Spatio-temporal grey model based on tensor Tucker decomposition for traffic flow prediction

Abstract

Short-term traffic flow prediction plays an increasingly important role in intelligent transportation systems, providing a decision-making basis for traffic signal optimization, route guidance, and emergency control, thereby improving the efficiency of road network operation. However, the traffic flow system is a complex dynamic system, and its data has complex spatio-temporal multi-modal characteristics. The existing grey models take time series and matrix series as the input of metadata, ignoring the multi-modal characteristics of traffic flow data, and have certain limitations in capturing the spatial correlation and dynamic spatio-temporal interaction of traffic flow. This paper introduces tensors with high-dimensional multi-modal features as metadata and establishes a novel tensor-structured discrete grey model. The new model reconstructs the multi-dimensional spatio-temporal attributes of traffic flow data and traditional multi-segment traffic data, follows the discrete grey model's modeling mechanism and prediction method, and retains the advantages of fast calculation efficiency and high prediction accuracy of the grey model. It also makes up for the defect that the grey model can only handle small-scale data, providing a new idea for the multi-dimensional prediction extension of the grey model. To illustrate the effectiveness of the new model, traffic data from multiple periods and multiple road sections are simulated and predicted. The simulation effectiveness index (Mean Absolute Percentage Error) values of the new model in different periods of the same road section and the same period of different road sections are all around 5 %, and it also has a better performance compared with the other three-time series prediction models.