Urban traveler portrait system based on mobile phone signaling data

Abstract

The invention discloses an urban traveler portrait system based on mobile phone signaling data. The urban traveler portrait system comprises a data preprocessing module, a travel chain extraction module, a travel mode identification module, a daily travel mode identification module, a travel path flow identification module and a traveler portrait module. The system carries out preprocessing by loading stationary point data processed by a mobile phone signaling data supplier and combining administrative division data, extracts a trip chain, identifies a trip mode by using an Gade map API and a logarithmic Gaussian mixture model, identifies a daily trip mode by combining a convolutional auto-encoder and a K-means algorithm, carries out statistics on trip path flow, and carries out real-time processing on the trip path flow. And finally, summarizing the tags of the travel characteristics to construct a traveler portrait. According to the system, the data processing flow is optimized, the accuracy of travel mode identification is improved, the interpretability of travel mode identification is enhanced, the macroscopic traffic flow and the individual travel path are associated, and an accurate and efficient analysis tool is provided for traffic demand management.

Description

Technical Field

[0001] The present invention relates to the technical field of urban traffic management, and in particular to an urban traveler portrait system based on mobile phone signaling data. Background Art

[0002] With the acceleration of urbanization, urban traffic problems have become increasingly prominent, among which the behavior patterns of urban travelers have an important impact on traffic planning and management. In order to better understand and optimize the urban transportation system, it is particularly important to conduct in-depth analysis of the behavioral characteristics of urban travelers and construct a portrait. Traditional traveler behavior analysis mainly relies on questionnaires, transportation card data, vehicle GPS data, etc. These methods have problems such as high data collection costs and limited coverage, and it is difficult to fully and accurately capture the dynamic behavior of travelers. In recent years, with the development of information technology, especially the popularization of smart phones and the advancement of mobile communication technology, traveler behavior analysis based on mobile phone signaling data has gradually become a research hotspot. Mobile phone signaling data has the advantages of wide coverage, strong real-time performance, and large data volume. It can provide detailed information such as the location, time, and travel mode of travelers, providing a new data source for the construction of urban traveler portraits.

[0003] Although there have been many studies on mobile phone signaling data in recent years, most of them focus on the analysis of human geography characteristics within cities or regions. These studies are essentially to determine the macroscopic crowd dispersion points and personnel flow characteristics, but mobile phone signaling data hides the daily behavior of travelers. By further mining it, we can obtain similar travel characteristics of a group of travelers. This characteristic is the portrait of the group, which can be used as a reference for the refined management of urban traffic demand.

[0004] However, currently most data suppliers provide station data that have been processed from raw mobile phone signaling data, that is, the data set only involves the longitude and latitude and time that a traveler stays at a certain location. Existing research lacks specific methods for constructing traveler portraits for this type of data. At the same time, there are shortcomings such as imperfect data cleaning and processing processes, poor accuracy of travel mode identification methods, fuzzy classification of daily travel patterns, and disconnection between macro results and micro analysis. Summary of the invention

[0005] The purpose of the present invention is to overcome the shortcomings and deficiencies of the prior art and to provide a city traveler portrait system based on mobile phone signaling data. The system aims to achieve a comprehensive analysis and portrait construction of the behavior characteristics of city travelers by integrating functions such as data cleaning, travel chain extraction, travel mode identification, travel pattern identification, and travel path flow identification, so as to provide a more accurate and efficient traveler behavior analysis tool and provide a scientific basis for urban traffic planning and management.

[0006] To achieve the above purpose, the technical solution provided by the present invention is: a city traveler portrait system based on mobile phone signaling data, comprising:

[0007] The data import module is used to load the station point data and administrative division data that have been processed by the mobile phone signaling data supplier, and pre-process the station point data in combination with the administrative division data to obtain station point data that is not repeated, has longitude and latitude within the administrative division of the studied city, and has at least 2 station point data for each traveler;

[0008] The trip chain extraction module is used to identify the job and residence attributes of each station data in the data import module, match two station data with adjacent time into one OD data, eliminate meaningless data after calculating the travel time, travel distance and travel speed, and identify the travel type according to the job and residence attributes of the starting and ending points of the OD data;

[0009] The travel mode identification module is used to divide the OD data obtained in the travel chain extraction module into multiple intervals according to the travel distance, randomly select OD data from each interval, call the Amap route planning API, and calculate the mean travel time of each group of selected OD data under the Amap route planning of different travel modes; use the logarithmic Gaussian mixture model to cluster the travel time to obtain n clusters, match the n travel modes with the n clusters according to the calculated mean travel time of different travel modes, and obtain the travel mode identification results of all OD data;

[0010] The daily travel mode recognition module is used to draw a curve image of "distance from home-time", compress the curve image into a feature image, input the feature image into a convolutional autoencoder to extract a feature vector, use the K-means algorithm to classify the feature vectors extracted by the convolutional autoencoder into different categories, trace back the feature image, and superimpose the feature images of the same type to obtain a superimposed image showing the daily travel mode;

[0011] The travel path flow identification module is used to divide the studied city into traffic zones, identify whether there are a large number of OD data between the two traffic zones where the starting and ending points of each OD data are located, and judge the flow size of the travel path from a macro perspective;

[0012] The traveler portrait module is used to extract the various travel characteristics of each traveler obtained by the travel chain extraction module, travel mode identification module, daily travel pattern identification module and travel path flow identification module as multiple labels, and summarize the labels to form a crowd portrait.

[0013] Furthermore, the data import module includes a data loading module and a data preprocessing module, wherein:

[0014] The data loading module reads the csv format station data processed by the mobile phone signaling data supplier from the local, and the specific required fields are: the variable imsi representing the traveler ID, the longitude lon, latitude lat, the station start time time1 and the station end time time2 of each station; reads the shp format administrative division data of the studied city from the local to form the administrative division boundary of the city;

[0015] The data preprocessing module is used to screen the validity of the station data, as follows:

[0016] For repeated abnormal data, they are directly eliminated in the context of tens of millions of mobile phone signaling data;

[0017] Based on Python language, geopandas library is used to match the longitude and latitude of the station point data with the administrative boundaries of the studied city to determine whether the longitude and latitude of the station point data are within the scope of the city, and the station point data whose longitude and latitude are not within the scope of the studied city are eliminated;

[0018] The amount of station data for each traveler is counted. Travelers with less than 2 station data are directly eliminated against the background of tens of millions of mobile phone signaling data.

[0019] Furthermore, the travel chain extraction module includes a job-residence attribute recognition module, an OD data matching module and a travel type recognition module, wherein:

[0020] The work-residence attribute identification module is used to identify the residence, work or social attributes of each stop point, as follows:

[0021] Define the time window for home attribute recognition as 0:00-7:00 a.m. and the time window for work attribute recognition as 9:00-5:00 p.m. for each station data point. Calculate the number of minutes overlapped with the above two time windows and store the calculation results in the new fields home_overlap and work_overlap respectively.

[0022] Calculate the dwelling duration of a single point data by using time2-time1, in minutes;

[0023] For travelers with two stop data in a single day, compare the values ​​of home_overlap and work_overlap. If home_overlap is greater than or equal to work_overlap and greater than half of the stop duration, mark the stop data as "home"; if work_overlap is greater than or equal to home_overlap and greater than half of the stop duration, mark the stop data as "work";

[0024] For travelers with more than two stopover data in a single day, if home_overlap is greater than or equal to work_overlap, the stopover data is marked as "home"; if work_overlap is greater than home_overlap, the stopover data is marked as "work";

[0025] Extract the unique longitude and latitude of each traveler under the job-residence attribute mark, and update the original data through Boolean indexing, so that the station data with the same longitude and latitude are marked with the same job-residence attribute, and the station data that have not been successfully identified with the job-residence attribute are all marked as "social", indicating social attributes;

[0026] The OD data matching module matches the adjacent stationary point data of each traveler. The adjacent stationary point data are used as the starting and ending points of a single trip according to the time sequence. The time interval between time2 of the first stationary point data and time1 of the second stationary point data is used as the travel time. The straight-line distance between the longitude and latitude of the adjacent stationary point data is calculated using the WGS84 coordinate system and the semi-haversine formula as the travel distance. The ratio of the travel distance to the travel time is used as the travel speed. Finally, the OD data with specific characteristics of travel time within 5 to 180 minutes, travel distance greater than 0, and travel speed between 3 and 80 km / h are retained;

[0027] The travel type identification module identifies the travel type corresponding to each OD data according to the attributes of the origin and destination points, as follows:

[0028] If one of the starting and ending points is home, the OD data is marked as "home travel", otherwise it is marked as "non-home travel";

[0029] If the starting point is of home attribute and the destination is of work attribute, or the starting point is of work attribute and the destination is of home attribute, then the OD data is marked as "commuting trip"; if the starting point is of work attribute and the destination is of social attribute, or the starting point is of social attribute and the destination is of work attribute, or both the starting and the destination are of work attribute and the arrival time is before 17:00, then the OD data is marked as "business trip"; the rest of the cases are marked as "social activities".

[0030] Furthermore, the travel mode identification module identifies the main travel mode used for a single trip from the five travel modes of walking, cycling, driving, bus and subway, as follows:

[0031] All OD data are divided into multiple intervals according to the travel distance. When the travel distance is within 3 km, every 500 m is divided into an interval; when the travel distance is greater than 3 km, every 1 km is divided into an interval.

[0032] According to the actual situation of the studied cities, the identification limits of walking and cycling are set based on the travel distance: when the travel distance is greater than the identification limit of walking, walking is not considered; when the travel distance is greater than the identification limit of cycling, neither walking nor cycling is considered;

[0033] Randomly select the required number of OD data for each interval, call the Amap route planning API, plan routes for various travel modes for each set of OD data selected, return the travel time corresponding to each travel mode, and calculate the average travel time for each travel mode;

[0034] For all OD data in the same interval, a logarithmic Gaussian mixture model is established, and the parameters of the logarithmic Gaussian mixture model are estimated by the expectation maximization algorithm. Then, clustering is achieved through probability calculation, and the OD data are divided into five clusters with significantly different travel times.

[0035] The five travel modes are mapped to five clusters in the order of the mean travel time. The travel mode corresponding to each cluster is used as the travel mode identification result of the OD data of the cluster. The same operation is performed on the OD data of all intervals to obtain the main travel mode identification results of all OD data.

[0036] Furthermore, the daily travel mode recognition module includes a feature image generation module, a feature vector extraction module and a travel mode classification module, wherein:

[0037] The feature image generation module is used to generate a feature image that reflects the distance from home as a function of time of day as input to the feature vector extraction module, as follows:

[0038] The WGS84 coordinate system and the haversine formula are used to calculate the straight-line distance between all the stationary data of each traveler on a single day and their home attribute stationary data as the "distance from home", with the distance from home as the ordinate and the time from 0 to 24 hours on the day as the abscissa. The origin corresponds to 0 for both the distance from home and the time. A curve image of "distance from home-time" is drawn, and horizontal segments are drawn within the duration of the stationary data. The endpoints of adjacent segments are connected by straight lines to represent the movement process between the coordinate points of the two stationary data. If the leftmost horizontal segment does not start at 0 hours or the rightmost horizontal segment does not end at 24 hours, it will be extended to 0 hours or 24 hours.

[0039] Convert each traveler's daily distance from home and time to the range of 0-25, create a blank image in RGB format and JPEG format with a size of 25*25, draw a dividing line in the blank image based on the curve image of "distance from home-time", fill the area above the dividing line with blue with RGB value (0,0,255), and fill the area below the dividing line with red with RGB value (255,0,0), and generate a feature image as the input of the subsequent convolutional autoencoder;

[0040] The feature vector extraction module extracts the feature vector representing the original features of the feature image by constructing a convolutional autoencoder. The feature vector extraction module includes an encoder part and a decoder part. The specific details are as follows:

[0041] a. The encoder part consists of the first convolutional layer, the first maximum pooling layer, the second convolutional layer and the second maximum pooling layer connected in series. It takes an image of size (25, 25, 3) as input and sets the green channel value of RGB to 0, where 25 and 25 are the length and width units of the image respectively, and 3 is the number of channels;

[0042] The first convolutional layer uses 16 convolution kernels with a kernel size of (3,3), ReLU activation function, same padding, and L2 regularization, with an output size of (25,25,16). It then passes through the first maximum pooling layer with a pooling window of (2,2) and the same padding, reducing the feature image size to (13,13,16).

[0043] The second convolutional layer uses 16 convolution kernels with a kernel size of (3,3), an activation function of ReLU, the same padding, and an output size of (13,13,16). It then passes through the second maximum pooling layer with a pooling window of (2,2) and the same padding, reducing the feature map size to (7,7,16).

[0044] b. The decoder part consists of the first upsampling layer, the third convolutional layer, the second upsampling layer, the fourth convolutional layer, the fifth convolutional layer and the sixth convolutional layer connected in series, and performs the following operations:

[0045] The first upsampling layer upsamples the feature map size from (7,7,16) to (14,14,16), and then passes through the third convolutional layer with 16 convolution kernels of size (3,3), ReLU activation function, same padding, and output size (14,14,16).

[0046] The second upsampling layer upsamples the feature map size from (14, 14, 16) to (28, 28, 16), and then passes through the fourth convolutional layer with 16 convolution kernels of size (3, 3), ReLU activation function, same padding, and output size (28, 28, 16).

[0047] Through the fifth convolutional layer, the number of convolution kernels is 16, the size is (2,2), the activation function is ReLU, and the output size is adjusted to (27,27,16);

[0048] Finally, through the sixth convolutional layer, the number of convolution kernels is 3, the size is (3,3), the activation function is ReLU, and the output size is (25,25,3);

[0049] The feature vector extraction module uses the Adam optimizer, sets the learning rate to 0.001, uses the mean square error as the loss function, sets the batch size to 256, and the training rounds to 50 to extract the feature vector of the feature image;

[0050] The travel mode classification module uses the K-means algorithm to cluster the feature vectors output by the feature vector extraction module, uses the elbow method to determine the number of clusters, marks the category to which each traveler's travel mode belongs after clustering, traces back the feature image, and superimposes the feature images of the same category to obtain a superimposed image to represent the daily travel mode.

[0051] Furthermore, the travel path flow identification module identifies whether each trip occurs between two traffic areas with high overall travel frequencies through the travel frequencies between the traffic areas at a macro level, as follows:

[0052] Extract all longitude and latitude in the OD data and merge them into a two-dimensional array;

[0053] The number of traffic zones is artificially set according to the urban land area, and the K-means algorithm is used to cluster all the longitudes and latitudes in the two-dimensional array to determine the cluster to which each longitude and latitude belongs. Each cluster represents each traffic zone, and the longitude and latitude of the centroid of each traffic zone are used to replace the longitude and latitude in the OD data to generate new OD data.

[0054] Count the travel frequency between each pair of centroids, sort each pair of centroids from large to small according to the travel frequency, and then draw the travel expectation line map on the map of the studied city based on the sorting using the cartopy library in Python language. In the drawing process, connect each pair of centroids with a straight line in turn. As the travel frequency increases, gradually increase the line width, reduce the transparency, and gradually transition the color from green with RGB value (0,255,0) to yellow with RGB value (255,255,0) and then to red with RGB value (255,0,0). Finally, draw the travel expectation line map within the studied city.

[0055] For each traveler's OD data of the day, the travel expectation line chart is traced back. If at least one corresponding RGB value in the traveler's OD data is between (255, 0, 0) and (255, 165, 0), the traveler is labeled with a "high-flow interval travel" label; if at least one corresponding RGB value in the traveler's OD data is between (255, 165, 0) and (165, 255, 0), the traveler is labeled with a "medium-flow interval travel" label; if it does not match either of the above two label situations, the traveler is labeled with a "low-flow interval travel" label.

[0056] Furthermore, the traveler portrait module extracts the following features of each traveler obtained by the travel chain extraction module, the travel mode identification module, the daily travel mode identification module and the travel path flow identification module as labels:

[0057] The travel time, travel distance, travel speed and travel type characteristics of each traveler for each trip obtained by the trip chain extraction module;

[0058] The travel mode characteristics of each traveler for each trip obtained by the travel mode identification module;

[0059] The daily travel pattern characteristics of each traveler for each trip obtained by the daily travel pattern recognition module;

[0060] The travel path flow characteristics of each traveler's travel for each trip are obtained by the travel path flow identification module;

[0061] Finally, travelers with the same label are formed into a group portrait of travelers.

[0062] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0063] 1. For the station data processed by the mobile phone signaling data supplier, a complete set of data preprocessing and travel chain extraction solutions is proposed to improve the data cleaning and processing process of this type of data.

[0064] 2. Combining the path planning function of Amap with the clustering algorithm of the logarithmic Gaussian mixture model enhances the accuracy of travel mode identification for OD data.

[0065] 3. The feature extraction function of the convolutional autoencoder is combined with the application of the K-means clustering algorithm to generate an overlay image, which gives the recognition results of each traveler's daily travel pattern stronger interpretability.

[0066] 4. Correlating the travel frequency between traffic zones at the macro level with the travel paths of individual travelers at the micro level will help technicians better identify the size of the flow of people on each traveler's travel path.

[0067] 5. Based on the station data processed by the mobile phone signaling data supplier, a set of urban traveler portrait system has been constructed, which is conducive to more accurate and efficient analysis of traveler behavior and empowering traffic demand management. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 Schematic diagram of the relationship between various modules of the system of the present invention.

[0069] Figure 2 This is the effect diagram of the travel mode identification module dividing the OD data into five clusters with significantly different travel times, where Cluster 1 to Cluster 5 represent the first cluster to the fifth cluster respectively, and Mean represents the mean calculated by taking the logarithm of the travel time within the cluster.

[0070] Figure 3 A curve image diagram of "distance from home-time" of a single traveler on a single day is drawn for the feature image generation module, where Distance from Home (m) represents the distance from home in meters, and Time represents the date and time of the day.

[0071] Figure 4 Schematic diagram of feature image of feature image generation module.

[0072] Figure 5 The figure is a schematic diagram of the feature vector clustering effect of the travel mode classification module, where Principal Component 1 and Principal Component 2 represent the two principal components after the feature vector size is reduced to two dimensions for the convenience of displaying the clustering effect, and Cluster 0 to Cluster 7 represent modes 1 to 8 respectively.

[0073] Figure 6 The overlay image representing the daily travel mode is the final output of the travel mode classification module.

[0074] Figure 7 A schematic diagram of the travel expectation line map of Guangzhou City is drawn for the travel path flow identification module. DETAILED DESCRIPTION

[0075] The present invention is further described in detail below in conjunction with embodiments and drawings, but the embodiments of the present invention are not limited thereto.

[0076] like Figures 1 to 6As shown, this embodiment discloses a city traveler portrait system based on mobile phone signaling data, which is a city traveler portrait system developed using Python language and can be run on Windows devices. The relationship between the modules of the system is as follows Figure 1 As shown, it includes:

[0077] The data import module is used to load the station point data and administrative division data that have been processed by the mobile phone signaling data supplier, and pre-process the station point data in combination with the administrative division data to obtain station point data that is not repeated, has longitude and latitude within the administrative division of the studied city, and has at least 2 station point data for each traveler;

[0078] The trip chain extraction module is used to identify the job and residence attributes of each station data in the data import module, match two station data with adjacent time into one OD data, eliminate meaningless data after calculating the travel time, travel distance and travel speed, and identify the travel type according to the job and residence attributes of the starting and ending points of the OD data;

[0079] The travel mode identification module is used to divide the OD data obtained in the travel chain extraction module into multiple intervals according to the travel distance, randomly select OD data from each interval, call the Amap route planning API, and calculate the mean travel time of each group of selected OD data under the Amap route planning of different travel modes; use the logarithmic Gaussian mixture model to cluster the travel time to obtain n clusters, match the n travel modes with the n clusters according to the calculated mean travel time of different travel modes, and obtain the travel mode identification results of all OD data;

[0080] The daily travel mode recognition module is used to draw a curve image of "distance from home-time", compress the curve image into a feature image, input the feature image into a convolutional autoencoder to extract a feature vector, use the K-means algorithm to classify the feature vectors extracted by the convolutional autoencoder into different categories, trace back the feature image, and superimpose the feature images of the same type to obtain a superimposed image showing the daily travel mode;

[0081] The travel path flow identification module is used to divide the studied city into traffic zones, identify whether there are a large number of OD data between the two traffic zones where the starting and ending points of each OD data are located, and judge the flow size of the travel path from a macro perspective;

[0082] The traveler portrait module is used to extract the various travel characteristics of each traveler obtained by the travel chain extraction module, travel mode identification module, daily travel pattern identification module and travel path flow identification module as multiple labels, and summarize the labels to form a crowd portrait.

[0083] Specifically, the data import module includes a data loading module and a data preprocessing module, wherein:

[0084] The data loading module reads the csv format station data within the Pearl River Delta urban agglomeration that has been processed by the mobile phone signaling data supplier from the local area. The specific fields are: the variable imsi representing the traveler ID, the longitude lon, latitude lat, the station start time time1 and the station end time time2 of each station; the studied city is set as Guangzhou, and the administrative division data of Guangzhou in shp format is read from the local area to form the administrative division boundary of Guangzhou;

[0085] The data preprocessing module is used to screen the validity of the station data, as follows:

[0086] For repeated abnormal data, they are directly eliminated in the context of tens of millions of mobile phone signaling data;

[0087] Based on Python language, geopandas library is used to match the longitude and latitude of the station data with the administrative boundaries of Guangzhou City to determine whether the longitude and latitude of the station data are within the scope of Guangzhou City, and the station data whose longitude and latitude are not within the scope of Guangzhou City are eliminated;

[0088] The amount of station data for each traveler is counted. Travelers with less than 2 station data are directly eliminated against the background of tens of millions of mobile phone signaling data.

[0089] Specifically, the travel chain extraction module includes a job-residence attribute recognition module, an OD data matching module and a travel type recognition module, wherein:

[0090] The work-residence attribute identification module is used to identify the residence, work or social attributes of each stop point, as follows:

[0091] Define the time window for home attribute recognition as 0:00-7:00 a.m. and the time window for work attribute recognition as 9:00-5:00 p.m. for each station data point. Calculate the number of minutes overlapped with the above two time windows and store the calculation results in the new fields home_overlap and work_overlap respectively.

[0092] Calculate the dwelling duration of a single point data by using time2-time1, in minutes;

[0093] For travelers with two stop data in a single day, compare the values ​​of home_overlap and work_overlap. If home_overlap is greater than or equal to work_overlap and greater than half of the stop duration, mark the stop data as "home"; if work_overlap is greater than or equal to home_overlap and greater than half of the stop duration, mark the stop data as "work";

[0094] For travelers with more than two stopover data in a single day, if home_overlap is greater than or equal to work_overlap, the stopover data is marked as "home"; if work_overlap is greater than home_overlap, the stopover data is marked as "work";

[0095] Extract the unique longitude and latitude of each traveler under the job-residence attribute mark, and update the original data through Boolean indexing, so that the station data with the same longitude and latitude are marked with the same job-residence attribute, and the station data that have not been successfully identified with the job-residence attribute are all marked as "social", indicating social attributes;

[0096] The OD data matching module matches the adjacent stationary point data of each traveler. The adjacent stationary point data are used as the starting and ending points of a single trip according to the time sequence. The time interval between time2 of the first stationary point data and time1 of the second stationary point data is used as the travel time. The straight-line distance between the longitude and latitude of the adjacent stationary point data is calculated using the WGS84 coordinate system and the semi-haversine formula as the travel distance. The ratio of the travel distance to the travel time is used as the travel speed. Finally, the OD data with the specific characteristics of travel time within 5 to 180 minutes and travel distance greater than 0 are retained. At the same time, considering that the travel speed is calculated from the straight-line distance and the travel time, the travel speed corresponding to the retained OD data should also be between 3 and 80 km / h.

[0097] The travel type identification module identifies the travel type corresponding to each OD data according to the attributes of the origin and destination points, as follows:

[0098] If one of the starting and ending points is home, the OD data is marked as "home travel", otherwise it is marked as "non-home travel";

[0099] If the starting point is of home attribute and the destination is of work attribute, or the starting point is of work attribute and the destination is of home attribute, then the OD data is marked as "commuting trip"; if the starting point is of work attribute and the destination is of social attribute, or the starting point is of social attribute and the destination is of work attribute, or both the starting and the destination are of work attribute and the arrival time is before 17:00, then the OD data is marked as "business trip"; the rest of the cases are marked as "social activities".

[0100] Specifically, the travel mode identification module is used to identify the main travel mode used for a single trip from five travel modes: walking, cycling, driving, bus, and subway, as follows:

[0101] All OD data are divided into multiple intervals according to the travel distance. When the travel distance is within 3 km, every 500 m is divided into an interval; when the travel distance is greater than 3 km, every 1 km is divided into an interval.

[0102] According to the actual situation of the prevalence of electric bicycles in Guangzhou, the identification limit for walking is set at 3km and the identification limit for cycling is set at 6km based on the travel distance. When the travel distance is greater than 3km, walking is not considered; when the travel distance is greater than 6km, walking and cycling are not considered.

[0103] Randomly select 5,000 OD data for each interval, call the Amap route planning API, plan routes for various travel modes for each set of OD data selected, return the travel time corresponding to each travel mode, and calculate the average travel time for each travel mode;

[0104] For all OD data in the same interval, a log-Gaussian mixture model is established, and the parameters of the log-Gaussian mixture model are estimated by the expectation maximization algorithm. Then, clustering is achieved through probability calculation, and the OD data are divided into five clusters with significantly different travel times, such as Figure 2 As shown;

[0105] The five travel modes are mapped to five clusters in the order of the mean travel time. The travel mode corresponding to each cluster is used as the travel mode identification result of the OD data of the cluster. The same operation is performed on the OD data of all intervals to obtain the main travel mode identification results of all OD data.

[0106] Specifically, the daily travel mode recognition module includes a feature image generation module, a feature vector extraction module and a travel mode classification module, wherein:

[0107] The feature image generation module is used to generate a feature image that reflects the distance from home as a function of time of day as input to the feature vector extraction module, as follows:

[0108] The WGS84 coordinate system and the haversine formula are used to calculate the straight-line distance between all the station data of each traveler on a single day and their home attribute station data as the "distance from home". The distance from home is used as the ordinate and the time from 0 to 24 hours on the day is used as the abscissa. The origin corresponds to 0 for both the distance from home and the time. The "distance from home-time" curve is plotted as shown in the figure below. Figure 3As shown, a horizontal line segment is drawn within the dwelling duration period of the stationary point data, and the endpoints of adjacent line segments are connected by straight lines to represent the movement process between the coordinate points of the two stationary point data. If the leftmost horizontal line segment does not start from the 0h time or the rightmost horizontal line segment does not end at the 24h time, it will be extended to the 0h time or the 24h time;

[0109] Convert each traveler's daily distance from home and time to the range of 0-25, create a blank image in RGB format and JPEG format with a size of 25*25, draw a dividing line in the blank image based on the "distance from home-time" curve image, fill the area above the dividing line with blue with an RGB value of (0,0,255), and fill the area below the dividing line with red with an RGB value of (255,0,0), and generate the following: Figure 4 The feature image shown is used as the input of the subsequent convolutional autoencoder;

[0110] The feature vector extraction module extracts the feature vector representing the original features of the feature image by constructing a convolutional autoencoder. The module includes an encoder part and a decoder part. The details are as follows:

[0111] a. The encoder part consists of the first convolutional layer, the first maximum pooling layer, the second convolutional layer and the second maximum pooling layer connected in series. It takes an image of size (25, 25, 3) as input and sets the green channel value of RGB to 0, where 25 and 25 are the length and width units of the image respectively, and 3 is the number of channels;

[0112] The first convolutional layer uses 16 convolution kernels with a kernel size of (3,3), ReLU activation function, same padding, and L2 regularization, with an output size of (25,25,16). It then passes through the first maximum pooling layer with a pooling window of (2,2) and the same padding, reducing the feature image size to (13,13,16).

[0113] The second convolutional layer uses 16 convolution kernels with a kernel size of (3,3), an activation function of ReLU, the same padding, and an output size of (13,13,16). It then passes through the second maximum pooling layer with a pooling window of (2,2) and the same padding, reducing the feature map size to (7,7,16).

[0114] b. The decoder part consists of the first upsampling layer, the third convolutional layer, the second upsampling layer, the fourth convolutional layer, the fifth convolutional layer and the sixth convolutional layer connected in series, and performs the following operations:

[0115] In the first upsampling layer, the feature map size is upsampled from (7,7,16) to (14,14,16), and then it passes through the third convolutional layer with 16 convolution kernels of size (3,3), ReLU activation function, same padding, and output size (14,14,16).

[0116] In the second upsampling layer, the feature map size is upsampled from (14, 14, 16) to (28, 28, 16), and then passes through the fourth convolutional layer with 16 convolution kernels of size (3, 3), ReLU activation function, same padding, and output size (28, 28, 16).

[0117] Through the fifth convolutional layer, the number of convolution kernels is 16, the size is (2,2), the activation function is ReLU, and the output size is adjusted to (27,27,16);

[0118] Finally, through the sixth convolutional layer, the number of convolution kernels is 3, the size is (3,3), the activation function is ReLU, and the output size is (25,25,3);

[0119] The feature vector extraction module uses the Adam optimizer, sets the learning rate to 0.001, uses the mean square error as the loss function, sets the batch size to 256, and the training rounds to 50 to extract the feature vector of the feature image;

[0120] The travel mode classification module uses the K-means algorithm to cluster the feature vectors output by the feature vector extraction module, and uses the elbow method to determine the number of clusters to be 8. The clustering results are as follows: Figure 5 As shown, the category of each traveler's travel mode on that day is marked, the feature image is traced back, and the feature images of the same category are superimposed to obtain the superimposed image as shown in Figure 6 As shown in the figure, the eight images from left to right and from top to bottom can represent the following sun travel modes:

[0121] Mode 1: Leaving Guangzhou earlier or going out to work earlier and not returning home on the same day;

[0122] Model 2: social activities in the evening after regular work hours;

[0123] Mode 3: Entering Guangzhou from other places or returning home after staying overnight outside;

[0124] Mode 4: Only one temporary trip;

[0125] Mode 5: Leaving Guangzhou late or going out to work late and not returning home on the same day;

[0126] Mode 6: Regular commuting trip;

[0127] Mode 7: Regular commuting with business trips or multiple social activities;

[0128] Pattern 8: Commuting at a later time and having business trips without returning home on the same day or having multiple social activities without returning home on the same day.

[0129] Specifically, the travel path flow identification module identifies whether each trip occurs between two traffic areas with high overall travel frequencies through the travel frequencies between traffic areas at a macro level, as follows:

[0130] Extract all the longitude and latitude in the OD data and merge them into a two-dimensional array.

[0131] The number of traffic zones is set to 400 according to the land area of ​​Guangzhou. The K-means algorithm is used to cluster all longitudes and latitudes in the two-dimensional array to determine the cluster to which each longitude and latitude belongs. Each cluster represents each traffic zone. The longitude and latitude of the centroid of each traffic zone are used to replace the longitude and latitude in the OD data to generate new OD data.

[0132] Count the travel frequency between each pair of centroids, sort each pair of centroids from large to small according to the travel frequency, and then draw the travel expectation line diagram on the Guangzhou map based on the Python language using the cartopy library. Figure 7 As shown, in the drawing process, each pair of centroids is connected with a straight line in turn. As the travel frequency increases, the line width is gradually increased, the transparency is reduced, and the color gradually transitions from green with an RGB value of (0,255,0) to yellow with an RGB value of (255,255,0) and then to red with an RGB value of (255,0,0). Finally, the travel expectation line map within the scope of Guangzhou is drawn;

[0133] For each traveler's OD data of the day, the travel expectation line chart is traced back. If at least one corresponding RGB value in the traveler's OD data is between (255, 0, 0) and (255, 165, 0), the traveler is labeled with a "high-flow interval travel" label; if at least one corresponding RGB value in the traveler's OD data is between (255, 165, 0) and (165, 255, 0), the traveler is labeled with a "medium-flow interval travel" label; if it does not match either of the above two label situations, the traveler is labeled with a "low-flow interval travel" label.

[0134] Specifically, the traveler portrait module extracts the following features of each traveler obtained by the travel chain extraction module, the travel mode identification module, the daily travel mode identification module and the travel path flow identification module as labels:

[0135] The travel time, travel distance, travel speed and travel type characteristics of each traveler for each trip obtained by the trip chain extraction module;

[0136] The travel mode characteristics of each traveler for each trip obtained by the travel mode identification module;

[0137] The daily travel pattern characteristics of each traveler for each trip obtained by the daily travel pattern recognition module;

[0138] The travel path flow characteristics of each traveler's travel for each trip are obtained by the travel path flow identification module;

[0139] Finally, travelers with the same label are formed into a group portrait of travelers.

[0140] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. The urban traveler portrait system based on mobile phone signaling data is characterized by: include: The data import module is used to load the station point data and administrative division data that have been processed by the mobile phone signaling data supplier, and pre-process the station point data in combination with the administrative division data to obtain station point data that is not repeated, has longitude and latitude within the administrative division of the studied city, and has at least 2 station point data for each traveler; The travel chain extraction module is used to identify the job and residence attributes of each station data in the data import module, match two station data with adjacent time into one OD data, eliminate meaningless data after calculating the travel time, travel distance and travel speed, and identify the travel type according to the job and residence attributes of the starting and ending points of the OD data; The travel mode identification module is used to divide the OD data obtained in the travel chain extraction module into multiple intervals according to the travel distance, randomly select OD data from each interval, call the Amap route planning API, and calculate the mean travel time of each group of selected OD data under the Amap route planning of different travel modes; use the logarithmic Gaussian mixture model to cluster the travel time to obtain n clusters, match the n travel modes with the n clusters according to the calculated mean travel time of different travel modes, and obtain the travel mode identification results of all OD data; The daily travel mode recognition module is used to draw a curve image of "distance from home-time", compress the curve image into a feature image, input the feature image into a convolutional autoencoder to extract a feature vector, use the K-means algorithm to classify the feature vectors extracted by the convolutional autoencoder into different categories, trace back the feature image, and superimpose the feature images of the same type to obtain a superimposed image showing the daily travel mode; The travel path flow identification module is used to divide the studied city into traffic zones, identify whether there are a large number of OD data between the two traffic zones where the starting and ending points of each OD data are located, and judge the flow size of the travel path from a macro perspective; The traveler portrait module is used to extract the various travel characteristics of each traveler obtained by the travel chain extraction module, travel mode identification module, daily travel pattern identification module and travel path flow identification module as multiple labels, and summarize the labels to form a crowd portrait.

2. The urban traveler portrait system based on mobile phone signaling data according to claim 1 is characterized in that: The data import module includes a data loading module and a data preprocessing module, wherein: The data loading module reads the csv format station data processed by the mobile phone signaling data supplier from the local, and the specific required fields are: the variable imsi representing the traveler ID, the longitude lon, latitude lat, the station start time time1 and the station end time time2 of each station; reads the shp format administrative division data of the studied city from the local to form the administrative division boundary of the city; The data preprocessing module is used to screen the validity of the station data, as follows: For repeated abnormal data, they are directly eliminated in the context of tens of millions of mobile phone signaling data; Based on Python language, geopandas library is used to match the longitude and latitude of the station point data with the administrative boundaries of the studied city to determine whether the longitude and latitude of the station point data are within the scope of the city, and the station point data whose longitude and latitude are not within the scope of the studied city are eliminated; The amount of station data for each traveler is counted. Travelers with less than 2 station data are directly eliminated against the background of tens of millions of mobile phone signaling data.

3. The urban traveler portrait system based on mobile phone signaling data according to claim 2 is characterized in that: The travel chain extraction module includes a job-residence attribute recognition module, an OD data matching module and a travel type recognition module, wherein: The work-residence attribute identification module is used to identify the residence, work or social attributes of each stop point, as follows: Define the time window for home attribute recognition as 0:00-7:00 a.m. and the time window for work attribute recognition as 9:00-5:00 p.m. for each station data point. Calculate the number of minutes overlapped with the above two time windows and store the calculation results in the new fields home_overlap and work_overlap respectively. Calculate the dwelling duration of a single point data by using time2-time1, in minutes; For travelers with two stop data in a single day, compare the values ​​of home_overlap and work_overlap. If home_overlap is greater than or equal to work_overlap and greater than half of the stop duration, mark the stop data as "home"; if work_overlap is greater than or equal to home_overlap and greater than half of the stop duration, mark the stop data as "work"; For travelers with more than two stopover data in a single day, if home_overlap is greater than or equal to work_overlap, the stopover data is marked as "home"; if work_overlap is greater than home_overlap, the stopover data is marked as "work"; Extract the unique longitude and latitude of each traveler under the job and residence attribute mark, and update the original data through Boolean indexing, so that the station data with the same longitude and latitude are marked with the same job and residence attributes, and the station data that have not been successfully identified with the job and residence attributes are all marked as "social", indicating social attributes; The OD data matching module matches the adjacent stationary point data of each traveler. The adjacent stationary point data are used as the starting and ending points of a single trip according to the time sequence. The time interval between time2 of the first stationary point data and time1 of the second stationary point data is used as the travel time. The straight-line distance between the longitude and latitude of the adjacent stationary point data is calculated using the WGS84 coordinate system and the semi-haversine formula as the travel distance. The ratio of the travel distance to the travel time is used as the travel speed. Finally, the OD data with specific characteristics of travel time within 5 to 180 minutes, travel distance greater than 0, and travel speed between 3 and 80 km / h are retained; The travel type identification module identifies the travel type corresponding to each OD data according to the attributes of the origin and destination points, as follows: If one of the starting and ending points is home, the OD data is marked as "home travel", otherwise it is marked as "non-home travel"; If the starting point is of home attribute and the destination is of work attribute, or the starting point is of work attribute and the destination is of home attribute, then the OD data is marked as "commuting trip"; if the starting point is of work attribute and the destination is of social attribute, or the starting point is of social attribute and the destination is of work attribute, or both the starting and the destination are of work attribute and the arrival time is before 17:00, then the OD data is marked as "business trip"; the rest of the cases are marked as "social activities".

4. The urban traveler portrait system based on mobile phone signaling data according to claim 3 is characterized in that: The travel mode identification module identifies the main travel mode used for a single trip from five travel modes: walking, cycling, driving, bus, and subway, as follows: All OD data are divided into multiple intervals according to the travel distance. When the travel distance is within 3 km, every 500 m is divided into an interval; when the travel distance is greater than 3 km, every 1 km is divided into an interval. According to the actual situation of the studied cities, the identification limits of walking and cycling are set based on the travel distance: when the travel distance is greater than the identification limit of walking, walking is not considered; when the travel distance is greater than the identification limit of cycling, neither walking nor cycling is considered; Randomly select the required number of OD data for each interval, call the Amap route planning API, plan routes for various travel modes for each set of OD data selected, return the travel time corresponding to each travel mode, and calculate the average travel time for each travel mode; For all OD data in the same interval, a logarithmic Gaussian mixture model is established, and the parameters of the logarithmic Gaussian mixture model are estimated by the expectation maximization algorithm. Then, clustering is achieved through probability calculation, and the OD data are divided into five clusters with significantly different travel times. The five travel modes are mapped to five clusters in the order of the mean travel time. The travel mode corresponding to each cluster is used as the travel mode identification result of the OD data of the cluster. The same operation is performed on the OD data of all intervals to obtain the main travel mode identification results of all OD data.

5. The urban traveler portrait system based on mobile phone signaling data according to claim 4 is characterized in that: The daily travel mode recognition module includes a feature image generation module, a feature vector extraction module and a travel mode classification module, wherein: The feature image generation module is used to generate a feature image that reflects the distance from home as a function of time of day as input to the feature vector extraction module, as follows: The WGS84 coordinate system and the haversine formula are used to calculate the straight-line distance between all the stationary data of each traveler on a single day and their home attribute stationary data as the "distance from home". The distance from home is used as the ordinate and the time from 0 to 24 hours on the same day is used as the abscissa. The origin corresponds to 0 for both the distance from home and the time. A curve image of "distance from home-time" is drawn. Horizontal segments are drawn within the duration of the stationary data. The endpoints of adjacent segments are connected by straight lines to represent the movement process between the coordinate points of the two stationary data. If the leftmost horizontal segment does not start at 0 hours or the rightmost horizontal segment does not end at 24 hours, it will be extended to 0 hours or 24 hours. Convert each traveler's daily distance from home and time to the range of 0-25, create a blank image in RGB format and JPEG format with a size of 25*25, draw a dividing line in the blank image based on the curve image of "distance from home-time", fill the area above the dividing line with blue with RGB value (0,0,255), and fill the area below the dividing line with red with RGB value (255,0,0), and generate a feature image as the input of the subsequent convolutional autoencoder; The feature vector extraction module extracts the feature vector representing the original features of the feature image by constructing a convolutional autoencoder. The feature vector extraction module includes an encoder part and a decoder part. The specific details are as follows: a. The encoder part consists of the first convolutional layer, the first maximum pooling layer, the second convolutional layer and the second maximum pooling layer connected in series. It takes an image of size (25, 25, 3) as input and sets the green channel value of RGB to 0, where 25 and 25 are the length and width units of the image respectively, and 3 is the number of channels; The first convolutional layer uses 16 convolution kernels with a kernel size of (3,3), ReLU activation function, same padding, and L2 regularization, with an output size of (25,25,16). It then passes through the first maximum pooling layer with a pooling window of (2,2) and the same padding, reducing the feature image size to (13,13,16). The second convolutional layer uses 16 convolution kernels with a kernel size of (3,3), an activation function of ReLU, the same padding, and an output size of (13,13,16). It then passes through the second maximum pooling layer with a pooling window of (2,2) and the same padding, reducing the feature map size to (7,7,16). b. The decoder part consists of the first upsampling layer, the third convolutional layer, the second upsampling layer, the fourth convolutional layer, the fifth convolutional layer and the sixth convolutional layer connected in series, and performs the following operations: The first upsampling layer upsamples the feature map size from (7,7,16) to (14,14,16), and then passes through the third convolutional layer with 16 convolution kernels of size (3,3), ReLU activation function, same padding, and output size (14,14,16). The second upsampling layer upsamples the feature map size from (14, 14, 16) to (28, 28, 16), and then passes through the fourth convolutional layer with 16 convolution kernels of size (3, 3), ReLU activation function, same padding, and output size (28, 28, 16). Through the fifth convolutional layer, the number of convolution kernels is 16, the size is (2,2), the activation function is ReLU, and the output size is adjusted to (27,27,16); Finally, through the sixth convolutional layer, the number of convolution kernels is 3, the size is (3,3), the activation function is ReLU, and the output size is (25,25,3); The feature vector extraction module uses the Adam optimizer, sets the learning rate to 0.001, uses the mean square error as the loss function, sets the batch size to 256, and the training rounds to 50 to extract the feature vector of the feature image; The travel mode classification module uses the K-means algorithm to cluster the feature vectors output by the feature vector extraction module, uses the elbow method to determine the number of clusters, marks the category to which each traveler's travel mode belongs after clustering, traces back the feature image, and superimposes the feature images of the same category to obtain a superimposed image to represent the daily travel mode.

6. The urban traveler portrait system based on mobile phone signaling data according to claim 5 is characterized in that: The travel path flow identification module identifies whether each trip occurs between two traffic areas with high overall travel frequencies through the travel frequencies between traffic areas at a macro level, as follows: Extract all longitude and latitude in the OD data and merge them into a two-dimensional array; The number of traffic zones is artificially set according to the urban land area, and the K-means algorithm is used to cluster all the longitudes and latitudes in the two-dimensional array to determine the cluster to which each longitude and latitude belongs. Each cluster represents each traffic zone, and the longitude and latitude of the centroid of each traffic zone are used to replace the longitude and latitude in the OD data to generate new OD data. Count the travel frequency between each pair of centroids, sort each pair of centroids from large to small according to the travel frequency, and then draw the travel expectation line map on the map of the studied city based on the sorting using the cartopy library in Python language. In the drawing process, connect each pair of centroids with a straight line in turn. As the travel frequency increases, gradually increase the line width, reduce the transparency, and gradually transition the color from green with RGB value (0,255,0) to yellow with RGB value (255,255,0) and then to red with RGB value (255,0,0). Finally, draw the travel expectation line map within the studied city. For each traveler's OD data of the day, the travel expectation line chart is traced back. If at least one corresponding RGB value in the traveler's OD data is between (255, 0, 0) and (255, 165, 0), the traveler is labeled with a "high-flow interval travel" label; if at least one corresponding RGB value in the traveler's OD data is between (255, 165, 0) and (165, 255, 0), the traveler is labeled with a "medium-flow interval travel" label; if it does not match the above two label situations, the traveler is labeled with a "low-flow interval travel" label.

7. The urban traveler portrait system based on mobile phone signaling data according to claim 6 is characterized in that: The traveler portrait module extracts the following features of each traveler obtained by the travel chain extraction module, the travel mode identification module, the daily travel mode identification module and the travel path flow identification module as labels: The travel time, travel distance, travel speed and travel type characteristics of each traveler for each trip obtained by the trip chain extraction module; The travel mode characteristics of each traveler for each trip obtained by the travel mode identification module; The daily travel pattern characteristics of each traveler for each trip obtained by the daily travel pattern recognition module; The travel path flow characteristics of each traveler's travel for each trip are obtained by the travel path flow identification module; Finally, travelers with the same label are formed into a group portrait of travelers.

Images