Passengers’ Travel Pattern Exploration And Hot-spots Recommendation Based On Taxi Origin-destination Data

With the continuous development of big data analysis technology,traffic big data mining and analysis have received extensive attention.Passengers travel pattern analysis and city hot spot recommendation are important research contents in urban management planning,traffic scheduling and other fields,and have a wide application prospect in residents’ daily travel life.Taxi travel is an important urban travel medium,which provides an important way of insight into a city’s traffic and population movement.With the rapid development and maturity of online taxi in recent years,the massive data of passenger travel has provided researchers with very important research data in order to ensure the safety of passengers.Based on these data,on the one hand,they can help researchers excavate the travel patterns of residents within the city,such as the number of daily travel or visiting passengers in a certain region within a week and the preferred travel time period of passengers.On the other hand,it can help to find hot areas of taxi demand in cities,which can help drivers find orders quickly and improve their operating efficiency.Therefore,the analysis of passenger travel patterns based on taxi OD data and the recommendation of travel hot spots are of great significance for improving the operating efficiency of taxi companies and reducing the waiting time of passengers.Traditional methods of residents’ travel pattern mining mostly focus on the study of the spatiotemporal rules of residents’ travel under normal conditions,but ignore the mining of abnormal travel data and passengers’ travel rules,so it is difficult to trace the causes of abnormal data.In addition,the search for taxi hot spots requires high accuracy and efficiency,so as to ensure that taxi drivers can find the nearest travel hot spots as soon as possible at the current moment.In view of the above problems,this paper focuses on the analysis of passenger travel patterns and the recommendation of taxi hot-spots.The specific work contents are as follows:（1）Passenger travel pattern mining and analysis based on sparse low-rank decomposition.The traditional analysis method of abnormal travel patterns only arranges passenger positions according to longitude and latitude,and lacks analysis of OD data related to functional areas,so it is difficult to accurately depict passenger travel purpose.To solve this problem,this paper proposes a method for analyzing passengers’ travel pattern analysis based on sparse low-rank tensor decomposition.Firstly,the multi-dimensional OD data is represented by a tensor.Each element of the tensor contains information of three dimensions,namely time,space and functional area attributes.Secondly,the sparse low-rank decomposition of the tensor data is carried out to obtain the normal pattern tensor representing the regular data and the abnormal pattern tensor representing the irregular data,as well as the corresponding pattern matrix.Finally,the pattern matrix is analyzed to obtain the rules of travel data in time,space,functional areas and other dimensions.The experiment on the Chengdu Didi-Taxi data set published by Gaia data shows that this method can more accurately excavate the travel purpose and potential motivation of passengers by adding functional area attributes to passenger travel data.（2）Urban hot-spots recommendation based on a density peak clustering algorithm.Hot-spots found that traditional methods such as poor efficiency in calculation and not practical problems,this thesis proposes an urban hot spots recommendation method based on a peak density clustering algorithm,which overcomes traditional peak density clustering algorithms by improving their computational efficiency.As a results,the proposed method can effectively fit the demand of taxi drivers.The method first calculates the pair-to-pair distances of all the data to calculate the out-of-local density and the minimum high-local density distance of each point,and then uses these two data to construct a decision tree to analyze the potential hot spots.The experiment on the data set of Didi-Taxi in Chengdu shows that the algorithm can accurately and efficiently find the hot spots of passengers’ taxi-hailing.