WO2020186770A1

WO2020186770A1 - Visual method for analyzing taxi pick-up or drop-off features

Info

Publication number: WO2020186770A1
Application number: PCT/CN2019/115056
Authority: WO
Inventors: 吕伟韬; 杨树; 李璐; 潘阳阳
Original assignee: 江苏智通交通科技有限公司
Priority date: 2019-03-21
Filing date: 2019-11-01
Publication date: 2020-09-24
Also published as: CN109885639B; CN109885639A

Abstract

The invention provides a visual method for analyzing taxi pick-up or drop-off features. The method comprises: extracting, from taxi GPS path data, information about taxi pick-up or drop-off; performing taxi travel feature clustering analysis by means of a K-medoids algorithm; then, performing daily commuting travel destination feature analysis after performing extraction and clustering on the pattern shape of an average fluctuation trend curve of taxi pick-up or drop-off road segments; and finally displaying, on an electronic GIS map, taxi travel feature information of each road segment. The method eliminates the need for personnel to manually configure kernel size and search radius parameter values, and avoids personnel having to manually configure kernel size and a search radius visualization of regions of taxi pick-up or drop-off positions on a conventional heat map. Taxi travel destination feature information is acquired by means of the shape of an average fluctuation trend diagram of taxi pick-up or drop-off data, then taxi travel feature information is superimposed and displayed on an electronic GIS map, thereby achieving a novel visual display method for intuitively and clearly displaying taxi pick-up or drop-off features.

Description

Visualized analysis method of taxi getting on and off characteristics

Technical field

The invention relates to a visualized method for analyzing the characteristics of taxi getting on and off.

Background technique

The research on the hot spots and travel purpose of taxis is the key to taxi scheduling and route optimization. The traditional visual display of taxi travel mode is mainly heat map or cold hotspot drawing research. The drawing of the taxi boarding and boarding heat map mainly depends on the two parameters of the kernel size and the search radius. Each pick-up point is regarded as the center of the two-dimensional Gaussian distribution. The kernel size determines the discrete type of the two-dimensional Gaussian distribution, and the search radius determines the area The number of the inner pick-up and drop-off points can be obtained by summing these two-dimensional Gaussian distributions to obtain the hotspot or heat map, but the optimal solution of this data method comes from two parameters, and the values of the two parameters are determined by experience .

At the same time, taxi pick-up and drop-off points are generally along urban roads, while traditional heat map or cold and hot research and drawing uses each pick-and-drop point as the center of a two-dimensional Gaussian distribution. The Gaussian distribution has no direction constraint, which means that In the heat map, there are also getting on and off behavior in areas without road coverage. On the other hand, this visual display method of heat map and heat map can only display regional hot spots, and cannot display time information. It is necessary to manually filter the data collection statistical time period before visual display.

To sum up, in view of the traditional visual display method of taxi getting on and off (heat map/heat map), it is urgent to provide a visual display method of getting on and off the road that displays time and space information together, so that traffic control users can clearly grasp the rental on each road Car getting on and off mode information.

Summary of the invention

Aiming at the visual display mode of the heat map of the traditional taxi boarding and boarding hotspots, the purpose of the present invention is to provide a visualized taxi boarding and boarding feature analysis method, so that users can effectively grasp the taxi boarding and boarding travel feature information in the road network and avoid the traditional The heat map display method needs to manually determine the empirical mode of parameter information, and solves the problem of how to display the time and space information of the road and the car together in the existing technology, so that the traffic control user can clearly grasp the information of the taxi alight and the car on each road problem.

The technical solution of the present invention is:

A visualized taxi boarding and boarding feature analysis method, by extracting the boarding and boarding information from the taxi GPS trajectory data, clustering analysis of taxi travel features based on the K-medoids algorithm, and then analyzing the average fluctuation trend curve based on the road section. The pattern shape of the extracted and clustered daily commuting travel purpose feature analysis, and finally the taxi travel feature information of each road section is presented in the GIS electronic map, including the following steps:

S1. Extract the GPS trajectory data of taxis and integrate them to get the average daily getting on and off status of each road section;

S2, based on the Fourier transform, convert the time series of the vehicle getting on and off data from the time domain to the frequency domain value;

S3. Use k-medoids to perform cluster analysis on the frequency domain values transformed in step S2 to determine the type of taxi travel, including daily commuting travel mode, holiday travel mode, mixed travel mode and data missing mode;

S4. Extract the number of taxis getting on and off at each time period in the daily commuting travel mode, draw a fluctuation trend graph of the number of cars on and off at each time period of the day, and determine the travel purpose of the taxi trip during the time period based on the pattern shape of the fluctuation trend graph;

S5. Based on the travel characteristics determined in step S3 and the commuter travel purpose of each road segment determined in step S4, mark the characteristic information of the taxi boarding and boarding travel purpose of each road segment in the GIS electronic map.

Further, step S1 is specifically:

S11. Connect to the taxi on-board terminal, obtain the taxi GPS trajectory data, and extract the taxi getting on and off data;

S12. Integrate the daily GPS trajectory data of each road section to obtain the average daily getting on and off status of each road section.

Further, in step S11, the taxi getting on and off data is extracted, specifically,

First, perform data cleaning work on taxi trajectory information and then perform data analysis; the GPS trajectory data of taxi vehicles is a time sequence traj = {p ₁ , p ₂ ,..., p _i ,..., sorted by time. p _n}, _{_{where, p i = (lng i,}} lat i, t i), lng i represents the vehicle longitude at the i-th position, lat _i represents the latitude of the vehicle at the time of the i-th position, t _i represents the vehicle At the time of the i-th position; the taxi boarding and disembarking conditions are determined according to the passenger loading status in the taxi GPS trajectory data, including heavy and empty vehicles. That is, the taxi status changes from empty to heavy, indicating that there are passengers When getting on the bus, the status of the taxi carrying passengers from heavy to empty means that the passengers get off, and then the points of getting on and off are extracted from the GPS track data and arranged in time series.

Further, in step S12, the average daily getting on and off status of each road section at each time period is obtained, specifically,

S121. Determine the road section according to the urban road network information, the distance between adjacent intersections is the road section, and then rely on the longitude and latitude (lng _i , lat _i ) in the GPS trajectory data to divide the GPS trajectory data into each road section; Time sorts and sorts the time information t _i in the GPS trajectory data, and then integrates the number of getting on and off vehicles in each time period of each road segment;

S122. Solve the average value of taxis getting on and off at each time period of each day, and obtain the average number of getting on and off vehicles at each time of each road section each day, and arrange them in time series.

Further, the time series of the getting on and off data is converted from the time domain to the frequency domain value based on the Fourier transform, specifically, based on the characteristics of the spatio-temporal data transformation, the condition value of the getting on and off data is publicized through the Fourier transform N _t ={p ₁ , p ₂ ,..., p _j } is converted to frequency value ω(t)={ω ₁ , ω ₂ ,..., ω _j }, that is, the number of vehicles getting on and off in each period is converted to the frequency value of each period for step S1 , The specific Fourier formula is:

In the formula, ω represents frequency, t represents time, e- ^iωt is a complex variable function, f(t) is the amount of data for getting on and off cars, and F(ω) is the frequency value of getting on and off cars.

Further, in step S3, the taxi travel mode is determined, including daily commuting travel mode, holiday travel mode, mixed travel mode and data missing mode, specifically,

S31. Integrate the frequency data of getting on and off the vehicle in each time period of each road section of the sample number in the statistical time period respectively, and determine the number of clusters k of the sample clustering analysis;

S32. Randomly select k points from the sample points in the statistical time range as the initial center points, calculate the Euclidean distances between the remaining points and the k center points respectively to divide the clusters, and then solve the remaining clusters according to the divided clusters. The sum of the Manhattan distance between the point and other sample points, and the point with the smallest sum of Manhattan distance as the center point of the cluster; the cluster is solved repeatedly until the center point no longer changes, and the clustering result is obtained, that is, the frequency of taxi GPS getting on and off The division of data.

The beneficial effects of the present invention are:

1. This kind of visualized taxi boarding and boarding feature analysis method is compared with the traditional heat map display method. This method does not need to manually set the kernel size and search radius parameter values based on experience, and can avoid the traditional heat map from configuring the kernel through manual experience The size and search radius result in the visualization of the area of the pick-up and drop-off point.

2. This kind of visualized taxi boarding and boarding feature analysis method uses Fourier transform to convert the time series boarding and boarding data in the GPS trajectory data into frequency values. Spatio-temporal information will not be lost during the conversion, which is more convenient for data clustering analysis.

3. The method of the present invention uses the K-medoids center point method to perform cluster analysis on GPS getting on and off trajectory data, and divides the four situations of daily commuting travel mode, holiday travel mode, mixed travel mode and data missing travel mode, and based on getting on and off. The trend graph shape (English letters) of the average fluctuation trend of the data divides the characteristic information of the taxi travel purpose, so that the characteristic information of the taxi travel is displayed by superimposing the English letters on the GIS electronic map.

Fourth, this kind of visualized taxi boarding and boarding feature analysis method, by extracting and integrating GPS trajectory information, analyzing taxi boarding and boarding data, clustering analysis of taxi travel patterns with K-medoids algorithm, and showing the trend graph of the fluctuations of the data on boarding and alighting The shape of the English letters analyzes travel characteristics, thereby forming an intuitive and clear innovative visual display of taxi boarding and boarding characteristics.

5. This kind of visualized taxi boarding and boarding feature analysis method is compared with the spatial clustering analysis of traditional heat map, through k-medoids center point clustering analysis of the boarding and boarding data on each road section and the average fluctuation trend graph of the boarding and boarding data Draw, find the information of taxi boarding and boarding travel patterns and travel characteristics, so as to facilitate the traffic planning managers to perform taxi scheduling and route optimization.

Description of the drawings

FIG. 1 is a schematic flowchart of a visualized method for analyzing the characteristics of getting on and off a taxi according to an embodiment of the present invention.

Fig. 2 is a schematic diagram of four types of entertainment travel (U-type), work travel (M-type), dining travel (W-type), and mixed-type (VVV-type) in the situation of getting on a taxi in the embodiment.

Fig. 3 is a schematic diagram of the distribution of taxi boarding points in a certain city as a specific example in the embodiment.

Fig. 4 is a histogram from time domain to frequency domain for a certain road section in a certain city on a certain two days in an embodiment.

Figure 5 shows the four modes of daily commuting travel mode, holiday travel mode, mixed travel mode and data missing mode after clustering analysis on the amount of data on the bus in each day and each time period of a certain road section statistical period (quarter) in the embodiment. Illustration of the situation.

Fig. 6 is a schematic diagram of a curve drawn in the daily commuting travel mode in the embodiment.

Fig. 7 is a schematic diagram illustrating an embodiment of the recreational travel (U-type), work travel (M-type), dining travel (W-type), and hybrid (VVV-type) types superimposed on the map in the embodiment.

detailed description

The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

Example

A visualized taxi boarding and boarding feature analysis method, by extracting the boarding and boarding information from the taxi GPS trajectory data, clustering analysis of taxi travel features based on the K-medoids algorithm, and then analyzing the average fluctuation trend curve based on the road section. The pattern shape of, analyzes the characteristics of the trip purpose of the daily commuting class after the extraction and clustering, and finally presents the trip feature information of each road section in the GIS electronic map, as shown in Figure 1. The specific steps are as follows:

S1. Extract the taxi GPS trajectory data and integrate them to obtain the average daily getting on and off status of each road section.

S11. Connect to the taxi on-board terminal, obtain taxi GPS trajectory data, and extract taxi boarding/getting off data.

Preferably, when extracting the taxi trajectory information, first perform data cleaning work and then perform data analysis; the GPS trajectory data of taxi vehicles is a time sequence traj = {p ₁ , p ₂ ,... , P _i ,..., p _n }, where p _i = (lng _i , lat _i , t _i ), lng _i represents the longitude of the vehicle at the i-th position, and lat _i represents the vehicle’s longitude at the i-th position Latitude, t _i represents the time when the vehicle is at the i-th position; the taxi getting on and off is determined according to the passenger status (heavy and empty) in the taxi GPS trajectory data, that is, the taxi's passenger status changes from empty A heavy vehicle means that there are passengers getting on the bus, and the status of a taxi from heavy to empty means that the passengers get off the bus, and then the points of getting on and off from the GPS track data (arranged in time series) are extracted.

S12. Integrate the daily GPS trajectory data of each road section to get on and off the vehicle at each time of the day on each road section. The road section is the distance between adjacent intersections.

S121. Determine the road section according to the urban road network information (the distance between adjacent intersections is the road section), and then rely on the longitude and latitude (lng _i , lat _i ) in the GPS trajectory data to divide the GPS trajectory data into each road section; according to the date Sort and sort the time information t _i in the GPS trajectory data with time, and then integrate the number of getting on and off vehicles in each time period of each road segment;

S122. Solve the average value of taxis getting on and off at each time of the day, and get the average number of getting on and off each road each day, and arrange them in time series. For example, extracting the boarding data in section A, average the data of boarding at the same time on section A to get the average fluctuation trend of boarding in section A in one day.

S2. Convert the time series of the vehicle getting on and off data from the time domain to the frequency domain value based on the Fourier transform.

Specifically, based on the characteristics of spatio-temporal data transformation, through Fourier transform publicity, the data condition value N _t ={p ₁ , p ₂ ,..., p _j } is converted into a frequency value ω(t)={ω ₁ , ω ₂ ,..., ω _j }, that is, the number of vehicles getting on and off in each time period is converted to the frequency value of each time period for the S1 step. The specific Fourier formula is:

S3. Use k-medoids to perform cluster analysis on the frequency domain values transformed in step S2 to determine the type of taxi trips, including daily commuting trips, holiday trips, mixed trips and missing data.

S31. Respectively integrate the frequency data of getting on and off the vehicle in each time period of each road section of the sample number of the statistical time period, and determine the number of clusters k of the sample clustering analysis;

S32. Randomly select k points from the sample points in the statistical time range as the initial center points, calculate the Euclidean distance between the remaining points and the k center points respectively to divide the clusters, and then solve the remaining clusters according to the divided clusters The sum of the Manhattan distance between the point and other sample points, and the point with the smallest sum of Manhattan distance as the center point of the cluster; the cluster is solved repeatedly until the center point no longer changes, and the clustering result is obtained, that is, the frequency of taxi GPS getting on and off The division of data.

Specifically, in the frequency cluster analysis of GPS trajectory data, it can be divided into four categories, including daily commuting travel mode, holiday travel mode, mixed travel mode and data missing mode. Among them, the difference between weekdays and weekends is less and the default is Daily commuting travel patterns, while the changes between days of the week are small.

S4. Extract the number of taxis getting on and off at each time period in the daily commuting travel mode, draw a fluctuation trend graph of the number of cars getting on and off at each time period of the day, and determine the travel purpose of the taxi trip during the time period based on the pattern shape of the fluctuation trend graph.

Take the taxi boarding situation as an example, specifically including entertainment trips with a "U-shaped" trend graph, work trips with a "M-shape" trend graph, dining trips with a "W-shape" trend graph, and trend graph presentation "VVV type" shape hybrid type, as shown in Figure 2.

S5. Based on the GIS electronic map, it presents the travel characteristic information of taxis getting on and off each road section.

Specifically, based on the travel characteristics determined in step S3 and the commuter travel purpose of each road segment determined in step S4, the characteristic information of the taxi boarding and boarding travel purpose of each road segment is marked in the GIS electronic map.

This kind of visualized taxi boarding and boarding feature analysis method extracts the GPS trajectory information of taxi vehicles, sorts the data of each road section in time, and clusters the taxi trips through Fourier transform and k-medoids algorithm In view of the daily commuting mode, the purpose of taxi trips on the road section is displayed in a visual way in the shape of English letters, so as to display and process in the GIS electronic map, so that users can effectively grasp the characteristics of taxis getting on and off the road network and avoid the traditional heat map display method Need to manually determine the empirical model of parameter information.

The innovative visual display method of taxi boarding and boarding features of the embodiment is compared with the spatial clustering analysis of the traditional heat map, and the k-medoids center point clustering analysis of the boarding and boarding data on each road section and the average fluctuation trend graph of the boarding and boarding data are performed Draw, find the information of taxi boarding and boarding travel patterns and travel characteristics, so as to facilitate traffic planning managers in taxi scheduling and route optimization.

A specific example of the embodiment is as follows:

Step S1. Select a certain city as the research object, obtain the operating data of taxis in a certain quarter of the city, including GPS trajectory data, and extract the boarding data for research according to the status of the heavy/empty vehicle. The distribution is shown in Figure 3.

Step S2, according to the road section, collect statistics on the data of each time of the day, and convert it to frequency value through Fourier transform, draw a histogram from time domain to frequency domain, and obtain the frequency index after conversion. The histogram of a certain road section is shown in Figure 4.

Step S3: Perform cluster analysis on the amount of vehicle boarding data in each day and each time period of a certain road section statistical period (quarter), and divide four modes, as shown in FIG. 5.

Step S4: Further draw the fluctuation trend of the number of taxis on the day in each mode. Taking daily commuting as an example, the drawn curve is shown in FIG. 6.

Among them, in the daily commuting travel mode, the peak time period for each type of travel mode is determined as follows:

Step S5: Further superimpose each type of data on the map, and the specific information is shown in FIG. 7.

Claims

A visualized taxi boarding and boarding feature analysis method, which is characterized by: extracting the boarding and boarding information in the taxi GPS trajectory data, clustering analysis of taxi travel features based on the K-medoids algorithm, and then averaging The pattern shape of the fluctuation trend graph analyzes the characteristics of the daily commuting travel purpose after the extraction and clustering, and finally presents the taxi travel characteristic information of each road section in the GIS electronic map, including the following steps:

S1. Extract the GPS trajectory data of taxis and integrate them to obtain the average daily getting on and off status of each road section;

S2, based on the Fourier transform, convert the time series of the vehicle getting on and off data from the time domain to the frequency domain value;

S3. Use k-medoids to perform cluster analysis on the frequency domain values transformed in step S2 to determine the type of taxi travel, including daily commuting travel mode, holiday travel mode, mixed travel mode and data missing mode;

S4. Extract the number of taxis getting on and off at each time period in the daily commuting travel mode, draw a fluctuation trend graph of the number of cars on and off at each time period of the day, and determine the travel purpose of the taxi trip during the time period based on the pattern shape of the fluctuation trend graph;

S5. Based on the travel characteristics determined in step S3 and the commuter travel purpose of each road segment determined in step S4, mark the characteristic information of the taxi boarding and boarding travel purpose of each road segment in the GIS electronic map.
The visualized taxi getting on and off characteristic analysis method according to claim 1, wherein the step S1 is specifically:

S11. Connect to the taxi on-board terminal, obtain the taxi GPS trajectory data, and extract the taxi getting on and off data;

S12. Integrate the daily GPS trajectory data of each road section to obtain the average daily getting on and off status of each road section.
The visualized taxi getting on and off characteristic analysis method according to claim 2, characterized in that: in step S11, extracting taxi getting on and off data, specifically,

First, perform data cleaning work on taxi trajectory information and then perform data analysis; the GPS trajectory data of taxi vehicles is a time sequence traj = {p 1 , p 2 ,..., p i ,..., sorted by time. p n}, where, p i = (lng i, lat i, t i), lng i represents the vehicle longitude at the i-th position, lat i represents the latitude of the vehicle at the time of the i-th position, t i represents the vehicle At the time of the i-th position; the taxi boarding and disembarking conditions are determined according to the passenger loading status in the taxi GPS trajectory data, including heavy and empty vehicles. That is, the taxi status changes from empty to heavy, indicating that there are passengers When getting on the bus, the status of the taxi carrying passengers from heavy to empty means that the passengers get off, and then the points of getting on and off are extracted from the GPS track data and arranged in time series.
The visualized taxi getting on and off characteristic analysis method according to claim 2, characterized in that: in step S12, the getting on and off conditions of each road section on average every day are obtained, specifically:

S121. Determine the road section according to the urban road network information, the distance between adjacent intersections is the road section, and then rely on the longitude and latitude (lng i , lat i ) in the GPS trajectory data to divide the GPS trajectory data into each road section; Time sorts and sorts the time information t i in the GPS trajectory data, and then integrates the number of getting on and off vehicles in each time period of each road segment;

S122. Solve the average value of taxis getting on and off at each time period of each day, and obtain the average number of getting on and off vehicles at each time of each road section each day, and arrange them in time series.
The visualized taxi boarding and boarding feature analysis method according to any one of claims 1 to 4, characterized in that: in step S2, the time series of the boarding and boarding data is converted from the time domain to the frequency domain value based on Fourier transform Specifically, based on the characteristics of spatio-temporal data transformation, through Fourier transform publicity, the vehicle getting on and off data condition value N t ={p 1 , p 2 ,..., p j } is converted into a frequency value ω(t)={ω 1 , ω 2 ,..., ω j }, that is, the number of vehicles getting on and off in each period is converted to the frequency value of each period for step S1. The specific Fourier formula is:

In the formula, ω represents frequency, t represents time, e- iωt is a complex variable function, f(t) is the amount of data for getting on and off cars, and F(ω) is the frequency value of getting on and off cars.
The visualized taxi getting on and off characteristic analysis method according to any one of claims 1-4, characterized in that: in step S3, the taxi travel mode is determined, including daily commuting travel mode, holiday travel mode, mixed travel mode and Data missing pattern, specifically,

S31. Integrate the frequency data of getting on and off the vehicle in each time period of each road section of the sample number in the statistical time period respectively, and determine the number of clusters k of the sample clustering analysis;

S32. Randomly select k points from the sample points in the statistical time range as the initial center points, calculate the Euclidean distances between the remaining points and the k center points respectively to divide the clusters, and then solve the remaining clusters according to the divided clusters. The sum of the Manhattan distance between the point and other sample points, and the point with the smallest sum of Manhattan distance as the center point of the cluster; the cluster is solved repeatedly until the center point no longer changes, and the clustering result is obtained, that is, the frequency of taxi GPS getting on and off The division of data.