Model Of Online Ride-Hailing Demand Based On Deep Network

Forecasting the demand for ride-hailing is an important part of enabling intelligent transportation systems in smart cities.The accurate demand forecasting model can reduce the waiting time of drivers and passengers by scheduling online car appointments in advance,helping cities to allocate resources in advance to meet travel needs,and can reduce empty taxis on the street.Energy waste and traffic congestion.First,this paper analyzes the Haikou city online car order data set provided by the Didi Gaia Opening Plan,as well as the current research status of some traffic flow predictions and online car demand predictions.The second step is to carry out the basic principles of neural networks Detailed introduction provides a theoretical basis for the establishment of later models.Next,the original data was analyzed and processed,the original order data was time sliced,and the order data was meshed according to the city map of Haikou City.The order data was integrated into a three-dimensional tensor for subsequent input models.Training.Then,a CNN-LSTM hybrid is proposed by using the convolutional neural network(CNN)for image data processing ability and the long shortterm memory neural network(LSTM)for time series data processing ability.Network,based on which a demand forecasting model for ride-hailing is constructed.Finally,the model is trained on the Haikou order data set to determine the final model parameters.It is obtained that the demand forecasting model proposed in this paper has good forecasting ability,and the root mean square error(RMSE)of the forecasting result is 9.0669.The Average Percentage Error(MAPE)is 0.1919,which is better than the results of other comparison models.This paper also compares the prediction results of different time periods and the prediction results of different regions.According to the comparison results,it can be obtained that the model's prediction accuracy rate for each time period is basically stable at about 81%,and the prediction accuracy rate of each region is basically Stable at around 82%.This shows that the CNN-LSTM hybrid network-based forecasting model for vehicle rental demand proposed in this paper can accurately capture the spatio-temporal characteristics of order data,and has good stability regardless of different time periods or different regions.