基于双向长短期记忆网络的城市快速路合流区车速预测

Traffic Speed Prediction in Merging Zone of Urban Expressway Based on Bidirectional Long Short-Term Memory Network

非典型复杂场景微观交通参数的准确预测是保证车路协同系统(IVICS)稳定运行的前提.为解决IVICS条件下合流区高峰时段瓶颈现象所致的车速分布紊乱而不易预测的问题,首先,基于无人机高空视频,从广域视角提取交织区高峰时段全样本高精度车辆轨迹数据;然后,考虑双向长短期记忆网络(bidirectional long short-term memory,Bi-LSTM)时间较长且人工设置训练参数对模型预测性能影响较大,提出基于贝叶斯超参数(bayesian hyperparameters optimization,BHO)优化的BHO-Bi-LSTM车速预测集成模型;最后,构建经典多元线性回归车速预测模型、Bi-LSTM车速预测模型作对比.结果表明:BHO-Bi-LSTM模型表现最优,拟合优度、秩相关度分别为91.05%、94.87%,误差均值、误差的标准差、均方误差、均方根误差、归一化均方根误差分别为0.056 1、0.455 6、0.210 6、0.458 9、0.078 5,有效改善了合流区高峰时段车速特性复杂而导致不易预测的缺陷.

Accurate prediction of microscopic traffic parameters in atypical complex scenes is a prerequisite to ensure stable operation of the intelligent vehicle infrastructure cooperative systems(IVICS).To solve the problem of vehicle speed distribution disorder and difficulty in prediction caused by bottleneck phenomenon during peak hours in the merging area under IVICS conditions,First,using the UAV video,the full-sample high-precision vehicle trajectory data of the intertwined area during peak hours are extracted from a wide-area view.Then,as bidirectional long short-term memory(Bi-LSTM) networks cost long time and affect the prediction performance of the model when training parameters are manually set,a BHO-Bi-LSTM(bayesian hyperparameter optimization bidirectional long short-term memory) integrated vehicle speed prediction model based on Bayesian hyperparameters optimization is proposed.Finally,the classical multiple linear regression model and Bi-LSTM model of vehicle speed prediction are constructed for comparison.The results show that the BHO-Bi-LSTM model outperforms other models,with a goodness-of-fit and rank correlation of 91.05% and 94.87%,respectively,and error mean,error standard deviation,mean square error,root mean square error,and normalized root mean square error of 0.0561,0.4556,0.2106,0.4589,and 0.0785,respectively,which can overcome the disadvantage in prediction of complicated traffic speeds during peak hours.