Research On Intelligent Analysis Of Regional Trajectory Hotspots

Regional hotspots are directly related to the future planning of the city and the road conditions of the city,and also reflect the importance of smart city construction and development.The trajectory data in the region is one of the important information for studying regional hotspots.Regional hot spots can not only provide a variety of effective reference opinions for road traffic management,but also provide a certain theoretical basis for the future development planning of the city,but also show the daily travel characteristics of residents from the side.In view of the challenges and difficulties faced by the current research on regional trajectory hotspots,this paper makes corresponding research on the analysis of regional trajectory hotspots based on the obtained taxi data.Hot spot analysis in the region refers to the mining of the travel characteristics of residents in the region,the location with large flow and high aggregation degree in the region.Through the study of residents’daily travel behavior,the potential hot spot locations in the area or locations with high aggregation density are analyzed.The main work of this paper is as follows:1)Data preprocessing.This paper uses the taxi trajectory data of the second ring road in Chengdu and the second ring road in Xi’an.In order to make the results not affected by data noise points,it is necessary to clean the obtained data first,which is called data preprocessing.For the position points that absolutely deviate from the track area,the processing of direct deletion is adopted.Since there are multiple attributes in the dataset,but not all attributes are required in the experiment,delete the unnecessary attributes.Map matching the data set after coarse-grained processing.This paper proposes a Kal-HMM（Kalman-Hidden Markov Model）map matching algorithm,which improves the problems of lack of accuracy and low time efficiency in the traditional map matching algorithm.Firstly,the deviation point of the track is removed,the deviation point is determined according to the angle difference,and the data of the removed deviation point is predicted by Kalman filter to supplement the removed points.Finally,the processed track data is matched with the actual map by HMM map matching algorithm.2)Travel law analysis.Use the online car Hailing track data mentioned in 1)to extract the OD（origin starting point,destination）data generated by residents’daily travel,and divide the data into National Day holiday part,working day part and weekend part through time attribute.In order to reduce the influence of non subjective factors such as environment and weather on the results,the average value of the data on working days and weekends is calculated respectively,and the average value is added to the evaluation process.Analyze the number and duration of trips on National Day holidays,working days and weekends at hourly intervals,and then compare the data processing results of weekends and working days to study the travel characteristics of residents in different time domains.3)Regional flow forecast.This paper proposes a tcn-lstm（temporary revolutionary Network-long short term memory）traffic prediction algorithm to predict the traffic of the second ring road in Chengdu.The features in the data set are extracted by TCN algorithm.Specifically,according to the longitude and latitude,time,minimum duration,maximum duration,average duration and other information in the data set,multiple one-dimensional convolution cores are constructed,and then the convolution cores are used to extract the feature information in each dimension.The expanded convolution is added in front of the loop layer of LSTM to obtain the amount of information contained in the current time,which can significantly simplify the algorithm structure,optimize the processing speed of the algorithm and improve the accuracy of the prediction results.Finally,the test set set set aside from the data set is put into the tcn-lstm algorithm to predict the traffic in the time period.4)Regional hotspot analysis.For the hot spot analysis in the region,a MC-Kmeans（Mahalanobis distance Canopy-K-means）algorithm is proposed to analyze the hot spots in the region.Firstly,using the second ring data set of Chengdu mentioned in 1),the data set is clustered by canopy algorithm to obtain the clustering result6)^′,that is,it is clustered into6)^′.Then take6)^′as the K value of K-means algorithm for the second clustering.In the process of K-means algorithm,Markov distance is used as the formula to select the clustering center.Finally,the clustering results of MC-Kmeans algorithm are obtained and visualized by means of thermodynamic diagram.