A Novel Stochastic Model Predictive Control Considering Predictable Disturbance With Application to Personalized Adaptive Cruise Control

A novel stochastic model predictive control (SMPC) scheme is proposed for automotive scenes based on high-performance and practical motion state prediction method. The significant properties of the proposed scheme are that: 1) it can accurately predict disturbances within the prediction horizon, and 2) the prediction results can be considered into the optimizing process to obtain a more efficient and accurate controller. As a result, the proposed adaptive cruise control (ACC) system can ensure driving safety and improve tracking accuracy and comfort performance while satisfying different driving styles. In detail, a large amount of naturalistic driving data is collected based on a real vehicle test platform at first. Then an adaptive optimization Gaussian process regression (AOGPR) is developed and trained with real measurements to predict the motion states of the preceding vehicle. The prediction module is embedded in SMPC to bind the collision conditions, tighten the states and finally construct a novel controller, i.e., AOGPR-SMPC controller. A bidirectional LSTM (BiLSTM) network is trained and tested for online recognizing driving styles to satisfy personalized car-following needs. The simulation and field tests verify and evaluate the proposed controller. The results demonstrate that the ACC system could realize personalized car-following according to the driver’s driving style, and the proposed controller can obtain better tracking accuracy and comfort performance compared with the GPR-SMPC controller and MPC controller.