A Deep Spatiotemporal Fuzzy Neural Network for Subway Passenger Flow Prediction With COVID-19 Search Engine Data

Abstract

Passenger flow prediction is of great significance in the operation and management of subways, especially in reducing energy consumption and improving service quality. Due to the impact of COVID-19, subway passenger flow fluctuates a lot, which makes passenger flow estimation or forecasting a very challenging task. This article mainly carries out two aspects of work to solve the task of subway passenger flow prediction under pandemic. First, this article introduces search engine data as a new data source and provides a systematic method to extract valid quires and search volumes that are closely associated with subway passenger flow under pandemic. Second, this article combines the fuzzy theory and neural network to propose a deep learning architecture called “deep spatiotemporal fuzzy neural network” to deal with the complex spatiotemporal features and uncertain external data of subway passenger flow prediction. Experiments on the actual dataset of the Beijing subway prove the superiority of the model and the effectiveness of search engine data in subway passenger flow forecasting.