Metro Disruption Passenger Flow Prediction With GNN Simulation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing transportation models like MATSim are computationally expensive and unable to predict passenger flow patterns in near real-time during service disruptions, limiting metro operators' ability to manage disruptions effectively.
Innovation Solution
A method using a graph neural network (GNN) and an event-based metro system simulation model to predict passenger flow patterns by integrating metro network topology, service schedules, alternative transportation modes, and passenger diversion behaviors, enabling near real-time simulations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If MATSim is used to simulate passenger flow patterns during service disruptions, then the simulation captures detailed behavioral parameters and route choices, but the computational time required exceeds 20 days which is too slow for near real-time prediction
Solution Approach 1:
The patent segments the complex MATSim simulation into multiple components: a graph neural network model for rapid prediction of passenger flow patterns, an event-based metro system simulation model for disruption scenarios, and a calibration module using historical data. This segmentation allows each component to be optimized independently, with the GNN providing fast predictions without requiring full MATSim execution.
Solution Approach 2:
The patent creates a simplified copy of the MATSim simulation framework by implementing a custom event-based simulation model that replicates the essential functionality for disruption scenarios. This copy uses pre-calibrated parameters from historical MATSim runs, allowing rapid prediction without re-running the full computationally expensive simulation.
2Measurement precision
If multiple iterations of MATSim simulation are performed to obtain an equilibrium solution, then the prediction accuracy improves, but the computation time compounds further making it impractical for timely disruption management
Solution Approach 1:
The patent performs preliminary calibration using historical disruption data to pre-determine key parameters such as passenger diversion rates, alternative route preferences, and mode-switching behaviors. These pre-calibrated parameters are stored and reused during actual disruption predictions, eliminating the need for repeated iterative simulations to reach equilibrium.
Solution Approach 2:
The patent replaces the expensive, time-consuming equilibrium-based simulation approach with a lightweight GNN model that provides sufficiently accurate predictions for disruption management. The model accepts approximate inputs and produces acceptable predictions rapidly, sacrificing the precision of equilibrium solutions for the benefit of timely decision-making.
3Adaptability or versatility
If MATSim is used for service disruption prediction, then comprehensive network information including road network attributes and dynamic signal timing are considered, but the model is not calibrated for disruption scenarios and requires excessive computational resources
Solution Approach 1:
The patent extracts only the essential elements needed for disruption prediction from the comprehensive MATSim model. Instead of using the full multi-modal transportation network with all road network attributes and signal timing, the patent focuses specifically on metro network topology, station-level data, and historical disruption patterns, removing unnecessary complexity while maintaining predictive accuracy for disruption scenarios.
Data Source
AI summary
A computer-implemented method (100) for predicting passenger flow patterns during service disruptions in a metro network of a multi-modal transportation network. The method comprises: providing (105) first information associated with the metro network; providing (110) second information associated with alternative modes of transportation to the metro network in the multi-modal transportation network; providing (115) third information associated with the topology of a portion in the metro network associated with a service disruption; providing (120) fourth information associated with estimated diverted origin-destination-station (ODS) demand patterns for the service disruption; providing (125) fifth information associated with predicted irregular route choices of passengers in the multi-modal transportation network during service disruptions; and predicting (130) passenger flow patterns associated with the service disruption, through computationally simulating, based collectively on the first information, the second information, the third information, the fourth information and the fifth information.