GCN-PSO: A Hybrid Graph Convolutional and Particle Swarm Optimization Framework for Urban Traffic Flow Forecasting

Abstract

With the acceleration of urbanization and the increase in car ownership, traffic management plays a crucial role in the urbanization process. Traditional traffic flow prediction methods are mainly based on historical data and statistical models. The Traffic Flow Forecasting Dataset was selected with data from 36 sensors on two highways in the Northern Virginia/Washington, D.C., U.S. Capital Region, measured every 15 minutes and covering 47 characteristics, including historical traffic volume sequences, time, roads, and more. The GCN part of the model architecture is set up with two layers, the first layer preliminarily extracts the spatial correlation of transportation network nodes, and the second layer further excavates the deep spatial dependence, and the input dimension is 47 dimensions, which corresponds to the feature dimension of the dataset. In the optimization process, the parameters of the GCN are optimized by the PSO algorithm, and the learning rate, convolution kernel and other parameters are adjusted to improve the accuracy of the model's prediction of urban traffic flow.By analyzing the changes in traffic flow in historical traffic data, a statistical model is established to predict future traffic flow. Standard methods include time series analysis, regression analysis, and neural networks. However, these methods have significant limitations regarding prediction accuracy and real-time performance and cannot adapt to the dynamic changes in the transportation system. Therefore, this study proposes a city traffic flow prediction model based on the combination of Graph Convolutional Network (GCN) and Particle Swarm Optimization (PSO) algorithm. GCN captures spatial dependencies in the traffic network, and the PSO algorithm is used to optimize model parameters and improve prediction performance. The research experimental results show that compared to a single GCN model, the optimized GCN-PSO model has significantly improved prediction accuracy, with a 15.3% reduction in mean square error (MSE) and a 12.7% reduction in mean absolute error (MAE). In real-time prediction scenarios, the response time of the GCN-PSO model was reduced by 8.9%, effectively improving prediction efficiency. Meanwhile, analyzing traffic data from different cities and time periods verified the universality and stability of the GCN-PSO model in various scenarios.