Predicting the electricity consumption of urban rail transit based on binary nonlinear fitting regression and support vector regression