多种残差补偿的贝叶斯网络下的短期交通预测

Short-term traffic prediction based on Bayesian networks with multiple residual compensations

为了解决道路车流量的数据生成条件时变场景下的交通预测问题,本文建立道路交通控制与交通流预测数据之间的联系,提出一种基于多种残差补偿的贝叶斯网络的短期交通预测方法。提取城市中大规模多路口主干道车道及车辆信息构造多个平行的贝叶斯网络,使用贝叶斯关系及期望最大化算法进行短期交通预测。再通过数据自相关残差补偿、车辆换道和多路口连通性的线性残差补偿提高了预测的精度,解决了传统研究对相邻路口和换道导致的误差等因素处理能力不足的问题。仿真结果表明:使用贝叶斯网络预测交通流,并基于车辆行为的残差进行精度补偿,可以更准确地预测复杂的交通演化场景的短期交通流。

The change in time-varying factors in traffic evolution has a great impact on the data generation condi-tions of road traffic flow.To determine the influence of these factors,this paper establishes the relationship between road traffic control and traffic flow prediction data and proposes a short-term traffic prediction method based on the Bayesian network(BN)with multiple residual compensations.The lane and vehicle information of large-scale multi-intersection trunk roads in the city are extracted to construct multiple parallel BNs,and the Bayesian relation-ship and expectation maximization(EM)algorithm are used for short-term traffic prediction.Then,the prediction accuracy is improved through data autocorrelation residual compensation,vehicle lane changes,and linear residual compensation of multi-intersection connectivity,which solves the problem of insufficient processing ability of tradi-tional research on factors such as errors caused by adjacent intersections and lane change.The simulation results show that the short-term traffic flow in complex traffic evolution scenarios can be predicted more accurately using BN,compensating for the accuracy based on residual vehicle behavior.