Research On Short-Time Forecast Of Taxi Travel Demand Based On Combination Model

As an important supplement to the public transport system,taxis plays an important role in improving travel services and the road capacity for vehicles.In actual life,due to the complex and diverse crowd activities and the uneven population distribution in various regions,the travel demand of the crowd are featured by relatively great randomness and volatility in temporal and space.The regional contradiction between the supply and demand of passenger service is aroused for taxis are unable to perceive the demand for travel in time.How to reasonably allocate the capacity for vehicles in the city to meet the demand for travel service has become a hot research topic in recent years.Aiming at the characteristics of crowd travel,this paper proposes a model of forecasting short-term travel demand on the basis of combined neural network,which combines the spatial feature forecasting model and the time-series feature forecasting model,extracts the temporal and spatial variation rules respectively,and expresses the temporal and spatial characteristics of travel demand,so as to improve the accuracy of regional travel demand forecasting.The main contents of this paper are as follows:(1)Preprocess of the multi source data.Firstly,the travel data of historical travel are sorted out,and the time and boarding location of travel in orders are extracted;in terms of the time granularity of short-term forecast,the time is divided into slices according to the granularity of short-term forecasting time to count the number of travel orders in each temporal slice in each area.For the missing information or abnormal data of continuous variables in the weather data,the mean value method is used for smoothing processing to fill them.Encoding conversion on string data such as weather type is conducted.Based on the number of each type of POI in the regional POI data,the regions are classified according to the regional functional attributes.(2)Analysis of the spatial and temporal characteristics of travel demand.The multi-source data are combined to compare and analyze the distribution characteristics of travel demand under different date attributes and different spatial attributes,after which the characteristics of travel demand changes in various temporal periods and the distribution of travel demand in different spatial attributes are studied.Furthermore,the correlation analysis is carried out on the potential factors that affect the number of travel demand,and the appropriate characteristic variables are screened out as the input data of the subsequent forecasting model.(3)Construction of a short-term combination forecast model of travel demand.The data of weather feature and historical travel in the target area are employed to establish a model of forecasting time series feature based on the Bi-LSTM network;the historical travel data of the target area and its neighboring areas and the POI data of each area are used to establish a spatial feature prediction model based on the Conv-LSTM network;the two sub-models are merged into a combined forecasting model through the fusion module so that the model has the ability to acquire spatial and temporal characteristics at the same time,and the accuracy of travel demand forecasting can be enhanced.(4)Simulation experiment of the combined model and result evaluation.The performance of the combined model is evaluated after using the preprocessed travel data.The multiple hyperparameters of the combined model are tuned through the method of controlling variables,and then the fitting effects of the combined model are tested under different date attributes,and the results are analyzed and evaluated.Finally,the combined model is compared with the traditional model.The experimental results show that compared with the traditional model,the root mean square error and the mean absolute percentage error of the combined forecasting model are lower than the traditional model,which verifies that the combined model can improve the accuracy of the result of forecasting temporal and spatial characteristics.