Deceleration Planning Algorithm Based on Classified Multi-Layer Perceptron Models for Smart Regenerative Braking of...

Area of Science:

• Automotive Engineering
• Artificial Intelligence
• Control Systems

Background:

• Smart regenerative braking systems (SRS) offer autonomous one-pedal driving in electric vehicles.
• Effective SRS implementation requires a deceleration planning algorithm for automatic regenerative control.
• Reducing driver discomfort necessitates aligning automatic regeneration with human driving patterns.

Purpose of the Study:

• To propose a novel deceleration planning algorithm for SRS based on multi-layer perceptron (MLP) models.
• To enhance SRS performance by mimicking human driving behavior and adapting to different deceleration scenarios.
• To validate the proposed algorithm's effectiveness through driving simulations.

Main Methods:

• Developed a deceleration planning algorithm utilizing multi-layer perceptron (MLP) models trained on driving data.
• Implemented a classified structure for the algorithm, with individual MLP models for car-following, speed bump, and intersection conditions.
• Validated the algorithm through driving simulations, analyzing time to collision and velocity root-mean-square error (RMSE).

Main Results:

• The algorithm successfully mimicked human driving behavior, achieving a velocity RMSE of 0.302 m/s.
• Ensured safe operation with a minimum time to collision of 1.443 seconds.
• Demonstrated superior planning performance compared to an integrated structure.

Conclusions:

• The proposed classified deceleration planning algorithm enables safe and comfortable SRS operation in electric vehicles.
• MLP models effectively learn and replicate human driving patterns for regenerative braking control.
• The classified structure significantly improves SRS planning performance across diverse driving conditions.