Optimization of spatial-temporal graph: A taxi demand forecasting model based on spatial-temporal tree

Abstract

Taxi is one of the important means of transportation for people's daily travel activities, and it is one of the important research objects of intelligent transportation system. Taxi demand forecasting research can promote the application of urban transportation basic services and the transportation department to analyze and allocate transportation resources more reasonably. Graph structure is an important method for capturing spatial correlations among urban regions. However, it has certain limitations in capturing the hierarchical features and the local path features of regional nodes. Additionally, existing research has failed to capture multiple factors influencing changes in taxi demand. Therefore, this study proposes a spatial-temporal model based on capturing multi-factor features. The model innovatively uses the tree structure as a topology structure and proposes the tree convolution for constructing data spatial distribution features. The spatial-temporal convolution module with tree convolution as the core can effectively capture the hierarchical features and the local path features among area nodes. In this study, four factors affecting taxi demand are designed. The deep features of the four factors are further fused through the spatial-temporal convolution module. The model integrates multiple influencing factors affecting taxi demand from the spatial-temporal level and shows certain advantages in experiments. Compared with existing baselines, the model designed in this paper shows certain advantages in three real urban taxi datasets.