改进麻雀搜索算法优化BP神经网络的短时交通流预测

The Improvement of sparrow search algorithm and the optimization of BP neural network for short term traffic flow prediction

针对BP神经网络预测短时交通流量过于依赖初始参数的问题,提出一种基于改进麻雀搜索算法(ISSA)来优化BP神经网络的短时交通流预测模型(ISSA-BP)。针对标准麻雀搜索算法(SSA)易收敛于原点,容易陷入局部最优等问题,对麻雀群体中的发现者和部分加入者的位置更新公式分别进行改进,改进后的发现者将基于搜索维度的大小和当前最优值的位置来进行全局搜索,部分加入者将根据其与最优位置之间的距离来进行全局搜索。通过实验对BP,PSO-BP,SSA-BP,ISSA-BP 4种短时交通流预测模型的预测效果进行对比分析,结果显示,ISSA-BP短时交通流预测模型的误差最小,ISSA-BP模型相较BP模型在MAE评价指标上的预测精度提升了48.85%,有着更好的预测精度。

The short-term traffic flow prediction by BP neural network is too dependent on the initial parameters.In order to solve this problem and optimize the BP neural network,a short-term traffic flow prediction model(ISSA-BP)is proposed based on the improved sparrow search algorithm(ISSA).Since the standard sparrow search algorithm(SSA)is easy to converge at the origin and fall into local optimum,the position update formulas of the discoverers and of some joiners in the sparrow group are improved respectively.The improved discoverers will perform a global search based on the size of the search dimension and the position of the current optimal value,and some joiners will perform a global search according to the distance between the optimal position and themselves.The prediction effects of four short-term traffic flow prediction models,BP,PSO-BP,SSA-BP and ISSA-BP,are compared and analyzed through experiments.The results show that the error of ISSA-BP short-term traffic flow prediction model is the smallest,and the prediction accuracy of ISSA-BP model is much better,48.85%higher than that of BP model in terms of MAE evaluation indicators.