Traffic Flow Prediction Based On Spatial-Temporal Characteristic Analysis

Traffic flow prediction is the cornerstone of the development of intelligent transportation systems,which aims to forecast the future conditions of urban traffic systems(e.g.,traffic flow and speed)and plays a vital role in traffic scheduling and management.Given the spatial-temporal characteristics of traffic flow,this paper analyzes the characteristics of different types of traffic flow data and constructs traffic flow prediction algorithms based on spatial-temporal characteristics analysis to achieve accurate traffic flow prediction.It is committed to extracting traffic flow patterns from historical data,sensing the future traffic situation,and providing the theoretical basis and technical support for traffic control and early warning control.The main contributions and innovations of this paper are summarized as follows:1.Data preprocessing.For the taxi GPS trajectory big data,the elastic distributed data set(RDD)is used for data reading,data sorting,data statistics,data integration,and data storage.Kalman Filter(KF)is employed to smooth the data and reduce the influence of noise data on prediction accuracy.For the traffic data of the highway network,the linear interpolation method is utilized to fill in the missing values,and the data is normalized by the Z-score standardization to improve the accuracy and the convergence rate of the algorithm during the training period.To fully capture the periodic characteristics of the traffic flow,the data is divided into three components based on three different time scales: recent period,daily period,and weekly period.2.Traffic flow prediction based on the DSA-DBLSTM algorithm.To solve the problem of inadequate modeling of traffic flow time pattern similarity and deep nonlinear characteristics,a deep bidirectional long short-term memory network optimization algorithm(DSA-DBLSTM)based on Fast Dynamic Time Warping(FastDTW)and attention mechanism is presented.Specifically,the distributed elastic data set(RDD)is exploited to process the taxi GPS trajectory data,and Kalman Filter(KF)is used to eliminate the noise data.Next,the DSA-DBLSTM algorithm is constructed to improve the accuracy and efficiency of traffic flow prediction,combined with the FastDTW algorithm and attention mechanism for implementing the optimization of DBLSTM.Finally,the performance of the DSA-DBLSTM algorithm is evaluated by the real-world GPS trajectories of taxicabs.3.Traffic flow prediction based on the ASTA-DGCN algorithm.To effectively mine the complex and dynamic topology structure of road networks and discover the dynamic spatial-temporal characteristics,an attention-based spatial-temporal adaptive dual graph convolutional network algorithm(ASTA-DGCN)is proposed.Specifically,the spatial-temporal attention module is introduced to extract the dynamic temporal dependence characteristics and the implicit influence of weights among road network nodes and to capture the dynamic influence of different spatial-temporal positions on the current spatial-temporal situation.Next,the adaptive graph modeling module is used to adaptively extract the sparse graph adjacency matrix based on the traffic flow data,and the FastDTW algorithm is adopted to measure the similarity of road network nodes.The adjacency matrix graph is designed to encode the semantic relevance to enhance the feature extraction ability of the traffic network topology.Finally,temporal and spatial correlations are explicitly modeled using dual graph convolution and time convolution based on the obtained graphs to mine the spatial-temporal patterns in dynamic traffic flow.The performance of the ASTA-DGCN algorithm is validated with the real-world highway traffic data.