Forecasting Method for Urban Rail Transit Ridership at the Station-Level Using a Weighted Population Variable and Genetic Algorithm Back Propagation Neural Network

Distance-decay affect of contribution rate of population in different distance to ridership has not been considered in current model. As a result, prediction of ridership at station-level is not accurate. Population in near distance to the station contributes more to ridership than that in far distance. So population used to predict should be weighted by corresponding contribution rate. Multivariate correlation analysis is used to analyze relationship between weighted population and ridership, and also to analyze relationship between other predictors and ridership, which can pick out significant predictors affecting ridership. To solve the irrationality of linear prediction model, model of Back Propagation Neural Networks (BP) which can express strong relation between independent and dependent elements and needs no formula in detail has been built. To avoid local solution, Genetic Algorithm (GA) is used to improve the model. Ridership result predicted by GA-BP with weighted population is compared with that predicted by linear model with weighted population, that predicted by GA-BP with total population and that predicted by linear model with total population. The comparation shows model in this essay exceeds others, taking minimum and maximum relative error, average relative error, and root of mean square error into consideration. So, GA-BP model with weighted population is perfect when forecasting ridership of urban railway transit at station-level.