Research On Passenger Prediction And Recommendation Technology For Taxi Based On Data Analysis

With the introduction of “smart city” and the rapid development of data analysis technology,intelligent transportation has become an indispensable part of building a modern city,and taxis have an irreplaceable important role in urban transportation systems.However,current urban taxis face problems such as high no-load rate and low operational efficiency.At the same time,due to the strong randomness of taxi business and the lack of reasonable dispatch,the traffic system will cause serious urban problems such as road congestion and environmental pollution.Therefore,the research on taxi operation system has a large proportion in the field of intelligent transportation,and it is an indispensable step to realize “smart city”.The goal of studying the taxi operating system is to achieve the maximum match between taxis and passengers,so that taxis can reduce the no-load rate and improve operational efficiency,and passengers can reduce waiting time for taxis.The technology related to urban taxis has attracted the attention of researchers at home and abroad.Because taxi looking for passengers with uncertainty,This article takes taxi as the research object,and provides a reasonable search strategy for taxis through data analysis to improve the operating reward of taxis.For the prediction of the number of passengers in hot spots,this paper studies the passenger prediction method of taxis and predicts the number of passengers in hotspots based on historical taxi trajectory data.First,after pre-processing the original data,based on the changes in the passenger status in the data,taxi pick-up points are extracted.Then,combined with the distribution characteristics of passenger points,a K-Means and DBSCAN(KAD)method is proposed for clustering hotspots and extracting information on the number of passengers in the hotspots by time period.Finally,in consideration of the time series characteristics of hotspot passenger information,a Segment Prediction Based on RNN(SPBR)is proposed.Multiple segment prediction is used to reduce prediction errors.Based on the prediction results of SPBR,in order to improve the overall profit of taxis,this paper applies the Markov Decision Process(MDP)to the field of taxi passenger search,and proposes a Time-varying Markov decision process(TMDP).The extraction and calculation methods of TMDP parameters were studied,and then a strategy iterative algorithm was used to solve TMDP to find the optimal strategy and recommend it to taxis.For the prediction of taxi passengers,this paper proposes a framework of passenger prediction based on neural network(PPNN),which is divided into two parts: offline processing and online prediction,and can guarantee the timely update of data.PPNN considers the characteristics of the actual distribution of trajectory points in the clustering of passenger boarding point points,and proposes a clustering method combining K-Means and DBSCAN(KAD).For the neural network prediction model in the PPNN framework,considering the sequence of data,this paper proposes a segment prediction based on RNN(SPBR),which reduces the prediction error by multiple segmentation prediction.Further,in this paper,the passenger prediction results are used in the recommendation model of taxi seeking passengers,and the Time-varying markov decision process(TMDP)is proposed to find the optimal strategy.Finally,in terms of the number of passengers in hot spots,the prediction simulation and comparison of SPBR,SVR,CART,and BP neural networks were first performed.The results show that SPBR has higher accuracy in predicting the number of passengers in hot spots.Then from various aspects,the factors affecting SPBR prediction performance are analyzed.In terms of passenger search strategy recommendation,the TMDP simulation was used to obtain the optimal passenger search strategy for each time zone in each hotspot area.Compared with historical data,taxis can get higher expected returns after being recommended by TMDP,thereby achieving an increase in overall revenue.