The Short-Term Traffic Flow Prediction Based on an Improved Data Decomposition Integration Strategy

In recent years, with the advancement of the economy and technology, more and more people choose to travel by means of transportation, which undoubtedly brings certain pressure to road traffic. Short-term traffic flow forecasting is an effective tool to alleviate traffic congestion and improve people's travel efficiency. Its performance faces essential challenges with the consideration of complex traffic flow. To this end, we design a new multi-stage decomposition and integration framework to deal with complex data before prediction. It divides the original sequence into stable and non-stationary parts by the CEEMDAN decomposition algorithm and Hodrick-Prescott filter, and further uses VMD decomposition for the non-stationary part. In order to improve the training efficiency of the framework, we creatively propose to use the clustering algorithm based on K-means and predict the clustered subsequences. We conduct multiple experiments by using three UK highway traffic flows as examples and compare them with other decomposition models using multiple ESN models, effectively demonstrating the effectiveness and stability of our framework.