A Spatio-Temporal Hybrid Neural Network for Crowd Flow Prediction in Key Urban Areas

Abstract

The prediction of crowd flow in key urban areas is an important basis for city informatization development and management. Timely understanding of crowd flow trends can provide cities with data support in epidemic prevention, public security management, and other aspects. In this paper, the model uses the Node2Vec graph embedding algorithm combined with LSTM (NDV-LSTM) to predict crowd flow. The model first analyzes the correspondence between key areas and grid centers, and the Node2Vec graph embedding algorithm was used to extract spatial features. At the same time, considering urban region type, weather, temperature, and other crowd flow data features, the long short-term memory (LSTM) network model was used for unified modeling. The model uses the crowd flow of the previous three days to predict the crowd flow of the next day. The model was evaluated on the 2020 CCF crowd density competition data set. The experimental results show that the NDV-LSTM model can capture the features of the region association digraph and various crowd flow correlation factors well, and the mean square error of the prediction of the crowd flow in key areas is reduced to 1.5194.