital Twins (DTs) of glucose metabolism in people with Type 1 Diabetes (T1D), coupling a large-scale mean population metabolic model of glucose dynamics developed from first principles, with deep learning (DL) architectures, to map actual patient data to their virtual counterparts. For the neural network component of our proposed strategy, two models were investigated: a Long Short-Term Memory (LSTM) network and a Generative Adversarial Network (GAN). Our best model outperformed significantly the mean population model with respect to evaluation metrics (LSTM vs. metabolic simulator), expressed as median (interquartile range): MAE 35.0 (28.8, 43.8) vs. 79.7 (62.4, 115.5) [mg/dL], RMSE 44.8 (37.2, 56.2) vs. 94.9 (76.3, 128.0) [mg/dL], eNRMSE 0.85 (0.83, 0.88) vs. 0.54 (0.48, 0.72), eFIT 0.25 (0.13, 0.40) vs. 0.00 (0.00, 0.00). We showed that the proposed physiology-informed deep learning approach successfully mapped real patient data to virtual subjects, with the potential to enable in-silico testing of novel therapeutic strategies on a virtual population.
|
|
14:20-14:40, Paper WeA104.2 | Add to My Program |
Split-Boost Neural Networks |
|
Cestari, Raffaele Giuseppe | Politecnico of Milan |
Maroni, Gabriele | IDSIA USI-SUPSI |
Cannelli, Loris | SUPSI University |
Piga, Dario | SUPSI-USI |
Formentin, Simone | Politecnico Di Milano |
Keywords: Neural Networks, Machine Learning and Data Mining, Algorithms
Abstract: The calibration and training of a neural network is a complex and time-consuming procedure that requires significant computational resources to achieve satisfactory results. Key obstacles are a large number of hyperparameters to select and the onset of overfitting in the face of a small amount of data. In this framework, we propose an innovative training strategy for feed-forward architectures - called split-boost - that improves performance and automatically includes a regularizing behaviour without modeling it explicitly. Such a novel approach ultimately allows us to avoid explicitly modeling the regularization term, decreasing the total number of hyperparameters and speeding up the tuning phase. The proposed strategy is tested on a real-world (anonymized) dataset within a benchmark medical insurance design problem.
|
|
14:40-15:00, Paper WeA104.3 | Add to My Program |
Structured State-Space Models Are Deep Wiener Models |
|
Bonassi, Fabio | Uppsala University |
Andersson, Carl | Uppsala University |
Mattsson, Per | Uppsala University |
Schön, Thomas Bo | Uppsala University |
Keywords: Neural Networks, Nonlinear System Identification, Multivariable System Identification
Abstract: The goal of this paper is to provide a system identification-friendly introduction to the Structured State-space Models (SSMs). These models have become recently popular in the machine learning community since, owing to their parallelizability, they can be efficiently and scalably trained to tackle extremely long sequence classification and regression problems. Interestingly, SSMs appear as an effective way to learn deep Wiener models, which allows us to reframe SSMs as an extension of a model class commonly used in system identification. To stimulate a fruitful exchange of ideas between the machine learning and system identification communities, we deem it useful to summarize the recent contributions on the topic in a structured and accessible form. At last, we highlight future research directions for which this community could provide impactful contributions.
|
|
15:00-15:20, Paper WeA104.4 | Add to My Program |
State Derivative Normalization for Continuous-Time Deep Neural Networks |
|
Weigand, Jonas | University of Kaiserslautern |
Beintema, Gerben Izaak | Eindhoven University of Technology |
Ulmen, Jonas | RPTU Kaiserslautern-Landau |
Görges, Daniel | University of Kaiserslautern |
Tóth, Roland | Eindhoven University of Technology |
Schoukens, Maarten | Eindhoven University of Technology |
Ruskowski, Martin | German Research Center for Artificial Intelligence |
Keywords: Nonlinear System Identification, Continuous Time System Estimation, Neural Networks
Abstract: The importance of proper data normalization for deep neural networks is well known. However, in continuous-time state-space model estimation, it has been observed that improper normalization of either the hidden state or hidden state derivative of the model estimate, or even of the time interval can lead to numerical and optimization challenges with deep learning based methods. This results in a reduced model quality. In this contribution, we show that these three normalization tasks are inherently coupled. Due to the existence of this coupling, we propose a solution to all three normalization challenges by introducing a normalization constant at the state derivative level. We show that the appropriate choice of the normalization constant is related to the dynamics of the to-be-identified system and we derive multiple methods of obtaining an effective normalization constant. We compare and discuss all the normalization strategies on a benchmark problem based on experimental data from a cascaded tanks system and compare our results with other methods of the identification literature.
|
|
15:20-15:40, Paper WeA104.5 | Add to My Program |
Physics-Informed and Black-Box Identification of Robotic Actuator with a Flexible Joint |
|
Weiller Corrêa do Lago, Antonio | Pontifical Catholic University of Rio De Janeiro |
Braz de Sousa, Daniel Henrique | Instituto Militar De Engenharia |
Domingues, Pedro Henrique | PUC-Rio |
Daneker, Mitchell | University of Pennsylvania |
Lu, Lu | Yale University |
Hultmann Ayala, Helon Vicente | Pontifical Catholic University of Rio De Janeiro |
Keywords: Parameter Estimation, Neural Networks, Nonlinear System Identification
Abstract: In robotics, precise models are critical for ensuring safety and functionality. However, acquiring a precise model characterizing a system’s dynamics can be challenging. One of the alternatives to address this issue is system identification, which aims to obtain models through physical and experimental observations. In this manner, the developments in machine learning algorithms, such as neural networks, have significantly improved the modeling of complex and nonlinear phenomena. In this work, a mass-spring-damper (MSD) system and a low-cost original elastomer-based Series Elastic Actuators (eSEA) assembly are used to evaluate the performance of system identification models. The black-box models selected are variations of the AutoRegressive Moving Average with eXogenous input (ARMAX) algorithm. The gray-box model aims to estimate the parameters of 4 friction models; the optimization is done utilizing physics-informed neural networks (PINNs). For both case studies, the PINNs outperformed the black-box models. In the didactic example, the parameters obtained are close to the ground truth, and the highest determinant coefficient obtained is 0.99. The friction model that best represents the robotic actuator is the LuGre model, with the parameters obtained using the PINNs, outperforming the best black-box model by lowering the mean absolute error (MAE) by 30.83%. With a determinant coefficient of 0.94, the model shows a high capacity for describing the multiple nonlinearities present in the system.
|
|
15:40-16:00, Paper WeA104.6 | Add to My Program |
Learning Reduced-Order Linear Parameter-Varying Models of Nonlinear Systems |
|
Koelewijn, Patrick | Sioux Technologies B.V |
Singh, Rajiv | Northeastern University |
Seiler, Peter | Univ. of Michigan |
Tóth, Roland | Eindhoven University of Technology |
Keywords: Reduced-order Modeling, Neural Networks
Abstract: In this paper, we consider the learning of a Reduced-Order Linear Parameter-Varying Model (ROLPVM) of a nonlinear dynamical system based on data. This is achieved by a two-step procedure. In the first step, we learn a projection to a lower dimensional state-space. In step two, an LPV model is learned on the reduced-order state-space using a novel, efficient parameterization in terms of neural networks. The improved modeling accuracy of the method compared to an existing method is demonstrated by simulation examples.
|
|
WeA201 Invited Session, ISEC 136: Classroom |
Add to My Program |
Emerging Trends in Deep Learning for System Identification:
Physics-Informed and Beyond |
|
|
Chair: Forgione, Marco | SUPSI-USI |
Co-Chair: Piga, Dario | SUPSI-USI |
Organizer: Forgione, Marco | SUPSI-USI |
Organizer: Mammarella, Martina | CNR |
Organizer: Dabbene, Fabrizio | CNR |
Organizer: Piga, Dario | SUPSI-USI |
|
16:30-16:50, Paper WeA201.1 | Add to My Program |
One-Shot Backpropagation for Multi-Step Prediction in Physics-Based System Identification (I) |
|
Donati, Cesare | Politecnico Di Torino |
Mammarella, Martina | CNR |
Dabbene, Fabrizio | CNR |
Novara, Carlo | Politecnico Di Torino |
Lagoa, Constantino M. | Pennsylvania State Univ |
Keywords: Nonlinear System Identification, Grey Box Modelling, Parameter Estimation
Abstract: The aim of this paper is to present a novel physics-based framework for the identification of dynamical systems, in which the physical and structural insights are reflected directly into a backpropagation-like learning algorithm. The main result is a method to compute in closed form the gradient of a multi-step loss function, while enforcing physical properties and constraints. The derived algorithm has been exploited to identify the unknown inertia matrix of a space debris, and the results show the reliability of the method in capturing the physical adherence of the estimated parameters.
|
|
16:50-17:10, Paper WeA201.2 | Add to My Program |
On the Adaptation of In-Context Learners for System Identification (I) |
|
Piga, Dario | SUPSI-USI |
Pura, Filippo | SUPSI-IDSIA |
Forgione, Marco | SUPSI-USI |
Keywords: Neural Networks, Machine Learning and Data Mining
Abstract: In-context system identification aims at constructing meta-models to describe classes of systems, differently from traditional approaches that model single systems. This paradigm facilitates the leveraging of knowledge acquired from observing the behaviour of different, yet related dynamics. This paper discusses the role of meta-model adaptation. Through numerical examples, we demonstrate how meta-model adaptation can enhance predictive performance in three realistic scenarios: tailoring the meta-model to describe a specific system rather than a class; extending the meta-model to capture the behaviour of systems beyond the initial training class; and recalibrating the model for new prediction tasks. Results highlight the effectiveness of meta-model adaptation to achieve a more robust and versatile meta-learning framework for system identification.
|
|
17:10-17:30, Paper WeA201.3 | Add to My Program |
Deep Learning of Vehicle Dynamics (I) |
|
Szecsi, Mate | Institute for Computer Science and Control |
Gyorok, Bendeguz Mate | Institute for Computer Science and Control |
Weinhardt-Kovacs, Adam | Institute for Computer Science and Control |
Beintema, Gerben Izaak | Eindhoven University of Technology |
Schoukens, Maarten | Eindhoven University of Technology |
Peni, Tamas | Institute for Computer Science and Control (SZTAKI) |
Tóth, Roland | Eindhoven University of Technology |
Keywords: Nonlinear System Identification, Neural Networks, Continuous Time System Estimation
Abstract: Recent advances in deep-learning-based identification of dynamic systems have resulted in a new generation of approaches utilizing state-space neural models with innovation noise structure, improved reformulation of multiple shooting, batch optimization, and a subspace identification-inspired form of encoders. The latter is used to learn a reconstructability map to estimate model states from past inputs and outputs. By using the SUBNET approach, which belongs to the state-of-the-art of these methods, we show how to effectively use these approaches to identify reliable vehicle models from data both in continuous and discrete time, respectively. We showcase the approach on the identification of the dynamics of a Crazyflie 2.1 nano-quadcopter and a F1tenth electric car, both in a high-fidelity simulation environment, and in case of the electric car, on real measured data. The results indicate that new-generation of deep-learning methods offer efficient system identification of vehicle dynamics in practice.
|
|
17:30-17:50, Paper WeA201.4 | Add to My Program |
On Identifying the Non-Linear Dynamics of a Hovercraft Using an End-To-End Deep Learning Approach (I) |
|
Schwan, Roland | EPFL |
Schmid, Nicolaj | EPFL |
Chassaing, Etienne | EPFL |
Samaha, Karim | EPFL |
Jones, Colin, N | EPFL |
Keywords: Other Applications, Mechanical and Aerospace, Nonlinear System Identification
Abstract: We present the identification of the non-linear dynamics of a novel hovercraft design, employing end-to-end deep learning techniques. Our experimental setup consists of a hovercraft propelled by racing drone propellers mounted on a lightweight foam base, allowing it to float and be controlled freely on an air hockey table. We learn parametrized physics-inspired non-linear models directly from data trajectories, leveraging gradient-based optimization techniques prevalent in machine learning research. The chosen model structure allows us to control the position of the hovercraft precisely on the air hockey table. We then analyze the prediction performance and demonstrate the closed-loop control performance on the real system.
|
|
17:50-18:10, Paper WeA201.5 | Add to My Program |
Physics-Guided State-Space Model Augmentation Using Weighted Regularized Neural Networks (I) |
|
Liu, Yuhan | Eindhoven University of Technology |
Tóth, Roland | Eindhoven University of Technology |
Schoukens, Maarten | Eindhoven University of Technology |
Keywords: Nonlinear System Identification, Neural Networks, Regularization and Kernel Methods
Abstract: Physics-guided neural networks (PGNN) is an effective tool that combines the benefits of data-driven modeling with the interpretability and generalization of underlying physical information. However, for a classical PGNN, the penalization of the physics-guided part is at the output level, which leads to a conservative result as systems with highly similar state-transition functions, i.e. only slight differences in parameters, can have significantly different time-series outputs. Furthermore, the classical PGNN cost function regularizes the model estimate over the entire state space with a constant trade-off hyperparameter. In this paper, we introduce a novel model augmentation strategy for nonlinear state-space model identification based on PGNN, using a weighted function regularization (W-PGNN). The proposed approach can efficiently augment the prior physics-based state-space models based on measurement data. A new weighted regularization term is added to the cost function to penalize the difference between the state and output function of the baseline physics-based and final identified model. This ensures the estimated model follows the baseline physics model functions in regions where the data has low information content, while placing greater trust in the data when a high informativity is present. The effectiveness of the proposed strategy over the current PGNN method is demonstrated on a benchmark example.
|
|
18:10-18:30, Paper WeA201.6 | Add to My Program |
Identifying the Dynamics of Interacting Objects with Applications to Scene Understanding and Video Temporal Manipulation (I) |
|
Comas Massague, Armand | Northeastern University |
Fernandez, Christian | Northeastern University |
Ghimire, Sandesh | Qualcomm |
Li, Haolin | Northeastern University |
Camps, Octavia I. | Northeastern University |
Sznaier, Mario | Northeastern University |
Keywords: Neural Networks, Other Applications, Dynamic Network Identification
Abstract: There is an ongoing effort in the machine learning community to enable machines to understand the world symbolically, facilitating human interaction with learned representations of complex scenes. A pre-requisite to achieving this is the ability to identify the dynamics of interacting objects from time traces of relevant features. In this paper, we introduce GrODID (GRaph-based Object-Centric Dynamic Mode Decomposition), a framework based on graph neural networks that enables Dynamic Mode Decomposition for systems involving interacting objects. The main idea is to model individual, potentially non-linear dynamics using a Koopman operator and identify its corresponding Dynamic Mode Decomposition using deep AutoEncoders, while the interactions amongst systems are captured by a graph, modeled by a Graph Neural Net (GNN). The potential of this approach is illustrated with several applications arising in the context of video analytics: video forward and backwards prediction, video manipulation and achieving temporal super-resolution
|
|
18:30-18:50, Paper WeA201.7 | Add to My Program |
Physics-Informed Neural Network for System Identification of Rotors (I) |
|
Liu, Xue | Xi'an Jiaotong University |
Cheng, Wei | Xi'an Jiaotong University |
Xing, Ji | China Nuclear Power Engineering Co. Ltd |
Chen, Xuefeng | Xi'an Jiaotong University |
Zhao, Zhibin | Xi'an Jiaotong University |
Zhang, Rongyong | China Nuclear Power Engineering Co. Ltd |
Huang, Qian | China Nuclear Power Engineering Co. Ltd |
Lu, Jinqi | Shanghai Apollo Machinery Co., Ltd |
Zhou, Hongpeng | The University of Manchester |
Zheng, Wei Xing | Western Sydney University |
Pan, Wei | University of Manchester |
Keywords: Mechanical and Aerospace, Fault Detection and Diagnosis, Monitoring
Abstract: The condition of the rotor system remains difficult to assess due to system nonlinearity and nosiy measurements. To deal with the problem, we proposed a hierarchical physics-informed neural network (HPINN) to discover the ordinary differential equation (ODE) of a healthy/faulty rotor system from noise measurements and then assess the machine condition based on the discovered ODE. Specifically, the ODE of a healthy rotor system is first stably identified from noise measurement through HPINN guided by rotor dynamics. Based on the identified healthy ODE, the extra fault terms in the ODE of the faulty rotor system are then sparsely regressed from the predefined library embedded in HPINN. Moreover, with the mathematical terms of discovered fault, the potential fault and the health indicator (HI) are diagnosed and constructed to assess the condition of the rotor system, respectively. Finally, the effectiveness of the proposed method is verified by the data set collected on the circulating water test bench, showing the potential for practical applications.
|
|
WeA202 Regular Session, ISEC 138: Classroom |
Add to My Program |
Recent Advances in System Identification Theory and Implementation - Part
II |
|
|
Co-Chair: Singh, Rajiv | Northeastern University |
|
16:30-16:50, Paper WeA202.1 | Add to My Program |
Subglottal Impedance-Based Model Parameter Estimation Via System Identification |
|
G. Fontanet, Javier | Universidad Técnica Federico Santa María |
Yuz, Juan I. | Universidad Técnica Federico Santa María |
Garnier, Hugues | University of Lorraine |
Espinoza, Víctor | Universidad De Chile |
Zañartu, Matias | Universidad Técnica Federico Santa María |
Keywords: Biomedical Systems, Continuous Time System Estimation, Frequency Domain Identification
Abstract: Accurate mathematical modeling of different systems of the human body stands as a key issue in medical and bioengineering applications. This paper specifically considers the continuous-time parameter estimation of the impedance-based mathematical model — a mechano-acoustic representation of the subglottal system. This approach allows the customization of the model for each patient. A key advantage of having a patient-dependent model of the subglottal system is to facilitate the ambulatory non-invasive monitoring of the glottal airflow and the assessment of vocal functions, using an accelerometer on the neck skin surface. For this model of the subglottal system, the glottal airflow is the excitation signal, while the acceleration on the neck skin is the system response. In this study, continuous-time parameter estimation of the impedance-based model for the subglottal system is applied, using both frequency response and sampled data through system identification techniques.
|
|
16:50-17:10, Paper WeA202.2 | Add to My Program |
Online System Identification of Global Lung Heat Transfers |
|
Victor, Stephane | Univ. Bordeaux |
Ndreko, Enso | Univ. Bordeaux |
Melchior, Pierre | Université De Bordeaux - Bordeaux INP/ENSEIRB-MATMECA |
Keywords: Biomedical Systems, Recursive Identification, Model Validation
Abstract: In cardiac surgery where extracorporeal circulation is used, the lungs are temporarily disconnected from the body and are connected to a device that provides air and blood. To minimize the risk of tissue damage, the lungs are subjected to mild hypothermia. Heat transfer modeling offers the potential to enhance temperature regulation through a more advanced approach. From a thermal quadrupole formalism combined with the heat equation or the bio-heat equation when considering blood perfusion according to the branch level, a more compact model is used for system identification. The feasibility of online system identification is proposed for coefficient estimation by using an extended Prediction Error Method algorithm more suited for estimating fractional order systems: the Long Memory Prediction Error Method (LMRPEM).
|
|
17:10-17:30, Paper WeA202.3 | Add to My Program |
Fast Dynamic Analysis of Damaged 1D Periodic Waveguides |
|
Gavilan Rojas, Alvaro | Inria |
Zhang, Qinghua | INRIA |
Robin, Olivier | Université De Sherbrooke |
Droz, Christophe | Inria |
Keywords: Reduced-order Modeling, Vibration and Modal Analysis, Fault Detection and Diagnosis
Abstract: We extend the Bloch wave-based reduced order models in the Wave Finite Element Method (WFEM) framework for fast wave-damage interaction analysis. It aims at fault detection and diagnosis of periodic structures. A finned tube heat exchanger, which can be seen as a 1D system, is used as a numerical application. Reduced model results and performance are compared to a standard WFEM model. Diffusion curves are obtained more than a hundred times faster with the proposed scheme, moving toward massive generation of wave-damage scenarios and indicators to perform damage detection.
|
|
17:30-17:50, Paper WeA202.4 | Add to My Program |
Weighted Least-Squares PARSIM |
|
He, Jiabao | KTH Royal Institute of Technology |
Rojas, Cristian R. | KTH Royal Institute of Technology |
Hjalmarsson, Håkan | KTH |
Keywords: Subspace Methods
Abstract: Subspace identification methods (SIMs) have proven very powerful for estimating linear state-space models. To overcome the deficiencies of classical SIMs, a significant number of algorithms has appeared over the last two decades, where most of them involve a common intermediate step, that is to estimate the range space of the extended observability matrix. In this contribution, an optimized version of the parallel and parsimonious SIM (PARSIM), PARSIMtextsubscript{opt}, is proposed by using weighted least-squares. It not only inherits all the benefits of PARSIM but also attains the best linear unbiased estimator for the above intermediate step. Furthermore, inspired by SIMs based on the predictor form, consistent estimates of the optimal weighting matrix for weighted least-squares are derived. Essential similarities, differences and simulated comparisons of some key SIMs related to our method are also presented.
|
|
17:50-18:10, Paper WeA202.5 | Add to My Program |
Least Squares Projection Onto the Behavior for SISO LTI Models |
|
Lagauw, Sibren | KU Leuven |
De Moor, Bart L.R. | Katholieke Universiteit Leuven |
Keywords: Time Series, Errors in Variables Identification, Parameter Estimation
Abstract: We consider the least squares projection onto the behavior for discrete-time linear time-invariant (LTI) single-input single-output (SISO) models, in which the observed input-output data are modified in a least squares (LS) sense by subtracting so-called misfits, so that the modified data satisfy a given linear dynamic relation. We show that the LS-criterion of the projection problem induces an orthogonal decomposition of the ambient data space and we characterize this decomposition by means of banded block-Toeplitz matrices, the elements of which are the coefficients of the difference equation describing the SISO LTI dynamics. We thereby generalize earlier results in the literature on autonomous LTI models to the more complicated SISO case. Additionally, we illustrate that the novel characterization is equivalent (up to a change of model representation) to results derived using (isometric) state space representations in the literature on behavioral systems theory.
|
|
18:10-18:30, Paper WeA202.6 | Add to My Program |
Noise Covariances Identification by MDM: Weighting, Recursion, and Implementation |
|
Kost, Oliver | University of West Bohemia |
Dunik, Jindrich | University of West Bohemia |
Straka, Ondrej | University of West Bohemia |
Keywords: Recursive Identification, Multivariable System Identification, Toolboxes
Abstract: The problem of noise covariance matrix identification of stochastic linear time-varying state-space models is addressed. The measurement difference method (MDM) is generalized to time-varying dimensions of the measurement and control. Three MDM identification techniques that differ in weighting used in the underlying least squares method are proposed. The techniques differ in estimate quality and computational complexity. In addition, recursive forms are designed for two techniques. The performance of the proposed techniques is analyzed using two numerical examples. The implementation of techniques is enclosed with the paper.
|
|
WeA203 Regular Session, ISEC 140: Classroom |
Add to My Program |
Identification for Control |
|
|
Chair: Breschi, Valentina | Eindhoven University of Technology |
Co-Chair: Hjalmarsson, Håkan | KTH |
|
16:30-16:50, Paper WeA203.1 | Add to My Program |
Data-Driven Explicit Predictive Control with Limited Resources: An Exploration-Based Strategy |
|
Sassella, Andrea | Politecnico Di MIlano |
Breschi, Valentina | Eindhoven University of Technology |
Formentin, Simone | Politecnico Di Milano |
Keywords: Data-driven Control
Abstract: Retrieving optimal control actions in a receding horizon fashion at run time might be a challenging task, especially when the sampling time of the system to be controlled is small and the optimization problem is large. Although explicit solutions have been proposed to tackle this challenge, the complexity of the explicit control law scales poorly with the dimension of the problem. In the attempt to cope with these limitations within the challenging data-driven setup, we propose to construct a limited-complexity approximation of the explicit predictive law by iteratively exploring the state/reference space while leveraging structural priors on the input parameterization. The same approximation can be exploited to compute the optimal action also when the closed loop system visits unexplored regions. The performance of the proposed strategy is assessed on a simple numerical example, highlighting its effectiveness for large problem sizes.
|
|
16:50-17:10, Paper WeA203.2 | Add to My Program |
Efficient Tuning for Motion Control in Diverse Systems: A Bayesian Framework |
|
Catenaro, Edoardo | Politecnico Di Milano |
Aarnoudse, Leontine | Eindhoven University of Technology |
Formentin, Simone | Politecnico Di Milano |
Oomen, Tom | Eindhoven University of Technology |
Keywords: Data-driven Control, Bayesian Methods, Mechanical and Aerospace
Abstract: Feed-forward control is widely used in motion control systems that involve repetitive tasks, leading to substantial performance improvements. This paper presents a model-free feedforward optimization framework centred around Bayesian Optimization (BO). Bypassing the need for exhaustive system modelling, the method directly optimizes the Iterative Learning Control (ILC) degrees of freedom based on a user-defined parametrization of the feed-forward controller. Experimental results on a motion control application show significant improvements with respect to more classical ILC. A notable advantage emerges when dealing with an industrially relevant case with multiple similar plants; the optimizer is shown to adeptly adjust the feed-forward control to be compliant with the response of the measured system.
|
|
17:10-17:30, Paper WeA203.3 | Add to My Program |
Control-Relevant Input Signal Design for Integrating Processes: Application to a Microalgae Raceway Reactor |
|
Banerjee, Sarasij | Arizona State University |
Otálora, Pablo | University of Almería |
El Mistiri, Mohamed | Arizona State University |
Khan, Owais | Arizona State University |
Guzman, Jose Luis | University of Almeria (Q-5450008-G) |
Rivera, Daniel E. | Arizona State University |
Keywords: Identification for Control, Experiment Design, Process Control
Abstract: Plants with integrators possess control-relevant modeling requirements that are typically ignored in the literature. Desired closed-loop speeds of response for such systems can differ significantly from their open-loop dynamics, causing conventional system identification guidelines to fail or be inaccurate. One such issue relates to experimental design. This paper presents guidelines for the design of excitation signals for system identification of plants with integrators, with application to the modeling and control of a microalgae raceway reactor. The concept is to excite the system through optimized test signals with control-relevant shaping of their power spectra. This facilitates a shift in emphasis of the identification objective from estimating a model with good open-loop performance to having a model possessing desired closed-loop characteristics. Such a consideration is particularly important when generating informative databases for estimating predictive models for closed-loop control. An illustration for this experimental design procedure is accomplished in this paper through the estimation of ARX-based models and, subsequently, model predictive control of the pH dynamics of an experimental raceway photobioreactor facility hosting sustainable microalgae production.
|
|
17:30-17:50, Paper WeA203.4 | Add to My Program |
Output-Only Identification of Lur'e Systems with Prandtl-Ishlinskii Hysteresis Nonlinearities |
|
Aljanaideh, Khaled | Jordan University of Science and Technology |
Al Janaideh, Mohammad | Memorial University |
Richards, Riley | University of Michigan |
Paredes Salazar, Juan Augusto | University of Michigan |
Bernstein, Dennis S. | University of Michigan |
Keywords: Nonlinear System Identification, Identification for Control, Nonparametric Methods
Abstract: Lur’e systems are dynamical systems that are characterized by the feedback interconnection between a linear, time invariant system and a feedback nonlinearity. Lur’e systems have been used to characterize the dynamics of several systems including gas turbine combustors and self-oscillatory systems. In this paper, we introduce an identification algorithm for Lur’e systems with hysteretic feedback nonlinearities. We assume that the input to the Lur'e system is an unknown constant signal, and the linear dynamics have unknown nonzero initial conditions. First, we use least squares with a transfer function model to identify the linear dynamics of the Lur’e system. Then, we use the identified linear dynamics along with the measured output to construct an estimate of the hysteretic nonlinearity. We show numerical examples to illustrate the proposed approach.
|
|
17:50-18:10, Paper WeA203.5 | Add to My Program |
Data Driven Positive Subspace System Identification |
|
Wang, Yueyang | Northeastern University |
Shafai, Bahram | Northeastern Univ |
Keywords: Subspace Methods, Identification for Control
Abstract: This paper considers the data-driven subspace system identification for positive systems. The application of subspace system identification (SSID) method to collected data of positive systems does not guarantee the positivity of the identified system. Thus, two approaches are proposed to obtain the state space parameters of the positive system using the data collected from its input and output. The first approach offers a procedure based on nonnegative least square algorithm applied to data-driven state space representation. The second approach applies a nonnegative matrix factorization to the data collected Hankel matrix followed by generalized realization algorithm. The effectiveness of two approaches in positive SSID are demonstrated by numerical examples.
|
|
WeA204 Regular Session, ISEC 142: Classroom |
Add to My Program |
Multivariable and Networked Systems |
|
|
Chair: Bitmead, Robert | University of California San Diego |
Co-Chair: Materassi, Donatello | University of Minnesota |
|
16:30-16:50, Paper WeA204.1 | Add to My Program |
Unveiling Fringe Social Network Dynamics Via Parameter Estimation with Hawkes Processes |
|
Grande, Davide | Politecnico Di Torino |
Ravazzi, Chiara | National Research Council of Italy (CNR) |
Malandrino, Francesco | National Research Council of Italy (CNR) |
Dabbene, Fabrizio | CNR |
Keywords: Dynamic Network Identification, Maximum Likelihood Methods, Parameter Estimation
Abstract: Fringe social networks, e.g., 4chan or Truth, position themselves as “free speech” alternatives to their mainstream counterparts like Facebook or X (formerly Twitter). Due to their very lax moderation policies, they however tend to become a hotbed for misinformation or otherwise malicious content, which then tends to spread towards the general public. In order to effectively counter such a process, it is important to properly understand and model how content appears and spreads over fringe social networks. Accordingly, in this study we focus on the now-defunct Parler social network, and conduct a statistical analsysis over 183 million posts dating from August 2018 to January 2021. The primary objective is to comprehensively analyze hashtag cascades related to the first impeachment of U.S. President Donald Trump. Our aim is to (i) uncover how external actors inject malicious and hateful tendencies into the network and (ii) quantify the levels of attention within these communities. We find that the hashtag cascade can be effectively modeled using the Hawkes process framework, specifically, employing an exponential decay kernel. Rigorous parameter estimation and statistical tools are employed to substantiate this assertion and evaluate the model’s goodness of fit. Importantly, the analysis reveals correlations between levels of hate, the dissemination of misleading information, and the attention garnered within these fringe social communities.
|
|
16:50-17:10, Paper WeA204.2 | Add to My Program |
Fault Detection and Diagnosis Using the Dynamic Network Framework |
|
Shi, Yibo | Eindhoven University of Technology |
Fonken, Stefanie | Eindhoven University of Technology |
Van den Hof, Paul M.J. | Eindhoven University of Technology |
Keywords: Dynamic Network Identification, Model Validation, Fault Detection and Diagnosis
Abstract: A local model-based method for fault detection and diagnosis (FDD) in large-scale interconnected network systems is introduced, using models in a dynamic network framework. To this end, model validation methods are developed for validating single modules in a dynamic network, which are generalized from the classical auto- and cross-correlation tests for open- and closed-loop systems. Invalidation of the model can indicate the detection of a fault in the system. A fault diagnosis algorithm is developed that includes fault isolation and optimal placement of external excitation signals. Numerical illustrations demonstrate the method's capability to detect a fault in a local module and isolate it within the entire network system.
|
|
17:10-17:30, Paper WeA204.3 | Add to My Program |
Characterization of Minimal Network Structures Modeling Stochastic Processes |
|
Biparva, Darya | University of Minnesota |
Materassi, Donatello | University of Minnesota |
Keywords: Dynamic Network Identification, Multivariable System Identification, Time Series
Abstract: Identifying the underlying structure of a network from observed data is an important problem across various disciplines. Given the general ill-posed nature of the problem, since in many cases, multiple plausible network models can explain the data, this article concentrates on characterizing classes of models providing possible explanations. Specifically, we explore linear models that can account for observed data in the form of wide-sense stationary processes accommodating the potential presence of feedback loops and direct feedthroughs. To achieve this, we leverage key insights from the theory of graphical models. In particular, we extensively employ Pearl-Verma Theorem in causal discovery which allows one to recover all minimal network structures compatible with the observed data. We adapt such a result to deal with stochastic processes and reinterpret it as a Gram-Schmidt orthogonalization procedure in a suitable Hilbert space. This reinterpretation allows us to characterize all minimal networks explaining a set of data, which have the property of not having any algebraic loops.
|
|
17:30-17:50, Paper WeA204.4 | Add to My Program |
ARMA Identification of Kronecker Graphical Models |
|
Zorzi, Mattia | Università Degli Studi Di Padova |
Keywords: Dynamic Network Identification, Regularization and Kernel Methods, Frequency Domain Identification
Abstract: We address the problem to estimate a Kronecker graphical model corresponding to an autoregressive and moving average (ARMA) stochastic process. The graph describes conditional dependence relations among the components of the process. We propose a Bayesian approach to estimate the model and the topology. We test the effectiveness of the proposed method by some numerical experiments.
|
|
17:50-18:10, Paper WeA204.5 | Add to My Program |
Stable State Estimation in a Network Context |
|
Bitmead, Robert | University of California San Diego |
Keywords: Filtering and Smoothing, Time Series, Process Control
Abstract: Power grid, communications, computer and product reticulation networks are frequently layered or subdivided by design. The OSI seven-layer computer network model and the electrical grid division into generation, transmission, distribution and associated markets are cases in point. The layering divides responsibilities and can be driven by operational, commercial, regulatory and privacy concerns. From a control context, a layer, or part of a layer, in a network isolates the authority to manage, i.e. control, a dynamic system with connections into unknown parts of the network. The topology of these connections is fully prescribed but the interconnecting signals are largely unavailable, through lack of sensing and even prohibition. Accordingly, one is driven to simultaneous input and state estimation methods. This is the province of this paper, guided by the structure of these network problems. We study a class of algorithms for this joint task, which if the system has transmission zeros outside the unit circle leads to an unstable and unimplementable estimator. Two modifications to the algorithm to ameliorate this problem were recently proposed involving (a) replacing the troublesome subsystem with its outer factor from its inner-outer factorization or (b) using the Kalman filter for a high-variance white signal model for the unknown inputs. The outer factor has only stable transmission zeros and so is stably invertible. The Kalman filter is stable by design. Here, we establish the connections between the original estimation problem for state and input signal and the outputs/estimates from the algorithm applied solely to the outer factor. We further show that both fixes coincide.
|
|
18:10-18:30, Paper WeA204.6 | Add to My Program |
Covariance Analysis of the Estimated Markov Parameters in a Subspace Identification Method |
|
Ikeda, Kenji | Tokushima University |
Tanaka, Hideyuki | Hiroshima University |
Keywords: Multivariable System Identification, Subspace Methods, Identification for Control
Abstract: It is important to provide a covariance of the estimates to ensure the quality of the identification results. This paper proposes a covariance of the estimated Markov parameters in a method previously proposed by the authors. The proposed covariance uses the gap between singular subspaces to estimate the perturbation of the extended observability matrix. The gap based analysis gives a simple expression in the sense that the estimated error in the singular subspace is strictly linear with respect to the perturbation of the original matrix and the left and right singular vectors. Numerical simulation shows the validity of the proposed covariance of the estimated Markov parameters.
|
|
18:30-18:50, Paper WeA204.7 | Add to My Program |
A Latent Representation of Brain Networks Based on EEG |
|
Falconi, Lucia | Università Degli Studi Di Padova |
Cisotto, Giulia | University of Milan-Bicocca |
Zorzi, Mattia | Università Degli Studi Di Padova |
Keywords: Biological Systems, Dynamic Network Identification, Regularization and Kernel Methods
Abstract: Electroencephalography (EEG) is one of the most popular techniques to investigate normal as well as pathological cerebral mechanisms, as it allows to measure, non-invasively and in real-time, the brain activity. However, modeling EEG is still extremely challenging, because of its high-dimensionality, low signal-to-noise ratio, and high individual variability. This paper proposes a novel latent representation to study brain networks using EEG by means of a robust dynamic factor analysis (RDFA) approach. We investigate the ability of this latent representation to discriminate between two groups of subjects, i.e. alcoholic and healthy. By RDFA, we can extract a limited number of highly explanatory factors, as low as 8, significantly discriminating between the two groups. Also, we show that different brain patterns can be identified across different stimulation scenarios and EEG locations. Although preliminary, this work could give support to domain experts while providing some clinically-meaningful insights to identify common patterns as well as individual characteristics in different groups of healthy and pathological subjects.
|
| |