Neural Network Optimized Model Predictive Multi-Objective Adaptive Cruise Control

Abstract

A model predictive multi-objective adaptive cruise control (MPC MO-ACC) system, designed to consider both the tracking performance and the fuel consumption, is optimized by a neural network in this paper, reducing the computational complexity without sacrificing the control performance. The optimized MO-ACC control system is built by training a neural network with the control results of the MPC MO-ACC system. Simulation tests are conducted in Matlab/Simulink in conjunction with the high-fidelity CarMaker software. Influences of four driving conditions (the learning track, NEDC, JP05, FTP75) and two kinds of sensor models (ideal radar sensor and 77GHz physical radar sensor) are analysed. Simulation results have shown that the neural network optimized model predictive MO-ACC has the same control capability and strong robustness as the original MPC MO-ACC. Meanwhile, the optimized control system has much lower computational complexity, which shows potentials for the application in real-time vehicle control and industry.