Real-Time Spatiotemporal Correlation Feature Analysis of Traffic Flow Based on CEEMD+BiGRU Combination Model

In the application process of real-time traffic flow data, the main reason affecting the analysis of spatiotemporal correlation features is the overlapping distribution of its own characteristic modes, which leads to poor representation of spatiotemporal features and the problem of inability to fit traffic flow with true values, failing to meet the requirements of confidence interval distribution. This paper proposes a CEEMD BiGRU combination model and uses the IMF components obtained by decomposing traffic flow data into CEEMD to represent spatiotemporal properties. A bidirectional time series model is constructed using BiGRU, and multi-scale features are used as inputs to fit traffic flow and true values. By bidirectionally calculating the hidden states of multi-scale features and considering the distribution requirements of confidence intervals, the spatiotemporal dependencies related to traffic flow are correlated and output. The case shows that the output flow of this method is highly consistent with the true value, which can improve the accuracy of prediction.