改进PSO优化神经网络的短时交通流预测

在短时交通流预测中，传统PSO优化神经网络预测模型对逃逸粒子直接取边界值且自身无相应的变异机制，这对于维持粒子群多样性、寻找最优解是不利的。为更进一步提高短时交通流预测精度，将在传统PSO优化BP神经网络的基础上，引入边界变异算子、自变异算子对粒子进行双重变异以优化网络配置参数。用实测的北京二环交通流数据对改进的预测模型进行验证，结果表明该模型更有利于搜寻全局最优解，且寻优时间更短，能有效改善短时交通流预测性能。

Short-term prediction of traffic flow based on neural network optimized improved particle swarm optimization

In short-term traffic flow prediction, the traditional PSO optimizes the neural network model for prediction sets escape particle on the boundary directly and has no corresponding variation mechanism by itself, which is bad for maintaining the diversity of particle swarm and finding the optimal solution. To further improve the accuracy of short-term traffic flow prediction, boundary mutation operator and self-adaptive mutation operator called double mutation are proposed in PSO to optimize the network configuration parameters based on the traditional PSO to optimize the BP neural network. The proposed prediction model is tested by measured Beijing 2nd ring road’s traffic flow data and the computational results show that this modified prediction method is more beneficial to search for the global optimal solution and save optimization time, and can improve the performance of short-term traffic flow prediction effectively.