Short‐term passenger flow forecasting using CEEMDAN meshed CNN‐LSTM‐attention model under wireless sensor network

Abstract

For a long time, the accurate prediction of passenger flow can provide early warning information for various industries such as the public service industry, tourism industry, and industrial business, thus opportunely arranging passengers and providing homologous services to relieve the overloading of places and the accidents caused by overcrowding of people. In recent years, by using the wireless sensor network to sense the passenger data in advance, the technique of machine learning and neural networks has been utilized to assist the short-term passenger flow prediction. In this study, building on convolutional neural network (CNN) and long short-term memory network (LSTM), a complete ensemble empirical mode decomposition with adaptive noise algorithm (CEEMDAN) and attention-based CNN-LSTM network to extract both temporal and spatial characteristics of passenger flow data, is proposed. Moreover, the problem of the inaccuracy of the noise part is properly solved by adding the CEEMDAN algorithm to the input layer. With the proposed network structure, the CNN-LSTM network is replaced with the Conv-LSTM network to reduce the information loss and get a further performance improvement. The result shows that 39% performance improvement can be achieved than the case with a single LSTM network, and 28% performance improvement can be achieved than the CNN-LSTM network.