IMPROVED METRO PASSENGER FLOW FORECASTING USING ADAPTIVE FEATURE FUSION NETWORKS

Abstract

Predicting Origin-Destination (OD) passenger flow accurately helps improve the effectiveness and quality of metro services. Predicting incoming and outgoing flows for individual stations has been the main focus of previous efforts; OD prediction in metro networks has received less attention. The difficulties arise from the fact that OD flows 1) have complicated geographical correlations and high temporal dynamics, 2) are influenced by outside variables, and 3) contain sparse and partial data slices. In order to a) adaptively fuse spatial dependencies from various knowledge-based graphs and even hidden correlations between stations, and b) accurately capture the periodic patterns of passenger flows based on the auto-learned impact from external factors, we propose an Adaptive Feature Fusion Network (AFFN) in this paper. To address the sparsity and incompleteness of OD matrices, we further enhance the accuracy of OD prediction by extending AFFN to multi-task AFFN, which predicts each station's intake and outflow as a side-task. Two real-world metro trip datasets gathered in Nanjing and Xi'an, China, were the subjects of our intensive research. The evaluation findings demonstrate the efficacy of AFFN and all of its essential components in OD prediction, as our AFFN and multi-task AFFN exceed the state-of-the-art baseline approaches and AFFN variations in a variety of accuracy measures.