Regional transportation mode identification method based on mobile phone signaling data

Abstract

The invention discloses a regional transportation mode identification method based on mobile phone signaling data. The method comprises the following steps: step 1, obtaining and preprocessing basic data of mobile phone signaling, a communication base station and a regional road network; step 2, matching a communication base station based on a tree structure with a road network; step 3, identifying a travel path of the user; 4, judging whether the base station sequence corresponding to the travel path of the user is a road station where a track line is located, and if so, identifying that the travel mode of the path is track travel; and if not, identifying the transportation modes of the car and the intercity bus based on an optimized gradient lifting algorithm. According to the method, the regional travel mode can be rapidly identified, and compared with the existing investigation and identification means, the method has the advantage of rapid and accurate identification, and has a high practical value for matching regional mass data.

Description

Technical Field

[0001] The present invention belongs to the field of data information identification of traffic travel characteristics of traffic big data, especially for the identification of cross-city travel modes at the regional level, including three travel modes: rail, car and intercity bus. Background Art

[0002] In recent years, with the rapid economic development and the evolution of urbanization, my country's urban spatial structure is developing from a single city to a regional group form such as an urban agglomeration and a metropolitan area. The accurate extraction of transportation modes is the basis for studying regional traffic behavior, and it is also an important guarantee for government decisions such as regional transportation construction and regional channel planning. Although the extraction of traffic features based on mobile phone signaling data within the city has gradually matured, for regional travel, because of its long travel distance, large time and space span, and the communication environment is significantly different from that within the city, the existing related research and theory may not be fully applicable to regional research. Based on this, the present invention proposes a set of traffic mode recognition algorithm models suitable for the region.

[0003] Mobile phone signaling data has the advantages of wide coverage, low cost, and large passive sample size. Compared with fixed-point video data, toll station data and other travel information data in regional travel, it has the advantages of more continuity and comprehensiveness. When users travel, mobile communication base stations will record the signaling data of the interaction between the user's mobile phone and the base station in real time, including the user's signaling unique identification code, the user's signaling interaction time, the longitude and latitude of the communication base station, the user's signaling type and other information. Based on this, combined with road, railway and other road network information, by matching the longitude and latitude of the user signaling data to the road network, and using the switching fluctuations and switching frequency between signaling data, the user's travel mode can be further identified. This method plays a supporting role in obtaining regional travel information such as the share ratio of regional traffic travel structure and the traffic volume of each travel mode, and can provide data analysis support for the new construction, renovation and expansion of major regional transportation infrastructure, optimization of transportation facility layout and regional transportation hub planning. Summary of the invention

[0004] In order to overcome the above-mentioned shortcomings of the prior art, the present invention proposes a method for identifying regional transportation modes based on mobile phone signaling data, aiming to provide a new technical method for identifying three transportation modes: rail, car and intercity bus in regional inter-city travel.

[0005] The technical solution adopted by the present invention to solve the technical problem is: a method for identifying regional transportation modes based on mobile phone signaling data, comprising the following steps:

[0006] Step 1: Acquisition and preprocessing of basic data of mobile phone signaling, communication base stations and regional road networks;

[0007] Step 2: Matching of communication base stations and road networks based on tree structure;

[0008] Step 3: Identify the user's travel path;

[0009] Step 4: Determine whether the base station sequence corresponding to the user's travel path is the station where the rail line is located. If so, identify the travel mode of the path as rail travel; if not, identify the travel modes of car and intercity bus based on the optimized gradient boosting algorithm.

[0010] Compared with the prior art, the present invention has the following positive effects:

[0011] In view of the limitations of current regional transportation travel mode identification methods using video data, toll station data and other data for identification, the present invention proposes to use mobile phone signaling data to identify regional transportation travel modes, and proposes a travel mode identification method based on tree structure (KD-Tree) communication base station and road network matching, regional travel trajectory matching based on Smith-Waterman algorithm, and dynamic time warping algorithm and optimized gradient boosting (eXtreme Gradient Boosting) algorithm to successfully identify regional transportation travel modes.

[0012] The advantages of the present invention are: first, data advantage. Mobile phone signaling data has the advantages of wide coverage, low cost, large passive sample size, etc. Compared with fixed-point video data, toll station data and other travel information data in regional travel, it has the advantages of more continuity and comprehensiveness, and proposes data preprocessing methods such as cleaning, deduplication and noise removal for basic data. Second, method advantage. A complete set of method systems from regional base station matching to regional travel path identification, and finally realizing regional travel mode identification are proposed. Through the proposed tree-structured (KD-Tree)-based communication base station and road network matching, regional travel trajectory matching based on the Smith-Waterman algorithm, and travel mode identification methods based on the dynamic time warping algorithm and the optimized gradient boosting (eXtreme GradientBoosting) algorithm, the regional travel mode can be quickly identified. Compared with the existing survey and identification methods, it has the advantages of fast and accurate identification, and has a high practical value for matching regional massive data. It is of great significance to analyze the current situation and problems of regional traffic, regional traffic management, and support future regional traffic planning and construction. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The present invention will now be described by way of example with reference to the accompanying drawings, in which:

[0014] Figure 1 is an overall identification flow chart of the present invention;

[0015] Figure 2 The figure is a process diagram of regional travel trajectory matching and identification based on the Smith-Waterman algorithm of the present invention, wherein: (a) initializing the score matrix, (b) calculating the score matrix, and (c) backtracking the optimal path;

[0016] Figure 3 This is a diagram of the process of identifying travel modes for cars and intercity buses using the optimized gradient boosting (eXtreme Gradient Boosting) algorithm of the present invention. DETAILED DESCRIPTION

[0017] A method for identifying regional transportation modes based on mobile phone signaling data, such as Figure 1 As shown, it includes four parts:

[0018] 1. Acquisition and processing of basic data of mobile phone signaling, communication base stations and regional road networks

[0019] The data used in the present invention include three types of data: mobile phone signaling data, communication base station data and regional road network data. Among them, the mobile phone signaling data refers to all signaling data generated by the interaction with the base station due to communication events such as surfing the Internet, calling, text messaging, and periodic updates (generally converted into data instead of representation) during the user's regional travel. Generally, all signaling data in 24 hours a day are collected in units of days. The data field includes the user's unique identification code, the user's signaling interaction time, the base station longitude and latitude of the user's signaling interaction, the base station area code, the base station cell code, and the user's signaling event type. The signaling data format is shown in Table 1; the communication base station data refers to all base station data within the user's regional travel range, generally the base station data of the city center, urban suburbs, and inter-city highway / railway areas of different cities during the regional travel process. The data field includes the base station area code number, the base station cell code, the base station longitude and latitude, and the base station address description. The base station data format is shown in Table 2; the regional road network data refers to all road / railway network data within the user's regional travel range, generally the road network information data of highways, express roads, urban trunk roads, railway networks, etc. The data field includes the road network name, section code, the longitude and latitude of the starting point of the section, and the longitude and latitude of the end point of the section. The road network data format is shown in Table 3.

[0020] Table 1 Example of mobile phone signaling data format

[0021]

[0022] Table 2 Base station data style example

[0023]

[0024] Table 3 Road network data style examples

[0025]

[0026] After obtaining the above data, the data is preprocessed by cleaning, deduplication, and noise removal: the signaling data is classified and aggregated according to the user, and then the signaling data of each user is sorted based on the time series; the switching speed between adjacent data is preliminarily calculated, and the data is filtered using Gaussian filtering to remove data with too high or too low switching speeds; the communication base stations are cleaned to remove duplicate data from the same base station; the road network data is preprocessed by segmentation to extract the longitude and latitude of the starting and ending points of each section, and the overly long sections are cut, and the starting and ending points of the cut sections are extracted respectively.

[0027] 2. Communication base station and road network matching algorithm based on tree structure (K-dimensional tree)

[0028] When matching communication base stations in other studies, the base station location is often simply used as the current traveler's location. However, due to the irregularity of road geometry, this traditional mobile phone signaling data processing algorithm has large errors, which has a significant negative impact on the reliability and accuracy of the processing results. Considering that the inter-regional base station layout environment is significantly different from that within the city, firstly, the regional road density is low and the road section length is long, which can reach tens of kilometers; secondly, due to environmental factors such as terrain and slope, the distance between parallel roads in the region is generally large, and one base station will hardly cover two roads. The reliability of the results of matching base stations and roads using the neighbor matching algorithm is guaranteed. Therefore, in order to more accurately describe the traveler's location with mobile phone signaling data, the present invention will construct a base station matching model based on the tree structure K-dimensional tree (abbreviated as KD-Tree) algorithm to match the positioning base station neighbors to the actual road. The specific principles are as follows:

[0029] KD-Tree is a special data structure based on dichotomy. When the sample size N is large enough and the samples are randomly distributed, the average time complexity of using KD-Tree to retrieve the nearest sample can be approximately O(logN), which has better computational performance than the traversal algorithm. The KD-Tree algorithm mainly includes two steps. The first is the initialization construction of KD-Tree. By modeling N data points in the plane, the sample coordinates are set to (x, y). First, the median of the x coordinates of the N initial samples is found, and the node i1 corresponding to the median is stored as the root node of the KD tree. The plane can be divided into two parts with the x coordinate of node i as the boundary. Secondly, consider a part of the initial samples, and find the median of the y coordinates of these initial samples, that is, node j1, based on the y axis. Node j1 is stored in the subtree of node i1, and the space of this part is divided into two subspaces. If there is only one sample in each of the two subspaces after the division, the two samples are stored as leaf nodes of i1 respectively; if the samples in the subspace can continue to be classified, continue to consider repeating the previous steps based on the x axis until there is only one sample in each subspace. In this cycle, a complete KD subtree can be constructed. The second is the nearest neighbor search, which uses the binary search method to search downward from the root node until a leaf node is found. The leaf node is used as the initial sample closest to the target node, and the current minimum distance is calculated, and then backtracking begins. The backtracking process needs to follow the following search principles: If the spatial range represented by this node (a hypersphere constructed with the target node as the center and the distance from the initial sample as the radius) intersects with the search range, the distance to the target node is calculated, otherwise it is not calculated. When the hypersphere with the target node as the center and the new current shortest distance as the radius does not intersect with other search ranges, the backtracking program terminates and outputs the final base station matching result.

[0030] Matching base stations to the nearest roads is essentially solving the nearest neighbor problem between two points in a two-dimensional plane. The previous approach was to use all road points as sample points and use KD-Tree to find the nearest points. However, due to the long road lengths between regions and the large number of base stations, the computational pressure and algorithm complexity of this algorithm are still relatively large. Considering the parameter characteristics of regional base stations and roads, the present invention optimizes and adjusts the KD-Tree model, constructs a base station matching method for use in a regional travel road network environment, and proposes the following two innovations:

[0031] (1) Use KD-Tree to find the k road segment endpoints closest to the base station

[0032] First, the road segments are divided into certain length ranges, and then the irregular actual roads are abstracted into plane lines containing only endpoint information, and all endpoints are used as samples to construct a KD-Tree; secondly, the base station location is used as the target node, and the nearest k points are queried through the KD-Tree. When the segment division threshold and k value are appropriate (for example, the segment division threshold is 3000m and the k value is 6), there is a high degree of confidence that the endpoint of the segment closest to the target node is included in these k points.

[0033] (2) Traverse the k road segments where the endpoints are found one by one to find the shortest distance

[0034] In the present invention, the road model is abstracted as a polyline set. Therefore, all roads containing the k endpoint information are traversed in sequence, and the shortest distance between the target node and each polyline on each road is queried to complete the base station matching. By finding the minimum value of all the shortest distances, that is, the shortest distance between the target point and the road network, traversing k roads, the minimum value of these k data is the shortest distance between the base station and the actual road.

[0035] 3. Regional Travel Trajectory Path Matching Method Based on Smith-Waterman Algorithm

[0036] Before identifying the user's travel mode, it is necessary to first match the user's travel trajectory to the road network. Through the above base station matching algorithm, the data sequence corresponding to the base station has been converted into a road network sequence. At this time, the further travel trajectory matching is to compare and match the road network sequence. For this problem, there are generally two methods: global search matching and local search matching. When the sample size is small, the global matching algorithm is suitable. When the regional travel involves a wide range, the base station coverage is large, and the amount of signaling data is huge, the use of the global matching algorithm will consume a lot of computing power, resulting in a slow computing speed, which is not suitable for large-scale data processing of regional travel. Therefore, the present invention proposes a travel trajectory matching algorithm based on the Smith-Waterman local comparison algorithm to construct a travel trajectory matching algorithm suitable for large-scale regional data. The algorithm is an algorithm for local sequence alignment (relative to global alignment) to find similar areas between two sequences. The purpose of the algorithm is not to compare the entire sequence, but to find fragments with high similarity in the two sequences. It has the following two advantages. First, it is a dynamic programming algorithm for local matching, which helps to distinguish and identify parallel similar sections; at the same time, because it focuses on local features, it can eliminate the interference of abnormal data to a certain extent. Figure 2 As shown in the figure, the algorithm is mainly divided into three steps: initializing the score matrix, calculating the score matrix, and backtracking the best path.

[0037] (1) Initialize the score matrix

[0038] Create an (m+1)×(n+1) matrix, where m and n are the lengths of the two sequences involved in the alignment. The first row and first column of the matrix are initialized to 0, such as Figure 2 (a) shown.

[0039] (2) Calculate the score matrix

[0040] Assign scores to matching and non-matching base station network sequences. Typically, matches receive positive scores and non-matches receive negative scores. Use a dynamic programming algorithm to fill the matrix according to the following rules:

[0041] If the two corresponding base station sequences are identical or similar, a matching score is added at the current position of the matrix.

[0042] If different, you can choose to add a mismatch score.

[0043] If the score of the current position is less than 0, it is set to 0, indicating that the comparison of the path is completed, such as Figure 2 (b) as shown.

[0044] (3) Best Path Backtracking

[0045] Backtrack to find the local alignment, starting from the path with the highest score until the starting point of the local alignment is found, such as Figure 2 (c) as shown.

[0046] When the present invention uses the above method to perform trajectory matching, it is necessary to first segment the long road section to ensure that the base station matching result has a high confidence level; on the other hand, even if the road density between regions is relatively low, a single base station may still cover multiple roads near the intersection of the road section. This mismatching result will cause great interference to the trajectory matching process and reduce the matching credibility and accuracy. Therefore, in order to improve the accuracy of trajectory matching.

[0047] The present invention proposes the following two innovations to optimize trajectory matching:

[0048] (1) Set up the sequence matching buffer

[0049] The present invention proposes to divide a long path into multiple sections according to a certain threshold value, and form a vacuum-like buffer zone near the intersection. The base station data matched to the road section within the buffer zone is not taken into consideration for trajectory matching. Near a road intersection, a single base station may cover multiple roads, which will interfere with path identification. Therefore, a buffer zone is established that includes the intersection and a 500m to 1000m road section near the intersection. All base station matching data in the buffer zone are considered invalid data and are not used for base station matching. Although the establishment of the buffer zone sacrifices some base station data, due to the small number of regional road intersections and the long regional travel time, more signaling data is generated. Such a loss is still within an acceptable range, and the accuracy of road section matching is guaranteed.

[0050] (2) Propose a score normalization comparison index

[0051] The Smith-Waterman algorithm often uses the number of sequence matching elements as an indicator to measure sequence similarity in gene sequence comparison, but this evaluation index system is not suitable for regional travel base station road network matching. Due to the difference in the length of each path sequence, the longer the sequence length, the greater the possibility of more matching elements. The scale of each path matching result is different and cannot be directly compared, so it needs to be normalized. By calculating twice the score of the two sequences to be compared and dividing it by the number of sequence elements corresponding to the two sequences, the normalized comparison score value is obtained, which is used to determine the final score and improve the accuracy of trajectory matching.

[0052] 4. Travel mode identification method based on dynamic time warping algorithm and optimized gradient boosting algorithm

[0053] Dynamic Time Warping (DTW) calculates the similarity between two sequences by constructing a distance matrix, with the aim of further determining the travel path. Specifically, DTW achieves alignment by finding an optimal path so that the cumulative distance on this path is minimized. Due to environmental interference or other force majeure factors, individual information data may be missing, which leads to differences in the length of the signaling data sequence generated by different travelers due to location updates. Therefore, the traditional Euclidean distance may not be able to effectively obtain the distance (similarity) between the two sequences. Based on the above analysis, the present invention proposes to use the DTW dynamic time warping algorithm to compare the distances between time series of different lengths. To align the sequences, an m×n matrix grid is constructed, and the matrix element d(i,j) represents the distance between the i-th element in the standard sequence and the j-th element in the test sequence. Any path that meets the continuity, monotonicity and boundary conditions is a set of valid DTW results, and the path with the shortest cumulative distance is defined as the distance between the two sequences. The present invention is based on the dynamic time warping (DTW) algorithm and backtracks the shortest path to obtain the path. The actual travel sequence and the travel paths in the region can be compared. The path with the highest score is the user's travel path. If the base station sequence corresponding to the travel path is the road station where the rail line is located, the travel mode of the path is successfully identified as rail travel; otherwise, it is road travel. At this time, it is necessary to further identify the two modes of road travel: cars and passenger buses.

[0054] The present invention proposes to use optimized gradient boosting eXtreme Gradient Boosting (XGboost for short) to identify regional travel cars and intercity buses on the basis of trajectory path recognition. In order to better distinguish between cars and intercity buses, the present invention comprehensively considers the overall distribution characteristics of the speeds of the two, and selects the following characteristic indicators as input, including: maximum speed, average speed, percentile speed, and the ratio of high-speed points to low-speed points.

[0055] (1) Maximum speed: the maximum speed during the trip.

[0056] (2) Average speed: the ratio of the total travel distance to the travel time.

[0057] (3) Percentile speed: the percentile speed of the travel speed sequence, including the 95%, 75%, 50%, 25% and 5% percentile speeds.

[0058] (4) Ratio of high-speed points to low-speed points: This feature can effectively measure the degree of dispersion of speed distribution and can better distinguish between cars and intercity buses.

[0059] XGBoos optimizes the objective function by iteratively adding new decision trees, thereby continuously improving the performance of the model. The objective function of XGBoost consists of two parts: the loss function and the regularization term. The loss function is used to measure the degree of fit of the model to the training data, while the regularization term is used to prevent overfitting. From the signaling data matched to road travel, half is used as a test set and half as a training set. The above feature parameters are used as classification feature variables to optimize the important hyperparameters in XGBoost. It is proposed to select the learning rate, random sampling ratio, the proportion of columns of random sampling, and the maximum tree depth as hyperparameters for optimization. The grid parameter adjustment method is used for hyperparameter optimization. The basic idea is exhaustive search: among all candidate parameter selections, all parameter combinations are tried through loop traversal and cross-validation to find the parameter combination that minimizes the model error. The optimal hyperparameter combination proposed is 0.1, 0.9, 0.5, and 5 respectively. Then identify the transportation modes of cars and intercity buses.

[0060] In summary, the three travel modes of regional rail, car and intercity bus are successfully identified by the dynamic time warping algorithm and the optimized gradient boosting algorithm.

Claims

1. A method for identifying regional transportation modes based on mobile phone signaling data, characterized in that: The steps include: Step 1: Acquisition and preprocessing of basic data of mobile phone signaling, communication base stations and regional road networks; Step 2: Matching of communication base stations and road networks based on tree structure; Step 3: Identify the user's travel path; Step 4: Determine whether the base station sequence corresponding to the user's travel path is the station where the rail line is located. If so, identify the travel mode of the path as rail travel; if not, identify the travel modes of car and intercity bus based on the optimized gradient boosting algorithm.

2. The method for identifying regional transportation modes based on mobile phone signaling data according to claim 1 is characterized in that: The mobile phone signaling data in step 1 are all signaling data generated by the interaction with the base station due to Internet access, calls, text messages, and periodic updates of communication events during the user's regional travel. All signaling data in 24 hours a day are collected in units of days, and the data field includes the user's unique identification code, user signaling interaction time, the longitude and latitude of the base station for user signaling interaction, the base station area code, the base station cell code, and the user signaling event type; the communication base station data are all base station data within the user's regional travel range, including base station data in the city center, suburbs, and inter-city highways and railway areas of different cities during the regional travel, and the data field includes the base station area code number, base station cell code, base station longitude and latitude, and base station address description; the regional road network data are all road and railway network data within the user's regional travel range, including road network information data of highways, express roads, urban trunk roads, and railway networks, and the data field includes the road network name, section code, longitude and latitude of the starting point of the section, and longitude and latitude of the end point of the section.

3. The method for identifying regional transportation modes based on mobile phone signaling data according to claim 1 is characterized in that: The data preprocessing method in step one includes: cleaning, deduplication and noise removal of data: classifying and grouping signaling data according to users, and then sorting the signaling data of each user based on time series; calculating the switching speed between adjacent data, filtering the data using Gaussian filtering, and removing data with too high or too low switching speeds; cleaning communication base stations and removing duplicate data from the same base station; preprocessing the road network data by segmentation, extracting the longitude and latitude of the starting and ending points of each section, and cutting the overly long sections, and extracting the starting and ending points of the cut sections respectively.

4. The method for identifying regional transportation modes based on mobile phone signaling data according to claim 1 is characterized in that: The matching method of the communication base station and the road network based on the tree structure in step 2 is: (1) Use KD-Tree to find the k road segment endpoints closest to the base station: First, the road segments are divided according to the set length threshold, and then the irregular actual road is abstracted into a plane straight line containing only endpoint information, and all endpoints are used as samples to construct a KD-Tree; secondly, the base station location is used as the target node, and the nearest k points are queried through the KD-Tree; (2) The k road segments where the endpoints are found are traversed in turn to find the shortest distance, and then the minimum value of all the shortest distances is taken, that is, the shortest distance between the base station and the actual road is obtained, and the matching of the base station and the road network is completed.

5. The method for identifying regional transportation modes based on mobile phone signaling data according to claim 3 is characterized in that: The length threshold for dividing the road section is 3000m, and the value of k is 6.

6. The method for identifying regional transportation modes based on mobile phone signaling data according to claim 3 is characterized in that: A buffer zone is established including the intersection and the 500m to 1000m road section near it. All base station matching data within the buffer zone are considered invalid data and are not used for base station matching.

7. The method for identifying regional transportation modes based on mobile phone signaling data according to claim 1 is characterized in that: The method for identifying the user's travel path in step 3 is: (1) Initialize the score matrix: Create an (m+1)×(n+1) matrix, where m and n are the lengths of the two sequences involved in the alignment, and the first row and column of the matrix are initialized to 0; (2) Calculate the score matrix: Fill the matrix using a dynamic programming algorithm: if the two corresponding base station sequences are the same or similar, add a matching score to the current position of the matrix; if they are different, add 0; (3) Best path backtracking: Start from the path with the highest score and backtrack until the starting point of local alignment is found.

8. The method for identifying regional transportation modes based on mobile phone signaling data according to claim 7 is characterized in that: When adding the matching score, the normalized comparison score value is obtained by calculating twice the scores of the two sequences to be compared and dividing by the number of sequence elements corresponding to the two sequences, which is used to determine the final matching score.

9. The method for identifying regional transportation modes based on mobile phone signaling data according to claim 1, characterized in that: In step four, the method for identifying the transportation modes of cars and intercity buses based on the optimized gradient boosting algorithm is as follows: the maximum speed, average speed, percentile speed, and the ratio of high-speed points to low-speed points are selected as input features, and half of the signaling data matched for road travel is used as a test set and half as a training set. The above feature parameters are used as classification feature variables, and the hyperparameters are optimized and adjusted. It is proposed to select the learning rate, random sampling ratio, the proportion of randomly sampled columns, and the maximum tree depth as hyperparameters for optimization, so as to identify the transportation modes of cars and intercity buses.

10. The method for identifying regional transportation modes based on mobile phone signaling data according to claim 9, characterized in that: The percentile speed is the percentile speed of the travel speed sequence, including 95%, 75%, 50%, 25% and 5% percentile speeds.

Images