Abstract:Subways have now reached an irreplaceable position in urban rail transportation. The aggregation of people choosing subway travel at the same time can easily lead to passenger congestion, so predicting passenger flow in advance is crucial. This study used LSTM (Long Short-Term Memory) as the data prediction model to forecast the card-swiping data of the Hangzhou subway for 10 consecutive days. The historical data was divided into intervals of 3 minutes, and the random forest was used to compare the correlation among features. The final prediction results showed that both the inbound and outbound flows exhibited a bimodal pattern on weekdays, while on holidays, the prediction results showed a bimodal pattern. Furthermore, by adjusting the parameters appropriately, the simulated trends closely matched the actual values, indicating the applicability of LSTM in simulating short-term passenger flow. Accurate prediction of passenger flow can enable more comprehensive management of the subway system and provide passengers with a higher-quality service experience and travel options.

Short-term Passenger Flow Prediction for Hangzhou Metro based on Machine Learning Algorithms