Method and system for extracting combined travel chains of urban agglomerations based on mobile phone signaling big data
By using data preprocessing and automated machine learning algorithms to identify travel modes within the urban agglomeration, the problem of not being able to identify multi-way combined travel chains in the existing technology is solved, and the precise extraction of multi-way combined travel chains within the urban agglomeration is achieved. Expand the application value of mobile phone signaling big data.
Patent Information
- Application Number
- CN202310128934.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-02-17
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2043-02-17
AI Technical Summary
The existing travel method identification method based on mobile signaling data is limited to a single mode of transportation, and it is impossible to identify and extract multiple combined travel chains within the urban agglomeration, and the data mining depth and breadth of mobile signaling big data are insufficient.
A multi-way combined travel chain extraction method based on mobile phone signaling big data was designed. An abnormal data was removed through data preprocessing algorithms, and an automated machine learning algorithm combining gradient enhancement, integrated learning, traditional classification and deep learning models were identified. , and set parameters according to the principles of transportation engineering to develop an intercity travel chain extraction algorithm for urban agglomerations.
It has realized the accurate identification and extraction of multi-meaning combined travel chains within urban agglomerations, expanded the scope of mobile phone signaling big data research, and met the multi-faceted application of intercity joint travel needs of urban agglomerations.
Smart Images

Figure CN116208927B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of integrated transportation planning, and in particular to a method and system for extracting combined travel chains of urban agglomerations based on mobile phone signaling big data. Background Art
[0002] Urban agglomerations are important spatial carriers for development based on my country's vast territorial conditions. They include 19 urban agglomerations, such as the Beijing-Tianjin-Hebei urban agglomeration, the Yangtze River Delta urban agglomeration, the Guangdong-Hong Kong-Macao Greater Bay Area, and the Chengdu-Chongqing urban agglomeration. With the continuous improvement of the country's comprehensive three-dimensional transportation network, it has become a national strategy to build a convenient and smooth urban agglomeration transportation network, promote the integrated development of multi-modal transportation modes, and promote the rapidity and convenience of travel services.
[0003] Traffic data collection and analysis technology is the key technology for realizing integrated multi-modal transportation services in urban agglomerations. Compared with traditional traffic data collection methods such as SP questionnaires, radar, ultrasound, induction coils, and video monitoring, mobile signaling has unique advantages such as high real-time performance, wide coverage, and rich full-sample information. It can well meet the wide-area and all-weather data needs in the research of intermodal travel chains within urban agglomerations, and fill the partial gaps in existing demand-oriented traffic data sources.
[0004] my country has a complete communication base station system that can realize the collection and storage of massive mobile phone signaling data, but the depth and breadth of data mining of mobile phone signaling big data are still insufficient, and the relevant technologies for transportation applications based on mobile phone signaling data are still imperfect, mainly reflected in: 1) Most of the current research on mobile phone signaling data is limited to traditional small areas, such as a city or urban area, and has not yet migrated to the extraction of travel chains in large-scale urban agglomerations; 2) There is a significant demand for multi-modal combined travel by inter-city passengers. The existing travel mode identification method based on mobile phone signaling data is limited to a single mode of transportation and cannot identify and extract multi-modal combined transportation travel chains. Therefore, it is necessary to propose a method for extracting multi-modal combined travel chains of urban agglomerations, which is conducive to further mining the data value of current mobile phone signaling big data, so as to realize various applications in the fields of transportation and urban planning. Summary of the invention
[0005] The technical problem to be solved by the present invention is: to develop a method for extracting multi-modal combined travel chains of urban agglomerations based on mobile phone signaling big data, to design a corresponding data preprocessing algorithm based on the acquired mobile phone signaling big data within the urban agglomeration, to develop a specific automated machine learning algorithm to identify the travel mode of each travel trajectory point in the mobile phone signaling data, to set corresponding parameters for different travel modes, and finally to extract the multi-modal combined travel chain of the urban agglomeration.
[0006] The present invention adopts the following technical solutions to solve the above technical problems:
[0007] The method for extracting travel chains of urban agglomeration combinations based on mobile phone signaling big data proposed in the present invention comprises the following steps:
[0008] S1. Obtain mobile phone signaling big data within the city cluster from communication operators;
[0009] S2. Develop corresponding data preprocessing algorithms for the four types of abnormal data commonly seen in mobile phone signaling big data, namely duplication, default, drift, and ping-pong, to achieve data preprocessing of mobile phone signaling big data and reduce data noise;
[0010] S3, integrates four types of models: gradient boosting, ensemble learning, traditional classification algorithm, and deep learning to develop an automated machine learning algorithm to identify the travel mode of each trajectory point;
[0011] S4. Based on the basic principles of traffic engineering, set corresponding parameter tables for different travel modes;
[0012] S5. Develop an algorithm for extracting inter-city travel chains of urban agglomerations, and extract multi-modal combined travel chains of urban agglomerations from mobile phone signaling big data.
[0013] Further, the mobile phone signaling big data described in step S1 specifically requires the following fields: variable u representing the user ID, the longitude lat of each mobile phone signaling track point i of user u i , latitude lon i , timestamp t i , where i∈[1, 2, 3, …, n], n is the total number of mobile phone signaling trajectory points.
[0014] Further, the specific process of the mobile phone signaling big data preprocessing in step S2 is as follows:
[0015] (1) For repeated abnormal data, directly delete it in the context of billions of mobile phone signaling big data;
[0016] (2) For abnormal data of default types, directly delete them in the context of billions of mobile phone signaling big data;
[0017] (3) For drift type abnormal data, by calculating the time difference T between two trajectory points i , spatial distance L i and speed v i To identify drift data:
[0018]
[0019] Here, i∈[2, 3, …, n], r is the radius of the earth, 6378.137 km, and the upper limit of the speed threshold between two trajectory points is limited to 600 km / h, that is, v i>600km / h when the drift data is deleted;
[0020] (4) For abnormal data of ping-pong type, the azimuth angle θ between two adjacent trajectory points is used. i To identify ping-pong data:
[0021]
[0022] Here, i∈[2, 3, …, n]. Considering that the ping-pong data feature is a large-range bounce of the single-point azimuth angle and the angle change between the actual travel trajectory points is generally a small angle uniform change, the upper limit of the azimuth angle difference threshold is limited to 60°, that is, θ gap When the angle is >60°, the ping-pong data will be deleted.
[0023] Further, the specific process of establishing the automated machine learning algorithm described in step S3 is as follows:
[0024] S301, gradient boosting models integrate three algorithms: XGBoost, CatBoost, and LightGBM; ensemble learning models integrate two algorithms: Random Forest and ExtraTrees; traditional classification models integrate two algorithms: KNN and LinearModel; deep learning models integrate three algorithms: NeuralNetTorch, NeuralNetMxnet, and NeuralNetFastAi;
[0025] S302: Manually calibrate the travel mode at some trajectory points i i , mode i ∈[walk, road, highway, rail, highrail], where walk is walking, road is urban road, highway is expressway, rail is ordinary railway, and highrail is high-speed railway;
[0026] S303, the calibrated travel mode i As the true value, the lat of user u at trajectory point i i ,lon i ,t i 、v i It is used as a variable to input into the gradient boosting, ensemble learning, traditional classification, and deep learning models, and the results output by different models are weighted averaged to obtain the travel mode at trajectory point i. i .
[0027] Further, the travel mode parameter table described in step S4 is set as follows:
[0028] (1) In walking mode, the speed threshold v walk , time threshold T walk , the spatial threshold is L walk =v walk T walk , minimum data point threshold N walk ;
[0029] (2) In urban road traffic, the speed threshold v road , time threshold T road , spatial threshold L road =v road T road , minimum data point threshold N road ;
[0030] (3) In highway traffic mode, the speed threshold v highway , time threshold T highway , spatial threshold L highway =v highway T highway , minimum data point threshold N highway ;
[0031] (4) In ordinary railway transportation, the speed threshold v rail , time threshold T rail , spatial threshold L rail =v rail T rail , minimum data point threshold N rail ;
[0032] (5) In high-speed rail transportation, the speed threshold v highrail , time threshold T highrail , spatial threshold L highrail =v highrail T highrail , minimum data point threshold N highrail .
[0033] Furthermore, the specific steps of the urban agglomeration intercity travel chain extraction algorithm described in step S5 are as follows:
[0034] S501. Extracting data D of user u n ={x1, x2…x n}, where x i = {lon i ,lat i , t i , mode i};
[0035] S502, spatial neighborhood threshold set Eps1 = [L walk , L road, L highway , L rail , L highrail ], representing the spatial neighborhoods corresponding to different travel modes;
[0036] S503, time neighborhood threshold set Eps2 = [T walk , T road , T highway , T rail , T highrail ], representing the time neighborhoods corresponding to different travel modes;
[0037] S504, minimum data point threshold set Eps3 = [N walk , N road , N highway , N rail , N highrail ], each element represents the data point threshold corresponding to different travel modes;
[0038] S505, trajectory data point threshold Limit, representing the minimum number of data points for each travel chain, used to exclude abnormal travel chain results caused by a small number of local noise data;
[0039] S506, initialize data point label category labels n ;
[0040] S507, finding the core point in the data set based on the parameter sequence of the data points;
[0041] S508, starting from any unmarked core point, find its density reachable points under the parameter set constraints based on different travel modes, that is, neighboring points, and then classify these points into one category;
[0042] S509, continue recursion and looping until each data point is marked;
[0043] S510, counting the number of data points in different categories, and deleting categories with less than the trajectory data point threshold Limit;
[0044] S511. The different categories finally obtained represent different travel chains;
[0045] S512. Output the category corresponding to each data point: labels n ={label1, label2…label n}, where each label j Represents a travel chain C(u, j) of user u, where j represents the jth travel chain.
[0046] Furthermore, the present invention also proposes a system for extracting travel chains of urban agglomeration combinations based on mobile phone signaling big data, comprising:
[0047] The mobile phone signaling big data receiving module is used to obtain the mobile phone signaling big data within the city cluster from the communication operator, including: the variable u representing the user ID, the longitude lat of each mobile phone signaling track point i of user u i , latitude lon i , timestamp t i , where i∈[1, 2, 3, …, n], n is the total number of mobile phone signaling trajectory points;
[0048] The mobile phone signaling big data preprocessing module is used to develop corresponding data preprocessing algorithms for the four types of abnormal data commonly seen in mobile phone signaling big data, namely duplication, default, drift, and ping-pong, to reduce data noise;
[0049] The automated machine learning module is used to integrate four types of models: gradient boosting, ensemble learning, traditional classification algorithms, and deep learning to identify the travel mode of each trajectory point;
[0050] The inter-city travel chain extraction module of the urban agglomeration is used to develop the inter-city travel chain extraction algorithm of the urban agglomeration by combining the parameter tables set by different travel modes, and extract the multi-mode combined travel chain of the urban agglomeration from the mobile phone signaling big data.
[0051] Furthermore, the specific operation steps of the mobile phone signaling big data preprocessing module are as follows:
[0052] (1) For repeated abnormal data, directly delete it in the context of billions of mobile phone signaling big data;
[0053] (2) For abnormal data of default types, directly delete them in the context of billions of mobile phone signaling big data;
[0054] (3) For drift type abnormal data, by calculating the time difference Ti and spatial distance L between two trajectory points i and speed v i To identify drift data:
[0055]
[0056] Here, i∈[2, 3, …, n], r is the radius of the earth, 6378.137 km, and the upper limit of the speed threshold between two trajectory points is limited to 600 km / h, that is, v i >600km / h when the drift data is deleted;
[0057] (4) For abnormal data of ping-pong type, the azimuth angle θ between two adjacent trajectory points is used. i To identify ping-pong data:
[0058]
[0059] Here, i∈[2, 3, …, n]. Considering that the ping-pong data feature is a large-range bounce of the single-point azimuth angle and the angle change between the actual travel trajectory points is generally a small angle uniform change, the upper limit of the azimuth angle difference threshold is limited to 60°, that is, θ gap When the angle is >60°, the ping-pong data will be deleted.
[0060] Furthermore, the specific process of establishing the automated machine learning module is as follows:
[0061] Step 1: The gradient boosting model integrates three algorithms: XGBoost, CatBoost, and LightGBM; the ensemble learning model integrates two algorithms: Random Forest and ExtraTrees; the traditional classification model integrates two algorithms: KNN and LinearModel; the deep learning model integrates three algorithms: NeuralNetTorch, NeuralNetMxnet, and NeuralNetFastAi;
[0062] Step 2: Manually calibrate the travel mode at some trajectory points i i , mode i ∈[walk, road, highway, rail, highrail], where walk is walking, road is urban road, highway is expressway, rail is ordinary railway, and highrail is high-speed railway;
[0063] Step 3: Set the calibrated travel mode i As the true value, the lat of user u at trajectory point i i ,lon i ,t i 、v i It is used as a variable to input into the gradient boosting, ensemble learning, traditional classification, and deep learning models, and the results output by different models are weighted averaged to obtain the travel mode at trajectory point i. i .
[0064] Furthermore, the specific steps of the city cluster intercity travel chain extraction module are as follows:
[0065] Step 1: Extract user u’s data D n ={x1, x2…x n}, where x i = {lon i ,lat i , ti , mode i};
[0066] Step 2: Spatial neighborhood threshold set Eps1 = [L walk , L road , L highway , L rail , L highrail ], representing the spatial neighborhoods corresponding to different travel modes;
[0067] Step 3: Time neighborhood threshold set Eps2 = [T walk , T road , T highway , T rail , T highrail ], representing the time neighborhoods corresponding to different travel modes;
[0068] Step 4: Minimum data point threshold set Eps3 = [N walk , N road , N highway , N rail , N highrail ], each element represents the data point threshold corresponding to different travel modes;
[0069] Step 5: The trajectory data point threshold Limit represents the minimum number of data points for each travel chain, which is used to exclude abnormal travel chain results caused by a small number of local noise data;
[0070] Step 6: Initialize data point label categories labels n ;
[0071] Step 7: Find the core points in the data set based on the parameter sequence of the data points;
[0072] Step 8: Starting from any unmarked core point, find its density-reachable points under the parameter set constraints based on different travel modes, i.e., neighboring points, and then classify these points into one category;
[0073] Step 9: Continue recursively and looping until each data point is marked;
[0074] Step 10: Count the number of data points in different categories and delete the categories with less than the trajectory data point threshold Limit;
[0075] Step 11: The different categories finally obtained represent different travel chains;
[0076] Step 12: Output the category corresponding to each data point: labels n ={label1, label2…label n}, where each label j Represents a travel chain C(u, j) of user u, where j represents the jth travel chain.
[0077] Compared with the prior art, the present invention adopts the above technical solution and has the following technical effects:
[0078] The present invention proposes a method for extracting combined travel chains of urban agglomerations based on mobile phone signaling big data. The method targets multi-modal combined travel modes such as walking, urban roads, highways, ordinary railways, and high-speed railways, expands the research scope of mobile phone signaling big data to the scope of urban agglomerations, and accurately identifies and extracts multi-modal combined travel chains of urban agglomerations, which is of great significance for exploring the inter-city travel needs of urban agglomerations. BRIEF DESCRIPTION OF THE DRAWINGS
[0079] Figure 1 It is an overall flow chart of the method for extracting travel chains of urban agglomeration combinations based on mobile phone signaling big data of the present invention.
[0080] Figure 2 This is a schematic diagram of mobile phone signaling drift data.
[0081] Figure 3 It is a schematic diagram of mobile phone signaling ping-pong data.
[0082] Figure 4 It is a schematic diagram of automated machine learning fusion.
[0083] Figure 5 It is a flow chart of the algorithm for extracting multi-modal combined travel chains in urban agglomerations.
[0084] Figure 6 This is a schematic diagram of the algorithm for extracting multi-modal combined travel chains in urban agglomerations.
[0085] Figure 7 It is a schematic diagram of a three-dimensional travel chain with multiple modes in a non-geographical urban agglomeration. DETAILED DESCRIPTION
[0086] The embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be interpreted as limiting the present invention.
[0087] like Figure 1 As shown, the overall flow chart of the method for extracting the combined trip chain of urban agglomerations based on mobile phone signaling big data of the present invention is shown. The specific steps of the present invention are described in detail below using the Beijing-Tianjin-Hebei urban agglomeration as an example:
[0088] S1. Obtain the mobile phone signaling big data of the Beijing-Tianjin-Hebei urban agglomeration for one month from China Unicom. The specific fields required are: variable u representing user ID, longitude lat of each mobile phone signaling track point i of user u i , latitude lon i , timestamp t i , where i∈[1, 2, 3, …, n], n is the total number of mobile phone signaling trajectory points. The longitude and latitude of the obtained China Unicom mobile phone signaling data are visualized on the map.
[0089] S2. Develop corresponding data preprocessing algorithms for the four types of abnormal data commonly seen in mobile phone signaling big data, namely duplication, default, drift, and ping-pong, to achieve data preprocessing of mobile phone signaling big data and reduce data noise. Figure 2 A schematic diagram showing drift data referred to in the present invention, Figure 3 The specific process of abnormal data processing is as follows:
[0090] For abnormal data of repeated types, in the context of billions of mobile phone signaling big data, the processing method adopted by the present invention is direct deletion.
[0091] For abnormal data of the default type, in the context of billions of mobile phone signaling big data, the processing method adopted by the present invention is direct deletion.
[0092] For drift type abnormal data, calculate the time difference T between two trajectory points i , the spatial distance L between two trajectory points i , the velocity v between two trajectory points i To identify drift data:
[0093]
[0094] Here, i∈[2, 3, …, n], r is the radius of the earth, 6378.137 km, and the upper limit of the speed threshold between two trajectory points is limited to 600 km / h, that is, v i The drift data will be deleted when the speed is >600km / h.
[0095] For ping-pong type abnormal data, calculate the azimuth angle θ between two adjacent trajectory points i To identify ping-pong data:
[0096]
[0097] Here, i∈[2, 3, …, n]. Considering that the ping-pong data feature is a large-range bounce of the azimuth angle of a single point and the angle change between the actual travel trajectory points is generally a small angle uniform change, the upper limit of the azimuth angle difference threshold is limited to 60°, that is, θgap When the angle is >60°, the ping-pong data will be deleted.
[0098] In this embodiment, Table 1 is the original data of user 1 before data preprocessing, wherein the data of sequence numbers 6-7 are duplicate data. According to this method, the first duplicate data, i.e., the data of sequence number 6, is retained, and the data of sequence number 7 is deleted; the data of sequence numbers 11-12 are default data, wherein the data of sequence number 11 lacks time information, and the data of sequence number 12 lacks spatial information. According to this method, these two default data are directly deleted.
[0099] Table 1 Original data before data preprocessing
[0100]
[0101]
[0102] For the processing of the two abnormal data, drift data and ping-pong data, some pre-processing calculations need to be performed before preprocessing, as shown in the formula in step S2. After calculation, the results shown in Table 2 are obtained, where L represents the spatial distance between the next data point and the previous data point, T represents the time difference between the next data point and the previous data point, v represents the speed obtained by dividing L by T, θ represents the azimuth between the next data point and the previous data point, and θ gap Indicates the difference in azimuth between the next data point and the previous data point. According to this method, delete the data with speed v greater than 600km / h, such as the data with sequence number 20 in Table 2; delete θ gap Data greater than 60°, such as data No. 5 in Table 2.
[0103] Table 2 Pre-calculation results of drift data and ping-pong data
[0104] Serial number User Time t Latitude lat Longitudelon L T v θ <![CDATA[θ gap ]]> 1 1 2021 / 5 / 1 5:33 39.15527 117.1705 --- --- --- --- --- 2 1 2021 / 5 / 1 5:34 39.15855 117.1773 0.697399 60 11.62332 58.42849 --- 3 1 2021 / 5 / 1 5:35 39.16998 117.1836 1.384222 60 23.07036 23.14455 35.28394 4 1 2021 / 5 / 1 5:36 39.18616 117.1939 2.006909 60 33.44848 26.15917 3.014622 5 1 2021 / 5 / 1 5:38 39.2099 117.1768 3.025203 120 25.21003 330.8748 124.7156 6 1 2021 / 5 / 1 5:43 39.36195 117.0708 19.23373 300 64.11242 331.6708 0.796037 7 1 2021 / 5 / 1 5:51 39.38209 116.985 7.717507 480 16.07814 286.8904 44.78037 8 1 2021 / 5 / 1 5:54 39.46364 116.9361 10.00366 180 55.57587 335.1707 48.28032 9 1 2021 / 5 / 1 5:56 39.51594 116.901 6.5571 120 54.6425 332.6134 2.557309 10 1 2021 / 5 / 1 5:57 39.56378 116.8613 6.322406 60 105.3734 327.3941 5.219344 11 1 2021 / 5 / 1 5:58 39.60425 116.8356 5.01644 60 83.60733 333.9176 6.52354 12 1 2021 / 5 / 1 6:33 39.86499 116.3775 0.387208 1260 0.307308 70.13414 0.000315 13 1 2021 / 5 / 1 6:34 39.86787 116.3832 0.582092 60 9.701527 56.52222 13.61192 14 1 2021 / 5 / 1 6:40 39.80959 116.6026 19.83977 360 55.11047 109.081 52.55881 15 1 2021 / 5 / 1 6:41 39.76683 116.6494 6.219105 60 103.6517 139.9255 30.84442 16 1 2021 / 5 / 1 6:43 39.69162 116.7485 11.92223 120 99.35193 134.5915 5.333957 17 1 2021 / 5 / 1 6:45 39.57425 116.8548 15.92856 120 132.738 145.0907 10.49918 18 1 2021 / 5 / 1 6:46 39.55343 116.8681 2.583218 60 43.05364 153.7585 8.667797 19 1 2021 / 5 / 1 6:47 39.49046 116.9252 8.557353 60 142.6226 144.9877 8.770817 20 1 2021 / 5 / 1 6:48 40.49046 116.4523 43.58414 60 632.7421 335.1707 48.28032 …… …… …… …… …… …… …… …… …… ……
[0105] S3, integrating gradient boosting, ensemble learning, traditional classification algorithm, and deep learning models to develop automated machine learning algorithms to identify the travel mode of each trajectory point. Figure 4 The flowchart of the automated machine learning model of the present invention is as follows:
[0106] S301. Integrate XGBoost, CatBoost and LightGBM algorithms into gradient boosting models; integrate Random Forest and ExtraTrees algorithms into ensemble learning models; integrate KNN and LinearModel algorithms into traditional classification models; integrate NeuralNetTorch, NeuralNetMxnet and NeuralNetFastAi algorithms into deep learning models.
[0107] S302: Manually calibrate the travel mode at some trajectory points i i , mode i ∈[walk, road, highway, rail, highrail], walk is walking transportation, road is urban road transportation, highway is expressway, rail is ordinary railway, and highrail is high-speed railway.
[0108] S303, the calibrated travel mode i As the true value, the lat of user u at trajectory point i i ,lon i ,t i 、v i It is used as a variable to input into the gradient boosting, ensemble learning, traditional classification, and deep learning models, and the results output by different models are weighted averaged to obtain the travel mode at trajectory point i. i .
[0109] S4. Based on the basic principles of traffic engineering, set up corresponding parameter tables for different travel modes. The specific contents are as follows:
[0110] (1) In walking mode, the speed threshold v walk =6km / h, time threshold T walk = 60s, the space threshold is L walk =v walk T walk =0.1km, minimum data point threshold N walk =3;
[0111] (2) In urban road traffic, the speed threshold v road =90km / h, time threshold T road = 60s, space threshold L road =v road T road =1.5km, minimum data point threshold N road =5;
[0112] (3) In highway traffic mode, the speed threshold v highway =120km / h, time threshold T highway = 60s, space threshold L highway =v highway T highway =2.0km, minimum data point threshold N highway =5;
[0113] (4) In ordinary railway transportation, the speed threshold vrail =120km / h, time threshold T rail = 60s, space threshold L rail =v rail T rail =2.0km, minimum data point threshold N rail =5;
[0114] (5) In high-speed rail transportation, the speed threshold v highrail =360km / h, time threshold T highrail = 60s, space threshold L highrail =v highrail T highrail =6·0km, minimum data point threshold N highrail =5.
[0115] S5. Develop an algorithm for extracting inter-city travel chains in urban agglomerations, and extract multi-mode combined travel chains in urban agglomerations from mobile phone signaling big data. The algorithm flow is as follows: Figure 5 As shown, the algorithm is intuitively shown in Figure 6 The specific steps are as follows:
[0116] S501. Extracting data D of user u n ={x1, x2…x n}, where x i = {lon i ,lat i , t i , mode i}.
[0117] S502, spatial neighborhood threshold set Eps1 = [L walk , L road , L highway , L rail , L highrail ], representing the spatial neighborhoods corresponding to different travel modes;
[0118] S503, time neighborhood threshold set Eps2 = [T walk , T road , T highway , T rail , T highrail ], representing the time neighborhoods corresponding to different travel modes;
[0119] S504, minimum data point threshold set Eps3 = [N walk , N road , N highway , N rail , N highrail ], each element represents the data point threshold corresponding to different travel modes;
[0120] S505 , trajectory data point threshold Limit=50, representing the minimum number of data points for each travel chain, and is used to exclude abnormal travel chain results caused by a small number of local noise data.
[0121] S506, initialize data point label category labels n ;
[0122] S507, finding the core point in the data set based on the parameter sequence of the data points;
[0123] S508, starting from any unmarked core point, find its density reachable points under the parameter set constraints based on different travel modes, that is, neighboring points, and then classify these points into one category;
[0124] S509, continue recursion and looping until each data point is marked;
[0125] S510, counting the number of data points in different categories, and deleting categories with less than the trajectory data point threshold Limit;
[0126] S511. The different categories finally obtained represent different travel chains;
[0127] S512. Output the category corresponding to each data point: labels n ={label1, label2…label n}, where each label j Represents a travel chain C(u, j) of user u, where j represents the jth travel chain.
[0128] Through the city cluster combined travel chain extraction method proposed in the present invention, it is finally identified that there are 18,320,374 travel chains in total in the China Unicom mobile phone signaling data for one month. Figure 7 This is a three-dimensional schematic diagram of the travel chain of the user within a one-month period extracted by the present invention. It can be intuitively seen that the algorithm proposed by the present invention has a good recognition effect on the spatiotemporal trajectory points of mobile phone signaling big data.
[0129] The embodiment of the present invention also proposes a system for extracting travel chains of urban agglomerations based on mobile phone signaling big data, including a mobile phone signaling big data receiving module, a mobile phone signaling big data preprocessing module, an automated machine learning module, an urban agglomeration intercity travel chain extraction module, and a computer program that can be run on a processor. It should be noted that each module in the above system corresponds to the specific steps of the method provided in the embodiment of the present invention, and has the functional modules and beneficial effects corresponding to the execution method. For technical details not fully described in this embodiment, please refer to the method provided in the embodiment of the present invention.
[0130] The above embodiments are only for illustrating the technical idea of the present invention, and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution in accordance with the technical idea proposed by the present invention shall fall within the protection scope of the present invention.
Claims
1. A method for extracting combined travel chains of urban agglomerations based on mobile phone signaling big data. It is characterized in that The steps include: S1. Obtain mobile phone signaling big data within the city cluster. The specific fields required for mobile phone signaling big data are: variable u representing user ID, longitude lat of each mobile phone signaling track point i of user u i , latitude lon i , timestamp t i , where i∈[1,2,3,…,n], n is the total number of mobile phone signaling trajectory points; S2. Adopt corresponding data preprocessing algorithms for common abnormal data in mobile phone signaling big data to realize data preprocessing of mobile phone signaling big data; common abnormal data in mobile phone signaling big data include the following four types: duplicate data, default data, drift data, and ping-pong data; The specific process of the mobile phone signaling big data preprocessing is as follows: (1) For repeated abnormal data, directly delete it in the context of billions of mobile phone signaling big data; (2) For abnormal data of default types, directly delete them in the context of billions of mobile phone signaling big data; (3) For drift type abnormal data, by calculating the time difference T between two trajectory points i , spatial distance L i and speed v i To identify drift data: Here, i∈[2,3,…,n], r is the radius of the earth 6378.137km, and the upper limit of the speed threshold between two trajectory points is limited to 600km / h, that is, v i >600km / h when the drift data is deleted; (4) For abnormal data of ping-pong type, the azimuth angle θ between two adjacent trajectory points is used. i To identify ping-pong data: Here, i∈[2,3,…,n], and the upper limit of the azimuth difference threshold is limited to 60°, that is, θ gap >60° delete the ping-pong data; S3, integrates the automated machine learning algorithm including gradient boosting, ensemble learning, classification algorithm, and deep learning models to identify the travel mode of each trajectory point; the specific contents are as follows: S301, gradient boosting models integrate three algorithms: XGBoost, CatBoost, and LightGBM; ensemble learning models integrate two algorithms: Random Forest and ExtraTrees; classification algorithm models integrate two algorithms: KNN and LinearModel; deep learning models integrate three algorithms: NeuralNetTorch, NeuralNetMxnet, and NeuralNetFastAi; S302: Manually calibrate the travel mode at some trajectory points i i , mode i ∈[walk,road,highway,rail,highrail], where walk is walking, road is urban road, highway is expressway, rail is ordinary railway, and highrail is high-speed railway; S303, the calibrated travel mode i As the true value, the lat of user u at trajectory point i i ,lon i ,t i 、v i It is used as a variable to input into the gradient boosting, ensemble learning, classification algorithm, and deep learning models. The results output by different models are weighted averaged to obtain the travel mode at trajectory point i. i ; S4. Based on the basic principles of traffic engineering, set up corresponding parameter tables for different travel modes; the specific contents are: (1) In walking mode, the speed threshold v walk , time threshold T walk , the spatial threshold is L walk =v walk T walk , minimum data point threshold N walk ; (2) In urban road traffic, the speed threshold v road , time threshold T road , spatial threshold L road =v road T road , minimum data point threshold N road ; (3) In highway traffic mode, the speed threshold v highway , time threshold T highway , spatial threshold L highway =v highway T highway , minimum data point threshold N highway ; (4) In ordinary railway transportation, the speed threshold v rail , time threshold T rail , spatial threshold L rail =v rail T rail , minimum data point threshold N rail ; (5) In high-speed rail transportation, the speed threshold v highrail , time threshold T highrail , spatial threshold L highrail =v highrail T highrail , minimum data point threshold N highrail ; S5. Use the city cluster inter-city travel chain extraction algorithm to extract the city cluster multi-mode combined travel chain in the mobile phone signaling big data; the specific content is: S501. Extracting data D of user u n ={x 1 ,x 2 …x n }, where x i = {lon i ,lat i ,t i ,mode i }; S502, spatial neighborhood threshold set Eps1 = [L walk ,L road ,L highway ,L rail ,L highrail ], representing the spatial neighborhoods corresponding to different travel modes; S503, time neighborhood threshold set Eps2 = [T walk ,T road ,T highway ,T rail ,T highrail ], representing the time neighborhoods corresponding to different travel modes; S504, minimum data point threshold set Eps3 = [N walk ,N road ,N highway ,N rail ,N highrail ], each element represents the data point threshold corresponding to different travel modes; S505, trajectory data point threshold Limit, representing the minimum number of data points for each travel chain, used to exclude abnormal travel chain results caused by a small number of local noise data; S506, initialize data point label category labels n ; S507, finding the core point in the data set based on the parameter sequence of the data points; S508, starting from any unmarked core point, find its density reachable points under the parameter set constraints based on different travel modes, that is, neighboring points, and then classify these points into one category; S509, continue recursion and looping until each data point is marked; S510, counting the number of data points in different categories, and deleting categories with less than the trajectory data point threshold Limit; S511. The different categories finally obtained represent different travel chains; S512. Output the category corresponding to each data point: labels n ={label 1 ,label 2 …label n }, where each label j Represents a travel chain C(u,j) of user u, where j represents the jth travel chain.
2. A city cluster combined travel chain extraction system based on mobile phone signaling big data, It is characterized in that include: The mobile phone signaling big data receiving module is used to obtain the mobile phone signaling big data within the city cluster from the communication operator, including: the variable u representing the user ID, the longitude lat of each mobile phone signaling track point i of user u i , latitude lon i , timestamp t i , where i∈[1,2,3,…,n], n is the total number of mobile phone signaling trajectory points; The mobile phone signaling big data preprocessing module is used to use corresponding data preprocessing algorithms for the four common abnormal data types of repetition, default, drift, and ping-pong in mobile phone signaling big data to reduce data noise; the specific contents are as follows: (1) For repeated abnormal data, directly delete it in the context of billions of mobile phone signaling big data; (2) For abnormal data of default types, directly delete them in the context of billions of mobile phone signaling big data; (3) For drift type abnormal data, by calculating the time difference T between two trajectory points i , spatial distance L i and speed v i To identify drift data: Here, i∈[2,3,…,n], r is the radius of the earth 6378.137km, and the upper limit of the speed threshold between two trajectory points is limited to 600km / h, that is, v i >600km / h when the drift data is deleted; (4) For abnormal data of ping-pong type, the azimuth angle θ between two adjacent trajectory points is used. i To identify ping-pong data: Here, i∈[2,3,…,n], and the upper limit of the azimuth difference threshold is limited to 60°, that is, θ gap When >60°, delete the ping-pong data; The automated machine learning module is used to integrate four types of models: gradient boosting, ensemble learning, traditional classification algorithm, and deep learning to identify the travel mode of each trajectory point; the specific contents are as follows: Step 1: The gradient boosting model integrates three algorithms: XGBoost, CatBoost, and LightGBM; the ensemble learning model integrates two algorithms: Random Forest and ExtraTrees; the classification algorithm model integrates two algorithms: KNN and LinearModel; the deep learning model integrates three algorithms: NeuralNetTorch, NeuralNetMxnet, and NeuralNetFastAi; Step 2: Manually calibrate the travel mode at some trajectory points i i , mode i ∈[walk,road,highway,rail,highrail], where walk is walking, road is urban road, highway is expressway, rail is ordinary railway, and highrail is high-speed railway; Step 3: Set the calibrated travel mode i As the true value, the lat of user u at trajectory point i i ,lon i ,t i 、v i It is used as a variable to input into the gradient boosting, ensemble learning, classification algorithm, and deep learning models. The results output by different models are weighted averaged to obtain the travel mode at trajectory point i. i ; The inter-city travel chain extraction module of the urban agglomeration is used to extract the multi-mode combined travel chain of the urban agglomeration from the mobile phone signaling big data by combining the parameter tables set by different travel modes and adopting the inter-city travel chain extraction algorithm of the urban agglomeration. The specific contents are as follows: Step 1: Extract user u’s data D n ={x 1 ,x 2 …x n }, where x i = {lon i ,lat i ,t i ,mode i }; Step 2: Spatial neighborhood threshold set Eps1 = [L walk ,L road ,L highway ,L rail ,L highrail ], representing the spatial neighborhoods corresponding to different travel modes; Step 3: Time neighborhood threshold set Eps2 = [T walk ,T road ,T highway ,T rail ,T highrail ], representing the time neighborhoods corresponding to different travel modes; Step 4: Minimum data point threshold set Eps3 = [N walk ,N road ,N highway ,N rail ,N highrail ], each element represents the data point threshold corresponding to different travel modes; Step 5: The trajectory data point threshold Limit represents the minimum number of data points for each travel chain, which is used to exclude abnormal travel chain results caused by a small number of local noise data; Step 6: Initialize data point label categories labels n ; Step 7: Find the core points in the data set based on the parameter sequence of the data points; Step 8: Starting from any unmarked core point, find its density-reachable points under the parameter set constraints based on different travel modes, i.e., neighboring points, and then classify these points into one category; Step 9: Continue recursively and looping until each data point is marked; Step 10: Count the number of data points in different categories and delete the categories with less than the trajectory data point threshold Limit; Step 11: The different categories finally obtained represent different travel chains; Step 12: Output the category corresponding to each data point: labels n ={label 1 ,label 2 …label n }, where each label j Represents a travel chain C(u,j) of user u, where j represents the jth travel chain.
Citation Information
Patent Citations
Resident trip mode comprehensive judging method based on handset signaling data
CN105117789A
Travel mode identification method based on big data machine learning
CN109389240A
