Abstract: Based on parameter prediction， a RBF neural network adaptive control scheme was proposed for the motion control problems of autonomous electric vehicles with uncertainties. Firstly, the influences of system parameter uncertainties and external interferences were considered, and a dynamic model which might reflect the tracking and following behaviors of vehicles was established by the preview method. Secondly, RBF neural network compensator was adopted to compensate system uncertainties adaptively, and a generalized coordinated control law was designed for the lateral and longitudinal motions of vehicles. Thirdly, the impacts from the front vehicle speeds and road curvatures were taken into account, and the minimization of the energy consumption and the average jerks in the tracking and following control processes were regarded as the optimization objects. Afterwards, PSO algorithm was utilized to rolling optimize the gain parameter K in the coordinated control law, and then a series of optimized sample data were obtained. Then, to ensure the economy and ride comfort of vehicles, a BP neural network was designed and trained to realize the real-time prediction of gain parameter K in the generalized coordinated control law. Simulation results validate the effectiveness of the proposed control scheme.

Key words: autonomous electric vehicle, uncertainty, radial basis function(RBF) neural network, particle swarm optimization(PSO) algorithm, parameter prediction

摘要： 针对具有不确定性的自动驾驶电动车辆的运动控制问题，提出了一种基于参数预测的径向基函数（RBF）神经网络自适应协调控制方案。首先，考虑系统参数的不确定性及外部干扰的影响，利用预瞄方法建立可表征车辆循迹跟车行为的动力学模型；其次，采用RBF神经网络补偿器对系统不确定性进行自适应补偿，设计车辆横纵向运动的广义协调控制律；之后，考虑前车车速及道路曲率影响，以车辆在循迹跟车控制过程中的能耗及平均冲击度最小为优化目标，利用粒子群优化（PSO）算法对协调控制律中的增益参数K进行滚动优化，并最终得到一系列优化后的样本数据；在此基础上，设计、训练一个反向传播（BP）神经网络，实现对广义协调控制律中增益参数K的实时预测，以保证车辆的经济性及乘坐舒适性。仿真结果证实了所提控制方案的有效性。

关键词: 自动驾驶电动车辆, 不确定性, 径向基函数神经网络, 粒子群优化算法, 参数预测