The Research And Implementation Of Taxi Pick-up Points Recommendation Method With Spatio-temporal Feature-constraints

With the development and popularization of artificial intelligence and big data technology,the revolution of smart city follows.The traditional transportation system is gradually transformed into "smart travel" which integrates technical support,urban construction and transportation reform.Taxi has become one of the important components of public transportation because of its flexibility,convenience and personalization.With the increasing maturity of global positioning system and data analysis technology,spatio-temporal data mining with taxi track data as the research object has gradually become an important research direction in the computer field.Due to various factors such as the imbalance in the supply and demand relationship between taxis and passengers,the lack of information exchange between drivers and passengers and other factors,taxi passengers generally have problems such as long waiting times and difficulty in taxis.There is an urgent need for a passenger recommend a scientific and reasonable waiting point method,that is,to increase the success rate of passengers and reduce the waiting time of passengers.Scholars at home and abroad have put forward a variety of methods in order to solve this problem,such as queuing model,prior rules,non-homogeneous Poisson process,clustering and probability distribution,which ignore the influence of urban residents’ time and space characteristics,so that the accuracy of the recommendation model of waiting point for taxi passengers is limited.Therefore,this paper proposes a spatial-temporal feature-constraints recommendation method for taxi passengers’ pick-up points,which is expected to provide a scientific decision-making model for urban taxi passengers’ smart travel.The main research contents are as follows:(1)Realize multi-source data fusion based on spatial grid.The data used in this paper include taxi trajectory data,Point of Internets(POI)data,weather data and administrative map information data.Because the format and content of different data sets are different,this paper matches and compares the spatial and temporal information in different data sets,and adds data information such as weather and functional area category on the basis of the historical track data of taxi to achieve multi-source data fusion.Finally,the fused data is mapped to a divided50m*50m unit grid to realize continuous trajectory data discretization.(2)A deep neural network recommendation model with spatial-temporal featureconstraints is proposed.This paper analyzes the factors that affect taxi passengers’ choice of pick-up points from three aspects of time,space and weather,and extracts the input eigenvalues of the recommended model of multiple pick-up points.Under the spatio-temporal features,the optimal deep neural network is used to model the spatio-temporal law of taxi transportation,to predict the waiting time and recommend the pick-up point of taxi passengers.(3)Experimental verification based on large-scale real taxi trajectory data.In this paper,Wuchang district of Wuhan city as the object of object area,and the collected historical taxi trajectory data,POI data,historical weather data,city administrative map and other data are used for model experiments.The design of three control group respectively,analyzing the characteristic of the temporal characteristics,spatial characteristics and weather characteristics affect the result of the experiment,experiment under the characteristics of spatio-temporal feature-constraints of the model has good applicability and accuracy.(4)Designed and implemented a prototype of a mobile application that taxi passengers recommend pick-up points.The system takes the recommended model as the core,uses the map to show the waiting time of empty taxi when the current location is the waiting point,and recommends the nearby reference pick-up point with high success rate and short waiting time for the user,with obvious recommendation effect.