Long-Time gap crowd prediction with a Two-Stage optimized spatiotemporal Hybrid-GCGRU. (October 2022)

• Record Type: Journal Article
• Authors: Cheng, Jack C.P.; Ho Poon, Kwok; Kok-Yiu Wong, Peter
• Is Part Of: Advanced engineering informatics. Volume 54(2022)
• Publication Date: 2022-10
• Subjects: Deep learning -- Graph neural network -- Long-time gap crowd prediction -- Spatiotemporal feature extraction -- Time-series modeling; Computer-aided engineering -- Periodicals; Engineering -- Data processing -- Periodicals; 620.00285
• DOI: 10.1016/j.aei.2022.101727
• Languages: English
• ISSNs: 1474-0346

Abstract

Highlights: Proposed two-stage framework for long-time gap crowd prediction 1 week ahead. Proposed GCGRU model to exploit spatiotemporal patterns for multi-location crowd prediction. Proposed hybrid model addressing fluctuating crowd volume variation during holidays. Different baseline models improved generally by 40% in prediction accuracy. Validated high adaptability of the framework to crowd volumes in different periods. Abstract: Crowd prediction is a crucial aspect of modern society, facilitating numerous decision-making processes, such as hazard detection and facility maintenance. Conventional crowd prediction relies on rule-based models which involve tedious formation steps and are error-prone. Hence, this study adopts spatiotemporal deep learning models for crowd prediction in order to exploits both temporal and spatial features of the dataset. There are three major limitations in previous studies: firstly, the prediction length of the time series models is relatively short (i.e., limited to single-digit time steps); Secondly, most studies using time series prediction do not consider the features in the neighboring locations other than the target location; Finally, only the temporal pattern on regular days is exploited, whilst the data variations during holidays can greatly reduce the prediction accuracy. Therefore, a particle swarm optimized (PSO) Hybrid-Graph Convolutional Gated Recurrent Unit (HGCGRU) model, entitled PSO-HGCGRU, is proposed to increase the prediction accuracy in various situations. The PSO-HGCGRU model is composed of two parts, 1) a hybrid model structure, which allows the model adapting to both regular and holiday features, and 2) the GCGRU model, in which the graph convolutional layer to extract the spatial features and the GRU layers to extract the temporal dependencies, to undertake spatiotemporal prediction. The Two-Stage Long-Time Gap Prediction method is applied to the proposed PSO-HGCGRU model with a newly designed two-stage mechanism to optimize the length of the time gap. Results show that the proposed model generally outperforms the baseline models by around 40% in accuracy.