Short-term forecasts on individual accessibility in bus system based on neural network model

Abstract

Precise forecasts on individual accessibility in bus system can help make policies to accommodate fluctuating bus travel demand and promoting social equity. In this study, we propose a three-stage method for short-term forecasts on individual accessibility in bus system based on neural network (NN) model. In the first stage, a NN model is designed to tackle the nonlinear mapping between passengers' bus trip appearances in historical periods and those in the predicted period. A rate function, which considers bus trip generation rates of passengers, is then applied using outputs of the designed NN model. In the second stage, probabilities of origin-destinations (ODs) chosen by passengers in the predicted period are calculated. In the third stage, land use information combined with results of previous two stages are used to obtain the individual accessibility in bus system in the predicted period. Compared to individual accessibility calculated by real data, it is found that the average errors of predicted results by the proposed method in weekdays and at weekends are only 8.37% and 10.13%, respectively. The results also demonstrate the capability of combining a NN model, traffic data and land use information to forecast the future spatial distribution of individual accessibility in transport system.