GMM Clustering Based on WOA Optimization and Space-Time Coupled Urban Rail Traffic Flow Prediction by CEEMD-SE-BiGRU-AM

Abstract

Accurately predicting the short-term passenger flow of urban subways is very important for urban subway stations to formulate passenger flow organization and evacuation plans effectively and rationally plan passenger travel routes. This paper establishes a novel framework to predict the hourly inbound and outbound passenger flow of subway stations based on WOA-GMM station classifiers, CEEMD-SE noise reduction, and BiGRU optimized by attention. Firstly, this paper classifies subway stations using the improved Gaussian mixture model (GMM) with Whale Optimization Algorithm (WOA) to realize the feature extraction of different types of subway stations. Secondly, this paper uses the Complementary Ensemble Empirical Mode Decomposition (CEEMD) to decompose the noise reduction of each station’s hourly inbound and outbound passenger flow. It combines the empirical modal components by calculating the sample entropy (SE), which makes the time series stable and reduces the time cost of forecasting. Finally, a Bidirectional Gated Recurrent Unit (BiGRU) model improved by attention mechanism (AM) is established for each station’s inbound and outbound passenger flow. The prediction model established in this paper is verified by subway passenger flow data in Shanghai, China, within 24 days. Finally, it is concluded that the model can predict the passenger flow of subway stations. Compared with the traditional Backpropagation Neural Network (BP), the Long Short-Term Memory (LSTM), and the normal BiGRU model, the model proposed in this paper has an average reduction of 65.90%, 64.54%, and 49.06% in Mean Absolute Percentage Error (MAPE) in the prediction of the hourly inbound and outbound passenger flow of each type of station, respectively.