基于GWO-LSTM模型的道路交通事故量预测

Road traffic accident prediction based on GWO-LSTM model

为降低道路交通事故率,减少事故损失,采用全局遍历性和收敛性较强的自适应学习策略灰狼优化(grey wolf optimizer,GWO)算法,对长短期记忆(long short term memory,LSTM)神经网络中的初始学习率、隐藏层节点数、正则化系数等参数进行优化训练,构建GWO-LSTM道路交通事故量预测模型。以2000—2019年美国道路交通致死事故数据为样本数据,分别采用月粒度、周粒度、时粒度划分交通事故数据,对比分析GWO-LSTM模型、自回归移动平均(autoregressive moving average,ARMA)模型、反向传播(back propagation,BP)神经网络和LSTM神经网络的道路交通事故预测结果。结果表明:在3种时间粒度下,GWO-LSTM模型预测结果的平均绝对百分比误差和均方根误差均最小,预测准确度较高,可用于道路交通事故量预测中。

In order to reduce the road traffic accident rate and minimize accident losses,the gray wolf optimizer(GWO) algorithm,which has strong global traversal and convergence properties,is used to optimize the initial learning rate,hidden layer nodes,regularization coefficient and other parameters in the long short term memory(LSTM) neural network.The GWO-LSTM road traffic accident prediction model is constructed.Taking the road traffic fatal accident data from 2000 to 2019 as the sample data,the traffic accident data is divided into monthly granularity,weekly granularity,and hourly granularity.The road traffic accident prediction results of the GWO-LSTM model,autoregressive moving average(ARMA) model,back propagation neural network(BPNN) model,and standard LSTM model are compared and analyzed.The results show that under the three time granularities,the GWO-LSTM model has the smallest average absolute percentage error and root mean square error,and high prediction accuracy.It can be used for road traffic accident prediction.