Metro Passenger-Flow Representation via Dynamic Mode Decomposition and Its Application

Area of Science:

• Data Science
• Transportation Systems
• Applied Mathematics

Background:

• Metro passenger flow analysis is crucial for intelligent transportation systems.
• Spatiotemporal dependencies and dynamic changes in passenger flow data pose representation challenges.
• Accurate passenger flow representation is fundamental for anomaly detection and prediction.

Purpose of the Study:

• To propose a novel passenger flow representation model for anomaly detection and short-term prediction.
• To leverage time-varying data characteristics for improved metro system operations.
• To enhance the accuracy of anomaly detection and prediction in metro passenger flow.

Main Methods:

• Developed a novel passenger flow representation model integrating low-rank dynamic mode decomposition (DMD).
• Incorporated global low-rank and sparsity properties to capture spatiotemporal consistency and abrupt changes.
• Applied strong temporal Toeplitz regularization for enhanced anomaly detection of periodic changes.

Main Results:

• The proposed model efficiently detects anomalies, particularly time sequence anomalies.
• Experimental results on Beijing metro data validate the model's effectiveness.
• The model demonstrates superiority over other methods in most short-term passenger flow prediction scenarios.

Conclusions:

• The low-rank DMD-based model offers a flexible and convenient approach for passenger flow anomaly detection and prediction.
• The integration of sparsity and regularization techniques improves the characterization of complex passenger flow dynamics.
• This research contributes to the advancement of intelligent metro operations through enhanced data representation and analysis.