Transfer learning with a spatiotemporal graph convolution network for city flow prediction

Abstract

Recently, deep learning based city flow prediction has been extensively used in the establishment of smart cities. These methods are data-hungry, making them unscalable to areas lacking data. Although transfer learning can use data-rich source domains to assist target domain cities in city flow prediction, the performance of existing methods cannot meet the needs of actual use, because the long-distance road network connectivity is ignored. To solve this problem, we propose a transfer learning method based on spatiotemporal graph convolution, in which we construct a co-occurrence space between the source and target domains, and then align the mapping of the source and target domains’ data in this space, to achieve the transfer learning of the source city flow prediction model on the target domain. Specifically, a dynamic spatiotemporal graph convolution module along with a temporal encoder is devised to simultaneously capture the concurrent spatiotemporal features, which implies the inherent relationship among the road network structures, human travel habits, and city bike flow. Then, these concurrent features are leveraged as cross-city invariant representations and nonlinearly spanned to a co-occurrence space. The target domain features are thereby aligned with the source domain features in the co-occurrence space by using a Mahalanobis distance loss, to achieve cross-city bike flow prediction. The proposed method is evaluated on the public bike flow datasets in Chicago, New York, and Washington in 2015, and significantly outperforms state-of-the-art techniques.

1. Introduction

In recent years, many countries and regions have been studying the construction of smart cities (Tascikaraoglu, 2018), hoping to grasp real-time data such as urban air quality (Kang et al., 2018) and urban flow (Xu et al., 2023) to predict the future state and achieve efficient management of the city. City flow prediction is an important part of a smart city, and can help government departments effectively control traffic congestion and pollutant emissions. The main task of city flow prediction is to accurately predict the corresponding state in the future using the given historical traffic flow.

The existing city flow prediction methods are divided mainly into the parametric model and the non-parametric model. Parametric models (Van Der Voort et al., 1996) refer to a model based on certain assumptions, usually with a fixed structure, and parameters are calculated based on empirical data. Non-parametric models (Mallick et al., 2019; Zhao et al., 2020) are represented by deep learning models, which can effectively capture nonlinear spatiotemporal features. They both have a strong dependence on historical data. When historical data are very scarce, these models often perform poorly. Considering that not every city has a large amount of historical data, when the target city has only a small amount of historical data, the means to use data-rich source cities to help target cities improve the accuracy of flow prediction is a problem worthy studying. The main challenge involved in tackling this problem comprises the fact that the source-domain and target-domain cities have different spatial structures, resulting in different spatiotemporal distributions of their flow, with the result that the source domain cannot directly assist in the prediction of the target domain.