Prediction-based early warning system for overflow of people in metro stations

Abstract

As cities rapidly urbanize, handling the management of crowded transportation hubs, notably metro stations, has become an immediate concern. High passenger traffic can lead to severe risks, including stampedes. While many past passenger flow forecasting systems aim to enhance prediction accuracy, the inherently noisy nature of passenger flow data makes it challenging for existing technologies to provide stable and precise predictions, making passenger flow management based solely on these models risky. This paper introduces a novel system that mitigates this risk by integrating a predictive model with managerial methods. The proposed management framework formulates a unique model for each station, determining a risk deviation coefficient grounded in the station's historical prediction accuracy. Station management is then holistically executed based on this coefficient and the predictive model. This paper employs the LSTM model for station-specific passenger flow prediction and defines the risk-related parameter , taking prediction accuracy into account. This adjusted LSTM prediction is then utilized to proactively streamline resource allocation, targeting improved passenger safety and overall station experience.