Prediction Method Of The Number Of Bike In Public Bicycle Rental Station

The public bicycle system can effectively solve the "last mile" problem of people’s travel and better enhance the overall level of urban public transport services. Currently "difficult in borrowing or returning a bicycle" in the process of implementing a public bicycle system is one of main problems, which directly affects the degree of satisfaction of users. However, bicycle schedule is one of methods to solve these problems, but the prediction of the number of bike at a rental station is one of key problems of bike scheduling. Therefore, it is of theory significance and application value of studying on the prediction of the number of bicycle based on the characteristics of public bicycle system.(1) Based on analysis of the pattern of the number of bicycle change from a rental station, this thesis presents a new method for predicting framework including the combination of data selection, the prediction model and error compensation. The framework is based on prediction model combining the error compensation mechanism which can greatly improve the prediction accuracy.(2) In order to tap the variation of public bike rental station and provide a good data support for the prediction model, a clustering method of rental stations based on the usage amount of bikes at a rental station. Through the different weather (sunny, cloudy, rain, snow, fog, wind, etc.) and season, and analysis of different types of rental stations combined with actual operational data from Lucheng District of Wenzhou City public bicycle system, all types of curves of bike change under different external conditions are described and analyzed by specific quantitative methods and represented into the feature string so as to cluster rental stations.(3) This thesis presents a prediction model of public bike at rental station based on time series forecasting model. On the basis of the existing prediction models of vehicle number, we use data selection method based on rental station clustering, combined with the historical trends of bike at a rental station, to forecast the number of bikes of the rental station. By comparing the predicting results with the predicted results of the existing model using the actual data, the prediction model in this paper is with high accuracy.(4) Due to the difference in the similar data selection generated by the difference of influencing factors between historical date and predictive date, this thesis presents an error compensation method of prediction result. By analyzing factors that could affect prediction errors, the specific impact of these factors is quantified to combine to historical data used by the predictive model. An error compensation method is adopted to calculate the error value as the compensation to the prediction model. Experimental results show that through error compensation the forecasting result can achieve higher accuracy.