The Research Of Short-term Public Bicycle Prediction Model Based On Multi-source Data

As economy develops and urbanization accelerates,the problem of road traffic becomes increasingly serious.Among problems of road traffic,the traffic congestion and environmental pollution are two prominent aspects.In order to alleviate urban traffic congestion and reduce environmental pollution,as the increment of people's consciousness of environmental protection and the promotion of government,the public bicycle system(PBS)has developed fast and effectively to alleviate the urban road traffic problems.However the problems of “no bicycle to rent" and "no place to return" caused by the tide trip of public bicycle restrict the further development of the PBS.It is necessary to guide the users to the appropriate public bicycle rental and to remind the PBS operating companies scheduling public bicycles in time by predicting the public bicycle rent with intelligent algorithms.However,the traditional traffic flow prediction research focused mostly on the motor vehicle road traffic flow forecast,and prediction research on public bicycle is less,what's more,there are few prediction research according to the characteristics of public bicycle.Therefore,this paper proposes a public bicycle short-term rental prediction model based on multi-source data through investigation on the motor vehicle traffic flow prediction model and public bicycle using characters.This paper uses public bicycle history rental records,weather,and holiday data to solve the problem of public bicycle short-term rental prediction.The main contributions and innovations include:(1)Propose a new prediction method based on the characteristics of public bicycle and combined with public bicycle rental history records,weather,and holiday data to make predictions.(2)Use K-means algorithm and naive Bayesian classification model,predict the public bicycle rental model according to the weather,holidays data.Experiments show that combing these two kinds of algorithm to forecast public bicycle rental model has good prediction effect.(3)Use classified training data to construct time series data set,and the data set will be used for training APSO-BP neural network model respectively.The models will be used for predicting public bicycle short-term rental.Use APSO algorithm tooptimize the BP neural network can overcome the disadvantages of BP neural network,which has the slow convergence speed and may easily fall into local.Experiments show that APSO-BP neural network model used for predicting public bicycle short-term rental has higher prediction accuracy.(4)Use time series data and spatial data to build time and space combined public bicycle rental prediction training data set for some lease stations which have large number of visitors.The data set will be used for training APSO-BP neural network model.Experiments show that in this kind of stations,time and space combined public bicycle short-term rental prediction have higher prediction accuracy than the time series prediction method.