A Data-Driven Method for Dynamic OD Passenger Flow Matrix Estimation in Urban Metro Systems.

Dynamic O-D flow estimation is the basis of metro network operation, such as transit resource allocation, emergency coordination, strategy formulation in urban rail system. It aims to estimate the destination distribution of current inflow of each origin station. However, it is a challenging task due to its limitation of available data and multiple affecting factors. In this paper, we propose a practical method to estimate dynamic OD passenger flows based on long-term AFC data and weather data. We first extract the travel patterns of each individual passenger based on AFC data. Then the passengers of current inflows based on these patterns are classified into fixed passengers and stochastic passengers by judging whether the destination can be inferred. Finally, we design a K Nearest Neighbors (KNN) and Gaussian Process Regression (GPR) combined hybrid approach to dynamically predict stochastic passengers’ destination distribution based on the observation that the distribution has obvious periodicity and randomicity. We validate our method based on extensive experiments, using AFC data and weather data in Shenzhen, China over two years. The evaluation results show that our approach with 85% accuracy surpasses the results of baseline methods and the estimation precision reaches 85%.