基于PSO-SVR动态模型的车辆排队长度预测

Vehicle queue length prediction based on PSO-SVR dynamic model

针对突发事件下城市道路车辆排队系统的特点,从时空角度综合考虑车辆排队系统的影响因素,建立支持向量回归(SVR)动态模型对车辆排队长度进行预测。考虑到参数选择对模型性能影响的敏感性,提出了以k折交叉验证(k-CV)均方误差平均值为适应度的粒子群优化(PSO)方法并对SVR模型参数进行寻优。用提出的PSO-SVR模型与K-CV和遗传算法(GA)优化的SVR模型以及BP网络预测模型对比,实验结果表明,该模型具有较高的预测精度和泛化能力,适用于车辆排队长度的预测。

According to the characteristics of city road vehicle queuing system under emergency, considering the affect factors from the view of space-time, it establishes Support Vector Regression（SVR）model to predict vehicle queue length.Considering the sensitivity of parameters effecting on model performance, Particle Swarm Optimization（PSO）algorithm is proposed to select SVR parameters. Furthermore, the k-fold cross validation（k-CV） mean square error averaged is used as the fitness of PSO. The proposed PSO-SVR model compares with k-CV SVR model, GA-SVR model and BP network. The test results show that PSO-SVR model has higher prediction accuracy and generalization ability, and the model is effective to predict vehicle queue length.