青岛理工大学学报

为提高车辆自主轨迹预测的准确度，运用机器学习方法，提出了一种基于粒子群算法(Particle Swarm Optimization，PSO)优化的径向基函数神经网络(Radial Basis Function，RBF)预测模型(PSO‐RBF)。借助HighD数据集构建实验数据；利用MATLAB编写程序训练并进行仿真分析；同时与传统的神经网络模型对比。结果表明，基于PSO算法优化的RBF神经网络预测模型具有更小的预测误差，PSO‐RBF模型的平均绝对误差相比原始的RBF模型下降了32.81%，相比BP模型下降了42.67%，相比SVM模型下降了12.25%，相比RF模型下降了23.21%；PSO‐RBF模型的均方根误差相比原始的RBF模型下降了35.85%，相比BP模型下降了52.78%，相比SVM模型下降了22.73%，相比RF模型下降了32%。该模型能够准确预测车辆的行驶轨迹，在预测的准确度上明显优于一般的机器学习模型。

In order to improve the accuracy of vehicle trajectory prediction， a Particle Swarm Optimization (PSO) based Radial Basis Function (RBF) neural network prediction model (PSO‐RBF) is proposed using machine learning methods. Construct experimental data with HighD data set， using MATLAB to write programs for training and simulation analysis， and compare PSO‐RBF with the traditional neural network model. The results show that the prediction model of RBF neural network optimized by PSO algorithm has smaller prediction error. Compared with original RBF model， the Mean Absolute Error value of PSO‐RBF model decreased by 32.81% and the Root Mean Square Error value decreased by 35.85%. Compared with BP model， the Mean Absolute Error value of PSO‐RBF model decreased by 42.67% and the Root Mean Square Error value decreased by 52.78%. Compared with SVM model， the Mean Absolute Error value of PSO‐RBF model decreased by 12.25% and the Root Mean Square Error value decreased by 22.73%. Compared with RF model， the Mean Absolute Error value of PSO‐RBF model decreased by 23.21% and the Root Mean Square value decreased by 32%. This model can accurately predict the driving trajectory of vehicles， and its prediction accuracy is significantly better than that of ordinary machine learning models.