Study On Traffic Flow Prediction And Congestion State Clustering Based On TS-GCN And SAGA-FCM Algorithms

The rapid development of cities has brought convenience to people’s lives,but it has also led to the problem of urban traffic congestion.Urban traffic congestion not only causes direct or indirect economic losses but seriously affecting the happiness of urban residents’ lives.Predicting and identifying traffic conditions can effectively assist residents in planning travel routes,saving travel time,and alleviating urban traffic congestion.Therefore,predicting and identifying traffic conditions have become important and practical methods in urban governance,attracting the attention of many scholars.To predict and identify traffic states,analysis and prediction of traffic flow data are required.Due to the combined effects of multiple complex factors on traffic flow and the complexity of urban road networks,the relationship between traffic states and traffic flow,vehicle speed,and time occupancy rate exhibits a highly nonlinear nature,increasing the complexity of traffic flow data prediction and identification.Traditional traffic flow prediction mainly relies on regression analysis of time series data.However,due to the strong nonlinearity and data sparsity of traffic flow data,traditional prediction and state recognition methods perform poorly in handling complex traffic flow scenarios and nonlinear relationships.In recent years,with the increasing computational power,deep learning has been widely applied in data analysis and prediction.For traffic flow data,deep learning methods can extract various information containing spatiotemporal features and intelligently integrate them,significantly improving the accuracy of short-term predictions.To accurately predict and identify traffic conditions,this article presents a rapid prediction and identification of traffic state information using two models: a traffic flow prediction model based on TS-GCN(Temporal-Spatial Graph Convolutional Neural Network)and a state recognition model based on SAGA-FCM(Spatiotemporal Attribute Graph Aggregation-Fuzzy C-Means)clustering algorithm.The TS-GCN model includes an input layer,stacked spatiotemporal convolutional network layers,and an output layer.The stacked spatiotemporal convolutional network layers consist of gate time convolution layers,spatiotemporal convolution blocks,and spatiotemporal attention blocks.The TS-GCN model considers time information and spatial position as nodes and connects them into a graph.By performing convolution operations on the graph and using the feature vectors of each node and adjacent node for information propagation and updating,a comprehensive spatiotemporal modeling result is obtained that captures the spatiotemporal information in traffic flow data in multiple dimensions.The SAGA-FCM model is based on the fuzzy C clustering(FCM)algorithm and improves the optimization of the membership matrix process,which is prone to local optimal values and difficult to converge,by introducing the genetic simulated annealing(SAGA)algorithm.By manually labeling the clustered data,the SAGA-FCM model can classify traffic flow data into five different traffic states.This article uses the TS-GCN model to predict future traffic information using historical traffic flow data,and the predicted results are input into the SAGA-FCM model to achieve status recognition of the predicted data.By combining the TS-GCN model and the SAGA-FCM model,this article achieves the prediction and identification of traffic conditions.Through the Pe MS04 dataset and the Pe MS08 dataset,the algorithm proposed can achieve a relative prediction error of 2.03% in short-term forecasting.Among 288 daily data,the state recognition and prediction of the proposed model only had 9 errors,and these erroneous data were all located in adjacent states.The effectiveness and accuracy of the proposed method are verified,which provides a new idea and method for solving the problem of urban traffic congestion.