Research On Online Ride-hailing Short-term Supply And Demand Forecast Based On KNN-LSTM

Network car Hailing is an important part of China's transportation system.It improves people's travel mode,enables people to realize intelligent travel,and effectively alleviates traffic problems.However,under the background of compliance,due to the asymmetric information between drivers and passengers,the unbalanced distribution of supply and demand in time and space,and the "difficulty in taking a taxi" of online car hailing,intelligent travel faces challenges,and short-term supply and demand forecasting is the key to meet the challenges.In this paper,we use the data of online car Hailing for visual analysis,select the appropriate algorithm to build the model,and predict the future time of online car hailing and the unmet order demand in a specific area,so as to provide data basis for online car Hailing platform scheduling.The main research contents are as follows:(1)In order to improve the passenger experience,this paper redefines the difference between supply and demand,including orders not received and orders with a time of more than 10 minutes.(2)The data source is analyzed visually based on time and space dimensions,the hidden rules in the data are studied deeply,and the data is preprocessed according to the requirements of the model for input indicators.(3)To explore the correlation between the relevant influencing factors of several time slices in front of the forecast point,the historical date of the forecast point and the supply-demand difference of the time slice,and the weather factors of the forecast point and the supply-demand difference of the forecast point,and to select the relevant indicators as the input data of the model.(4)According to the temporal and spatial characteristics of the prediction data,the knn-lstm short-term supply and demand multi-step prediction model with multi factor input is constructed,in which the KNN algorithm mines the spatial correlation of the data,the "memory gate" of LSTM is used to fit the timing of the data,and the encoder decoder framework is used to realize the multi-step prediction of the difference between supply and demand.(5)The best learning rate,the best dropout parameter and the best input area are selected by training the model for many times.The best model is predicted,and the multi-step prediction results are visually analyzed to verify the practicability of the model.The performance of the hybrid model is evaluated with single factor input LSTM model,multi factor input LSTM model and SVR model.The research of this paper shows that,considering the space-time characteristics of the network car Hailing data,a multi-step prediction model for the short-term supply and demand of knn-lstm network car Hailing data is constructed to predict the regions with different data rules,which has high accuracy and expansibility.The prediction of the difference between supply and demand in the future can provide the platform with different strategies of information transmission or intelligent scheduling,reduce the information asymmetry of the driver and passenger,improve the matching degree of the supply and demand of the online car hailing in time and space,promote the compliance process of the online car Hailing,build the online car Hailing traffic standard,guarantee the driver and passenger experience,and build the intelligent traffic city.