Integrating Spectral Clustering and Hybrid CNN‐LSTM‐PSO Model for Short‐Term Passenger Flow Prediction in Urban Rail Transit

With the rapid development of model cities, urban rail transit (URT) systems have emerged as a crucial component in urban public transport, and passenger flow prediction serves as the cornerstone for planning the travel, avoiding the congestion, and improving the travel efficiency. In order to predict short‐term passenger flow of the URT system, a hybrid convolutional neural network (CNN)‐long short‐term memory (LSTM)‐particle swarm optimisation (PSO) model is proposed to accommodate both the spatial and temporal features of passenger flow. First, spectral clustering is employed to extract four different types of stations, in which the Calinski–Harabasz (CH) index is considered. Second, the hybrid CNN‐LSTM‐PSO model is constructed to predict the short‐term passenger flow for different types of stations, in which CNN can extract the abstract feature with a multi‐layer convolutional structure, LSTM can deal with time series data, and the PSO algorithm is employed to optimise some parameters. Third, the data from Hangzhou urban rail transit in 2019 are employed for prediction. The results show that the proposed hybrid model reveals the best performance in accuracy by comparing the equivalent CNN‐LSTM, LSTM and autoregressive integrated moving average (ARIMA) models. At last, some empirical suggestions are provided to benefit both the passengers and operation and management departments of the URT system.