基于卷积神经网络的交叉口过车速度短时预测模型研究

Research on Short-term Speed Prediction of Intersections Based on Convolutional Neural Network

交叉口是城市交通网络系统的关键节点,在绿灯通行环境下的交叉口过车速度可以精准地反应出路网的畅通程度。该文基于长时间、大样本的车辆轨迹数据,利用卷积神经网络,将数据处理为包含其特征的矩阵,构建车速时序预测模型,能够高效、精准地预测出交叉口过车速度。实验结果表明,MAE平均值为2.06,MAPE平均值为10.57%,说明模型具有较好的预测效果,可以为警力的提前派遣和交叉口的高效疏堵提供数据支撑。

Intersections are the key points of urban transportation network system.In a green traffic environment,the passing speed of intersections can accurately reflect the smoothness of the road network.In order to predict the short term vehicle speed at the intersection,we process the traffic data into a matrix which contains characteristics based on the large vehicle trajectory data and construct a convolutional neural network.The experimental results show that the average MAE is 2.06 and the average MAPE is 10.57%,which mean the model has a good predictive effect and can provide data support for the early dispatch of police force and the efficient dredging of intersections.