Noda, Yuto (The University of Tokyo), Yamasaki, Yudai (The University of Tokyo)

Characteristics of Accelerator Pedal Operation Prediction Model by Comparing to Driving Data Clustering

Scheduled for presentation during the Invited session "Modeling and Control of Advanced Powertrain System" (ThA4), Thursday, October 31, 2024, 10:30−10:50, Room T4

7th IFAC Conference on Engine and Powertrain Control, Simulation and Modeling, Oct 30 - Nov 1, 2024, Dalian, China

This information is tentative and subject to change. Compiled on May 16, 2025

Abstract

To reduce the fuel consumption of vehicles in real world, a powertrain control based on a prediction of a driver’s behavior is useful. In the final purpose of predicting accelerator pedal position(Accel) considering each driver’s characteristics in real world, we are building the Long Short-Term Memory(LSTM)-based accel operation predictive model. In this paper, to understand the response of a driver’s characteristics according to operation situations, we analyzed memory cells of the LSTM models and applied Gaussian Mixture Model(GMM) clustering to divide situations(“pre-clustering”) of learning data. By comparing the groups represented by LSTM's memory cells with ones by pre-clustering, it shows that the memory cells can identify the accel position, the preceding vehicle’s distance, and the vehicle speed. On the other hand, it suggests that the memory cells don’t identify the brake, the gradient, and the relative speed. These results are useful for understanding the predictive characteristics of LSTM and the way to modify a model with LSTM and the learning data treatment.