Spatiotemporal sequence data mining: traffic flow prediction algorithm based on Recurrent Neural Networks

Traffic congestion is one of the most common problems in the transportation system. In urban planning and construction, traffic congestion increases the difficulty of control and scheduling, hindering the pace of urban modernization development. The development of intelligent transportation systems has to some extent solved these problems, among which traffic flow prediction is particularly crucial. A period processing layer containing concatenation, dimension amplification, and diffusion convolution was studied and designed. On the grounds of this, a graph neural network prediction model combining spatiotemporal period characteristics was designed. And it further introduced a traffic flow prediction method on the grounds of dynamic topology maps. This model combined adaptive topology generation and parameter learning mechanisms, synchronously processing the complex correlation of traffic data and the heterogeneity between nodes through gated recurrent neural networks. The outcomes indicated that the research model is more excellent than other graph learning methods in three evaluation indicators: mean square error, mean absolute percentage error, and root mean square error. Their values were 15.63%, 9.87%, and 25.08%, respectively. The evaluation on all four datasets showed excellent performance of the research model, confirming the effectiveness of its adaptive topology and parameter learning strategy. This study had important practical significance for improving the traffic management system, alleviating traffic pressure, and enhancing the travel experience of urban residents.