Short-term Prediction Of Car-hailing Travel Demand Based On Deep Neural Network

With the rapid development of mobile Internet,especially the widespread popularity of smart phones,the online car-hailing industry has developed rapidly in recent years,and more and more people choose personalized online car-hailing services.Although the time cost is reduced compared to traditional taxis,due to the asymmetry of information and the randomness of travel demand,it is difficult to ensure the consistency of passenger travel demand and vehicle supply in time and space.Areas with large local demand are in problems such as “difficult to take a taxi” and long waiting time in bad weather or peak travel periods.This paper uses the online car-hailing order and other relevant data to study the spatial and temporal distribution characteristics of online car-hailing travel demand,and establish and verify models for local demand and regional demand short-term prediction respectively,which improves the accuracy and efficiency of prediction.It is of great significance to improve the efficiency of passenger travel,the dispatch level of online car-hailing companies,reduce environmental pollution,and reduce the pressure on urban road traffic.The specific work of this article is as follows:(1)In this paper,we first defined the definition of net travel demand,short-term prediction of local demand and short-term prediction of regional demand;According to the Geohash algorithm,the region within the fifth ring in Beijing was divided and the predicted time granularity was determined;based on the qualitative analysis of factors affecting the demand for ride-hailing trips,data collection,data cleaning,map matching and other pre-processing operations were carried out for multiple influencing factors such as ride-hailing order data,POI data and weather data.(2)This paper will study the temporal and spatial characteristics of ride-hailing demand distribution,that is,from the overall change trend,different week attributes,different date types,respectively study the time distribution characteristics of travel demand.;The spatio-temporal distribution characteristics of travel demand were studied by comparing the spatial distribution of travel demand under different date types.Correlation analysis was conducted between the potential influencing factors such as historical demand order data,weather data and POI data and the demand.The input characteristic variables of the model were selected and summarized.(3)For the problem of short-term forecast of local demand,a single-factor local demand short-term forecast model based on LSTM was established using only the time series of historical car-hailing demand,and multi-factor local demand based on LSTM was established using processed multi-source data Short-term forecast model;for the short-term forecast of regional demand,LSTM,CNN and Conv-LSTM networks were used to capture the different spatial and temporal changes of demand respectively,and the output of the three sub-models was fused to establish a regional demand short-term forecast model.(4)The actual data of Beijing didi platform are used for case analysis.In view of the multiple short time travel demand forecast models,the network is optimized for multiple super parameters;The results of the local demand short-term forecast model and the regional demand short-term forecast model was analyzed and evaluated from the perspectives of the overall forecast results,the forecast results of different date types,and the forecast results of different time granularity;comparing the above models with traditional models,the results show that the regional travel demand short-term forecast model has significantly improved accuracy and efficiency,and the addition of multisource data can significantly improve the forecast Precision.