Short-time traffic flow combination prediction based on CEEMDAN-PE-IGWO-BIGRU

We propose a combined prediction model based on CEEMDAN-PE-IGWO-BIGRU to improve the prediction accuracy of short-term traffic flow, CPIB. Firstly, after completing the preprocessing of the raw traffic flow data, we apply the adaptive noise full set empirical modal decomposition technique in order to extract the trend and detail information in different modal sequence components. In order to further analyse the stochastic nature of these modal components, we introduce the alignment entropy algorithm to calculate the alignment entropy values of the modal components in each frequency band. In view of the parameter optimisation difficulties of the BIGRU neural network, the parameter configuration of the BIGRU model is optimized using an enhanced Grey Wolf Optimization algorithm. This optimization process resulted in an improvement in the model's prediction accuracy. And we reconstruct final prediction output by adding the predictions of modal components in different frequency bands together. The experimental results show that the combined CEEMDAN-PE-IGWO-BIGRU prediction model achieves significant results in the MAPE, MSE, RMSE, MAE and R2 metrics on the Sg-A dataset, with values of 14.6725, 14.4718, and 7.5646, respectively, which indicate that the model effectively solves the problems of instability, nonlinearity, and the inherent non-linearities in the traffic flow dynamics, instability, nonlinearity, noise problems.