Short-Term Passenger Flow Prediction Based on a Hybrid Model of Support Vector Machine and Long Short-Term Memory Network

This study focuses on Line 10 of the Nanjing subway system, analyzing the spatiotemporal distribution characteristics of passenger flow data from the perspectives of network layout and station functionality. The research summarizes the unevenness, temporal variability, periodicity, and similarity characteristics of passenger flows along this line. Addressing these characteristics, a preprocessing method combining Self-Organizing Map (SOM) algorithm with Isolation Forest is proposed for handling passenger flow data, which involves data classification and anomaly detection. Input features such as periodic flows, weekday identification, weather conditions, and passenger flow categories are constructed. A Short-Term Passenger Flow Prediction Model (PSO-SVM) is developed, employing Support Vector Machine (SVM) algorithm optimized by Particle Swarm Optimization (PSO) for enhanced parameter tuning capability. The integrated predictive model is applied within the operational decision-making framework of Line 10, facilitating data storage, retrieval, and analysis of future station passenger flows. This enables rail transit operators to adjust operational strategies more accurately based on predicted passenger flows, thereby improving overall energy efficiency.