A multistep forecasting method for online car-hailing demand based on wavelet decomposition and deep Gaussian process regression

The main objective of this paper is to develop a high-precision forecasting method that can forecast the probability distribution of the demand value of ride hailing. Due to the influence of various complex factors, the demand time series of ride-hailing generates noise and thus affects the accuracy of the forecast. Forecastings of uncertainty can provide a valuable reference for vehicle scheduling decisions. In the present study, the application of wavelet-DGPR models to forecast the demand time series of ride hailing was investigated. The effectiveness of the model was verified by using the demand data of ride-hailing in Hangzhou. The results show that wavelet decomposition can reduce the difficulty of forecasting; DGPR can obtain a probability distribution of demand forecast values for uncertainty forecasting. Wavelet-DGPR has better forecasting accuracy, stability, and robustness than typical methods.