Crowd prediction is a crucial aspect of modern life with innumerable applications. By predicting future human occupancy in advance, crowd prediction can support the decision-making processes of facility stakeholders, e.g., the campus operator can schedule facility maintenance during the period of lowest pedestrian flow to eliminate any disturbance. Conventional crowd prediction utilizes statistical models and rule-based data mining techniques, which are tedious in data processing and error-prone. Hence, this research contains two stages formulating crowd prediction into a time series analysis based on deep learning.

In the first stage, the deep learning time series model is adopted to conduct crowd prediction. Despite the wide adaptability of deep learning-based time series techniques i...[ Read more ]

Crowd prediction is a crucial aspect of modern life with innumerable applications. By predicting future human occupancy in advance, crowd prediction can support the decision-making processes of facility stakeholders, e.g., the campus operator can schedule facility maintenance during the period of lowest pedestrian flow to eliminate any disturbance. Conventional crowd prediction utilizes statistical models and rule-based data mining techniques, which are tedious in data processing and error-prone. Hence, this research contains two stages formulating crowd prediction into a time series analysis based on deep learning.

In the first stage, the deep learning time series model is adopted to conduct crowd prediction. Despite the wide adaptability of deep learning-based time series techniques in various research fields, they are seldom adopted in crowd prediction. There are two major limitations in previous studies: firstly, the prediction accuracy notably degrades with increased prediction length, and secondly only the temporal pattern along a single time dimension is exploited, i.e., the consecutive time steps in the most recent input data. Therefore, a Long-Time Gap Two-Dimensional method, entitled LT2D-method, is proposed to increase the crowd prediction length with high accuracy. The LT2D-method is composed of two parts, 1) Long-Time Gap Prediction, which extends the prediction length to 240 time steps (1 day) with high accuracy, and 2) 2D inputs method, which exploits the prior knowledge from different time dimensions to further improve the prediction accuracy of Long-Time Gap Prediction. The proposed LT2D-method can be generally adapted to deep learning models, such as LSTM, BiLSTM, and GRU, to improve the prediction accuracy. By incorporating the proposed LT2D-method into different baseline models, the accuracy is generally improved by around 22%, demonstrating the robustness and generalizability of the method.

In the second stage, spatiotemporal deep learning models are adopted for crowd prediction in order to exploit both temporal and spatial features of the dataset. There are two major limitations in previous studies: Firstly, most studies using time series prediction do not consider the features in the neighbouring locations other than the target location; Secondly, 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 to adapt to both regular and holiday features, and 2) the GCGRU model, in which the graph convolutional layer extracts the spatial features and the GRU layers to extract the temporal dependencies, to undertake spatiotemporal prediction. The Two-Stage Long-Time Gap Prediction method, which improves the Long-Time Gap Prediction method in the first stage, 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