Relationship analysis of short-term origin–destination prediction performance and spatiotemporal characteristics in urban rail transit

Accurately predicting passengers’ origin and destination (OD) demand and analyzing their spatiotemporal characteristics are the key to efficient operation and management of urban rail transit. While obtaining these spatiotemporal characteristics and thus making a short-term OD prediction are a big challenge for a model due to its high dimensionality and uncertainty. This paper proposes a pattern match algorithm based on t-distribution stochastic neighbor embedding and K nearest neighbors (TSNE-KNN) to promote the prediction performance and introduces similarity indicators to explore these features of OD flow and their relationship with the forecasting performance. Analysis of automatic fare collection data of Beijing rail transit shows that the TSNE-KNN model is superior to other state-of-the-art approaches, even including deep neural network models, and the similarity, which is affected by the functional attributes of station, the surrounding land use attributes and the degree of development of the road network, can be a universal indicator to indirectly reflect the time–space properties of the OD flow. It is found that as the similarity of daily OD flows decreases, the performance of short-term OD prediction of decreases, and rail transit stations are gradually shifting from the periphery to the center of the city.