A Deep Learning Method For Taxi Order Prediction

With the rapid development of big data and artificial intelligence system,the problem of traffic congestion makes people have higher requirements for daily travel,and the prediction of taxi orders in the future is one of the effective means to solve people’s needs.In recent years,in order to improve the taxi market management,taxi operation efficiency,passenger ride convenience to maximize,industry scholars have put forward many solutions.Forecast taxi future orders to produce,which is based on past historical data,so as to forecast the future taxi orders,the forecast can not only promoted the taxi market management efficiency and benefit,more convenience for passengers traveling quality greatly,the urban traffic management,and people’s travel arrangements to provide the reasonable suggestion.However,there are still many deficiencies in the current taxi order forecasting method.Most of the taxi order prediction methods only focus on the time series feature learning to analyze the future;although other prediction methods for the improvement,or on the characteristics of time and space information extraction in order to forecast future traffic flow,the however both have their own deficiencies: the former is missing in the information acquisition,a large amount of information,the latter for information extraction method is not perfect enough,the accuracy of both methods to predict the bad consequences.To solve the above problems,this paper proposes two taxi order forecasting methods based on deep learning.The first method is to rasterize the real city map image,and pixelated images have been obtained,using the global positioning system(GPS)to obtain a large number of the taxi vehicle trajectory data of time and space,by driving a taxi trajectory and the distribution characteristics of passenger boarding,we design a VGG-FCN model,under different time and space in the future in different parts of the taxi driving route is forecasted,in optimizing the parameters of the model at the same time,guarantee the accuracy of the model has not reduced.The other method is GCN-LSTM method.After transforming the urban road map into a line graph,the time and space of taxi order data are extracted,and the future taxi order situation is predicted.The method proposed in this paper can simultaneously predict the future order quantity of multiple regions.After the comparative experiment with the representative method,it has been improved in multiple evaluation objectives.