Online Metro Origin-Destination Prediction via Heterogeneous Information Aggregation

Area of Science:

• Intelligent Transportation Systems
• Data Science
• Machine Learning

Background:

• Metro origin-destination (OD) and destination-origin (DO) prediction is vital for intelligent transportation systems.
• Online metro systems often lack complete historical OD matrices, hindering accurate forecasting.
• Conventional methods forecast OD and DO ridership separately using limited information.

Purpose of the Study:

• To propose a novel neural network module, the Heterogeneous Information Aggregation Machine (HIAM), for joint OD and DO ridership prediction.
• To address the challenge of incomplete historical data in online metro systems.
• To improve the accuracy of forecasting both OD and DO ridership simultaneously.

Main Methods:

• Developed the Heterogeneous Information Aggregation Machine (HIAM) to leverage diverse historical data, including incomplete OD matrices, unfinished order vectors, and DO matrices.
• Implemented an OD modeling branch to estimate destinations from unfinished orders, complementing incomplete OD matrices.
• Integrated a DO modeling branch to analyze spatial-temporal DO ridership patterns and a Dual Information Transformer to model OD-DO correlations.

Main Results:

• The proposed HIAM effectively utilizes heterogeneous information for joint OD and DO ridership pattern learning.
• The unified Seq2Seq network based on HIAM demonstrates superior performance in simultaneous OD and DO ridership forecasting.
• Extensive experiments on large-scale benchmarks validate the method's effectiveness for online metro OD prediction.

Conclusions:

• HIAM offers a robust solution for online metro origin-destination prediction by jointly modeling OD and DO ridership.
• The method's ability to handle incomplete data and capture interdependencies between OD and DO flows enhances prediction accuracy.
• The developed approach represents a significant advancement in intelligent transportation systems for real-time ridership forecasting.