基于GRU递归神经网络的城市道路超车预测

Overtaking prediction of urban road based on GRU neural network

城市道路中的超车行为尤其是违规超车等,对交通秩序与安全造成严重影响。随着电子警察与卡口等车牌识别系统的广泛应用,通过上下游车牌识别与时间对比,可以较为精确地获取车辆在路段之间的超车关系。基于电子警察处理的超车数据建立了基于GRU递归神经网络的城市道路超车率预测模型,预测城市道路超车率的变化趋势,并与循环神经网络(recurrent neural network,RNN)、反向传播(back propagation,BP)神经网络进行对比。在苏州工业园区星湖街-现代大道路段的测试结果表明,基于GRU递归网络的超车预测模型的绝对值误差为12.52%,相比于其他2种模型,精度高、泛化能力强、鲁棒性强。

Overtaking behavior in urban roads,especially violations of regulations,has a serious impact on traffic order and safety.With the wide application of the license plate recognition system such as electronic police and bayonet,the overtaking relationship between the road segments can be accurately obtained through the comparison of the upstream and downstream license plate recognition and time.Based on the overtaking data processed by electronic police,this paper establishes a prediction model of overtaking rate on urban roads based on gated recurrent unit(GRU),which is used to predict the trend of overtaking rate on urban roads,and compared with the model of recurrent neural network(RNN)and back propagation(BP)neural network.The test on Xinghu Street-Modern Highway Section in Suzhou Industrial Park shows that the absolute error of the overtaking prediction model based on GRU network is 12.52%,which has higher precision,stronger generalization and robustness compared with the other two models.