A Short-term Traffic Flow Prediction Method Based on Time Series Data Decomposition and Reconstruction

In order to extract signal components with rich feature information from short-term traffic flow data and further improve the prediction accuracy, a short-term traffic flow prediction method based on temporal data decomposition reconstruction is constructed by combining the parameter optimization based variational mode decomposition (VMD), recurrence quantification analysis (RQA), and bidirectional gated recurrent unit (BIGRU) models. The osprey cauchy sparrow search algorithm (OCSSA), which integrates the osprey and cauchy variants, is used to determine the number of modal components and the penalty factor of the variational modal decomposition, and to obtain the relatively smooth intrinsic modal components. The decomposed modal components are reconstructed into the deterministic components, fluctuating components, and trend components through the recursive quantitative analysis. Based on this, for each reconstructed component the BIGRU prediction model is constructed, and the predicted values of each reconstructed component are nonlinearly integrated using the BIGRU prediction model to obtain the final prediction results. The measured data of the flow of Shanghai North-South Expressway and California Expressway network are used for validation, The results show that in the NBDX08(1) dataset, the corresponding mean absolute error, root-mean-square error, and mean absolute percentage error are reduced by 29.1%, 24.5%, and 46.1% on average, respectively, compared with the other models; and the errors in the dataset of No. 760101 are reduced by 19.05%, 19.69%, and 16.46% on average. These verify that the proposed method for the decomposition and reconstruction of different components can accurately capture and learn the characteristics of traffic flow components, which further improves the prediction accuracy while controlling the computational complexity of the model.