A method and system for identifying accompanying relationships of heterogeneous spatiotemporal trajectories

Abstract

The present invention discloses a method and system for identifying accompanying relationships of heterogeneous spatiotemporal trajectories, which are used to improve the recall rate of accompanying relationship identification and accurately identify the accompanying relationships between heterogeneous spatiotemporal trajectories. The method for identifying accompanying relationships of heterogeneous spatiotemporal trajectories of the present application includes acquiring heterogeneous data, creating a corresponding heterogeneous spatiotemporal trajectory table; determining accompanying tasks, performing spatiotemporal collision and similarity evaluation on the trajectories in the trajectory table corresponding to the accompanying tasks, and generating an accompanying result candidate set according to the evaluation results; and performing heterogeneous spatiotemporal trajectory accompanying relationship identification tasks in different application scenarios based on the accompanying result candidate set.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and more particularly to a method and system for identifying accompanying relationships of heterogeneous spatiotemporal trajectories. Background Art

[0002] In the context of contemporary mobile communication networks, signaling data has become a key source of information for recording user-network interactions, covering key operations such as connection establishment and disconnection, and real-time location updates. These data not only reveal user behavior patterns and mobility habits, but also reflect the communication quality of the network. However, due to the ubiquity of mobile device use and the diversity of people's activities, the volume of signaling data is huge and the structure is complex, which poses considerable challenges to the management and analysis of communication data.

[0003] In this regard, effective signaling data management, optimization and use have become important ways to improve network service quality, strengthen public safety and support urban management. Against this background, the concept of heterogeneous spatiotemporal trajectory accompanying relationship recognition has emerged. This technology aims to parse and use signaling data through advanced data analysis technology, extract trajectory information in combination with other heterogeneous data, and identify accompanying relationships through the extracted trajectory information.

[0004] However, heterogeneous spatiotemporal trajectories usually have sparsity in time and space, which makes the trajectory data lack observation points to varying degrees at certain time points or spatial regions; at the same time, the sparsity in time and space will affect the identification of accompanying relationships, that is, the sparsity of trajectories in time and space will cause insufficient data integrity, reduce the accuracy and credibility of accompanying relationships, and increase the difficulty of identifying accompanying relationships.

[0005] At present, the mining of adjoint relationships mainly depends on specific data and scenarios. Most existing methods use adjoint analysis based on longitude and latitude, that is, the geohash algorithm is used to process the data under the premise that the precise longitude and latitude of each base station are known. This method usually uses simple trajectory collision detection to determine the adjoint relationship. However, this method has certain limitations. For example, in a high-density urban environment, building occlusion, signal reflection, and multipath effects may cause errors in location information. In addition, in scenarios with sparse trajectories, there may be a lack of sufficient collision data, which in turn affects the accuracy of the adjoint relationship.

[0006] To address these challenges, this paper proposes a heterogeneous spatiotemporal trajectory adjoint method, aiming to solve the main problem of low recognition accuracy in related tasks. Summary of the invention

[0007] In view of this, the present invention provides a method and system for identifying accompanying relationships of heterogeneous spatiotemporal trajectories, aiming to improve the recognition accuracy and generate more accurate and reliable accompanying results based on massive heterogeneous data.

[0008] In order to achieve the above object, the present invention adopts the following technical solution:

[0009] On the one hand, the present application discloses a method for identifying heterogeneous spatiotemporal trajectory accompanying relationships, comprising:

[0010] Obtain heterogeneous data and create corresponding trajectory tables;

[0011] Determine the accompanying task, perform spatiotemporal collision and similarity evaluation on the trajectories in the trajectory table corresponding to the accompanying task, and generate a candidate set of accompanying results based on the spatiotemporal collision and similarity evaluation results;

[0012] Based on the accompanying result candidate set, heterogeneous spatiotemporal trajectory accompanying relationship recognition in different application scenarios is performed.

[0013] The recognition process is aimed at two scenarios. One is that in the case of unknown positive samples, that is, the accompanying relationship is unknown, a multi-way recall strategy is adopted to use a multi-level trajectory similarity algorithm to obtain an accompanying relationship archive based on daily calculation; the other is that on the basis of known positive samples or accompanying relationship archives calculated on a daily basis, the TrajCL method based on contrastive learning is used to further measure the trajectory similarity using the feature vectors of positive and negative sample pairs, so as to further improve the accuracy of accompanying relationship recognition of the corresponding task and obtain the final accompanying relationship archive.

[0014] Preferably, the multi-level heuristic trajectory similarity algorithm includes DTW and Hausdorff.

[0015] Preferably, the heuristic trajectory similarity algorithm also includes LCSS and EDR.

[0016] Preferably, the process of creating the corresponding trajectory table includes:

[0017] The heterogeneous data is preprocessed, including data analysis, data cleaning, data completion, data extraction and other operations, wherein the extraction process is to extract relevant information including a unique identifier (ID), longitude, latitude and timestamp to form a heterogeneous data trajectory table corresponding to the unique identifier (ID).

[0018] Preferably, the process of generating the candidate set of accompanying results includes: taking a union of the spatiotemporal collision result and the similarity evaluation result to obtain the candidate set of accompanying results.

[0019] Among them, a preferred implementation process is:

[0020] Determine a first trajectory table and a second trajectory table corresponding to the accompanying task;

[0021] Performing a spatial-temporal collision on the trajectories in the first trajectory table and the trajectories in the second trajectory table to obtain a first accompanying result candidate set;

[0022] Evaluate the similarity between the trajectory in the first trajectory table and the trajectory in the second trajectory table to obtain a similarity score, and obtain a second accompanying result candidate set according to the similarity score;

[0023] The first companion result candidate set and the second companion result candidate set are combined to obtain a companion result candidate set.

[0024] Preferably, the similarity evaluation method is:

[0025] Determine the straight-line distance on the earth's surface based on the longitude and latitude of the beginning and end of the two tracks.

[0026] Add the straight-line distance and the spherical distance between the first and last ends to obtain the total distance between the first and last trajectory points;

[0027] The reciprocal of the total distance between the first and last trajectory points is used as the similarity score of the two trajectories.

[0028] On the other hand, the present application discloses a system for identifying heterogeneous spatiotemporal trajectory accompanying relationships, including:

[0029] The trajectory table creation module is used to obtain heterogeneous data and create corresponding trajectory tables;

[0030] The initial candidate screening module is used to determine the accompanying tasks, perform spatiotemporal collision and similarity evaluation on the trajectories in the trajectory table corresponding to the accompanying tasks, and generate a candidate set of accompanying results based on the spatiotemporal collision and similarity evaluation results;

[0031] The accompanying recognition module performs accompanying relationship recognition of heterogeneous spatiotemporal trajectories based on the accompanying result candidate set.

[0032] It can be seen from the above technical solutions that the present invention discloses a method and system for identifying the accompanying relationship of heterogeneous spatiotemporal trajectories. Compared with the prior art, the present invention can accurately capture the complex relationship between trajectories in time and space, thereby effectively improving the recall rate and accurately identifying the accompanying relationship between heterogeneous spatiotemporal trajectories.

[0033] Specifically, the present invention can comprehensively utilize multi-source heterogeneous data such as mobile trajectory data, sensor data, geographic information data, and traffic flow data as data sources, perform fusion calculations on massive heterogeneous data, and obtain candidate sample sets. On this basis, by solving the problem of whether there are positive samples or not, the trajectory similarity calculation is used to determine and measure the accompanying relationship between trajectories, thus forming the final associated basic file.

[0034] In particular, this patent can solve the accompanying task of heterogeneous data sources with and without positive samples, and effectively combines the advantages of the heuristic accompanying algorithm and the learning accompanying algorithm, improves the recall rate of the accompanying task, and accurately judges the accompanying relationship; specifically, the comparative trajectory similarity learning method based on dual-feature attention is used to solve the accompanying relationship archives obtained in the presence of positive samples or based on daily calculations, and improve the accuracy of trajectory similarity measurement. This method is different from the common heuristic trajectory similarity algorithm. It pays more attention to the spatial and temporal characteristics of the trajectory. By using contrastive learning and dual-feature attention mechanisms, it effectively solves the problem that the number of trajectory points in the face code accompanying task is small and it is difficult to capture complex spatiotemporal relationships, and can obtain more reliable accompanying results. In addition, the present invention also proposes a multi-way recall method of a multi-level algorithm, that is, a strategy of combining the heuristic algorithm and the learning method for trajectory similarity calculation. This strategy solves the need to measure the similarity of the target trajectory in the absence of positive samples. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0036] Figure 1 This is a flow chart of the vehicle face code accompanying recognition method of the present invention;

[0037] Figure 2 This is a structural block diagram of the vehicle face code accompanying recognition system of the present invention. DETAILED DESCRIPTION

[0038] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0039] Embodiment 1

[0040] Reference Figure 1 The embodiment of the present invention discloses a method for identifying accompanying relationships of heterogeneous spatiotemporal trajectories, which can solve accompanying tasks within a day, including hourly level tasks, and includes the following steps:

[0041] Obtain heterogeneous data and create corresponding trajectory tables;

[0042] Determine the accompanying task, perform spatiotemporal collision and similarity evaluation on the trajectories in the trajectory table corresponding to the accompanying task, and generate a candidate set of accompanying results based on the spatiotemporal collision and similarity evaluation results;

[0043] The trajectory similarity is calculated based on the accompanying result candidate set, and the heterogeneous spatiotemporal trajectory accompanying relationship is identified according to the calculation result.

[0044] Specifically,

[0045] Step 1: This application first obtains a large amount of heterogeneous data. Heterogeneous data is spatiotemporal trajectory data with different characteristics and sources. The heterogeneous data includes mobile trajectory data, sensor data, geographic information data, traffic flow data and other related data, which are diverse and complex.

[0046] Therefore, as a preferred embodiment of the present invention, after obtaining heterogeneous data, data parsing, extraction, cleaning, and completion operations are performed first to ensure data quality and consistency. Data completion can solve the sparsity of trajectory time and space to a certain extent.

[0047] Furthermore, the key fields in the processed data are extracted and the corresponding trajectory table is created. The process includes:

[0048] Extract the unique identifier ID, longitude, latitude and timestamp from each original heterogeneous data to form a corresponding signaling trajectory table;

[0049] Step 2: Generate a candidate set of accompanying results based on the trajectory table to facilitate more accurate trajectory matching in the future. Specifically, it includes: determining the accompanying task, performing spatiotemporal collision and similarity evaluation on the trajectories in the trajectory table corresponding to the accompanying task, and generating a candidate set of accompanying results based on the spatiotemporal collision and similarity evaluation results.

[0050] A preferred implementation process is:

[0051] Determine the first trajectory table and the second trajectory table corresponding to the accompanying task; the accompanying task is a custom combination in the heterogeneous data. If there are three types of heterogeneous data, the accompanying task includes the accompanying task of heterogeneous data I and heterogeneous data II, the accompanying task of heterogeneous data I and heterogeneous data III, and the accompanying task of heterogeneous data II and heterogeneous data III;

[0052] Performing a spatial-temporal collision on the trajectories in the first trajectory table and the trajectories in the second trajectory table to obtain a first accompanying result candidate set;

[0053] Evaluate the similarity between the trajectory in the first trajectory table and the trajectory in the second trajectory table to obtain a similarity score, and obtain a second accompanying result candidate set according to the similarity score;

[0054] The first companion result candidate set and the second companion result candidate set are combined to obtain a companion result candidate set.

[0055] In this embodiment, the spatiotemporal collision means that the trajectory in the first trajectory table and the trajectory in the second trajectory table have intersections within a certain time and space threshold range. In this case, it is counted as one collision during calculation, and the total number of intersections of the trajectory points is the total number of collisions. For the corresponding accompanying task, the number of spatiotemporal collisions between the target trajectory and the trajectory data to be accompanied is calculated, and they are sorted in reverse order to obtain the first accompanying result candidate set;

[0056] For heterogeneous data trajectories of a specified date, matching and retrieval are performed based on trajectory accompanying coarse screening rules. That is, based on the specific accompanying task, reasonable time and space thresholds are set to perform spatiotemporal collisions between trajectories to obtain candidate data sets under spatiotemporal collisions; for example, in accompanying task A, the main purpose is to achieve the accompanying relationship between the two types of heterogeneous data by analyzing heterogeneous data trajectory I and heterogeneous data II. According to the set time and space thresholds, the two types of data are analyzed and spatiotemporally collided to determine which heterogeneous data I trajectories have an accompanying relationship with which heterogeneous data II trajectories in time and space, that is, the two types of data appear at the same location at the same time, so as to establish the first accompanying result candidate set;

[0057] Then, through similarity evaluation, the second companion result candidate set under this method is obtained; wherein the similarity evaluation method is:

[0058] Determine the straight-line distance on the earth's surface based on the longitude and latitude of the beginning and end of the two tracks.

[0059] Add the straight-line distance and the spherical distance between the first and last ends to obtain the total distance between the first and last trajectory points;

[0060] The reciprocal of the total distance between the first and last trajectory points is used as the similarity score of the two trajectories.

[0061] The closer the distance between two trajectories in time and space, the higher the degree of similarity is considered. This method is more effective in short-term accompanying tasks, and because the algorithm does not rely on positive samples and can quickly measure the similarity between two trajectories, it can be considered a heuristic algorithm.

[0062] Furthermore, in this embodiment, the spherical distance refers to the shortest distance between two points on the spherical surface, and the calculation steps include:

[0063] 1. Determine the longitude and latitude coordinates of two points: These coordinates are used to determine two points on the sphere.

[0064] 2. Convert longitude and latitude to radians: Most spherical distance formulas require converting longitude and latitude to radians, which can be done by multiplying the angle by π / 180.

[0065] 3. Use the Haversine formula to calculate the great circle distance: The general form of the Haversine formula is as follows:

[0066]

[0067]

[0068] dR·c

[0069] Where Δlat and Δlong are the differences between the latitude and longitude of the two points, lat1 and lat2 are the latitudes of the first and second points, and R is the radius of the Earth, usually 6371 km when using metric units.

[0070] 4. The calculated distance is the spherical distance between the two points.

[0071] Finally, the two types of candidate results are combined as a final candidate set. In this embodiment, the collision candidate set push threshold is set to 1000; the similarity evaluation threshold is set to 500, and the two types of candidate sets are combined as a final candidate set.

[0072] Step 3: Calculate the trajectory similarity based on the accompanying result candidate set, and identify the trajectory accompanying relationship according to the calculation result.

[0073] In this embodiment, the recognition process is oriented to two types of scenarios. One is that in the case of unknown positive samples, that is, the accompanying relationship is unknown, a multi-way recall strategy is adopted to use a multi-level trajectory similarity algorithm to obtain an accompanying relationship archive based on daily calculation; the other is that on the basis of known positive samples or accompanying relationship archives calculated on a daily basis, the TrajCL method based on contrastive learning is used to further measure the trajectory similarity using the feature vectors of positive and negative sample pairs, so as to further improve the accuracy of accompanying relationship recognition of the corresponding task and obtain the final accompanying relationship archive.

[0074] Specifically, when the positive samples are unknown, a multi-level recall heuristic trajectory similarity algorithm is used to determine the similarity between trajectories, and the number of accompanying results is funnel-distributed in each level of the algorithm, thereby obtaining an accompanying relationship archive based on daily calculations. In order to accurately measure the similarity of trajectories when there are no positive samples, the linear difference method is first used to fill the trajectories, and then the strategy of the multi-level heuristic similarity algorithm is adopted to enhance the robustness and accuracy of the similarity calculation in the scenario without positive samples.

[0075] Among them, the multi-level heuristic trajectory similarity algorithms include DTW and Hausdorff, and can also be extended to LCSS, EDR, etc.

[0076] DTW (Dynamic Time Warping) is a dynamic programming algorithm that calculates the similarity of two time series data (especially two time series data with different time lengths). Given two time series Q and C, their lengths are n and m; the DTW algorithm can calculate the distance between each time value of the time series Q and C. The specific steps are: according to the lengths n and m of the two time series Q and C, form an n×m matrix D, calculate the Euclidean distance between each time series element, and obtain the n×m elements of the matrix D, where the calculation formula of the matrix value is:

[0077]

[0078] Where: d ij is Q in the time series Q i and C in the time series C j The smaller the distance, the higher the similarity.

[0079] The Hausdorff distance can be understood as for each point in set A, find the point in set B that is closest to it and calculate the maximum value of these distances; at the same time, for each point in set B, find the point in set A that is closest to it and calculate the maximum value of these distances, and then take the larger value of the two maximum values ​​as the Hausdorff distance. The Hausdorff calculation formula is:

[0080] H(A,B)=max{supa∈Ainfb∈Bd(a,b),sup∈Binfa∈Ad(b,a)}

[0081] Where d(a,b) represents the point distance function between point a and point b.

[0082] The larger the Hausdorff distance, the greater the maximum distance between the two sets, that is, the higher the degree of dissimilarity between the sets. Therefore, the larger the Hausdorff distance, the lower the similarity between the sets, and the greater the difference in their shapes or distributions. That is, the increase in the Hausdorff distance means an increase in the degree of dissimilarity between the sets.

[0083] In the process of rearranging the accompanying results, different algorithms are used in sequence from top to bottom to obtain the candidate sets of TopN1, TopN2, ..., TopN, where N1>N2>...>N. In the specific accompanying task, the first-level algorithm uses DTW to obtain the TopN1 candidate accompanying numbers, and then the Hausdorff algorithm is used to calculate the accompanying results of TopN2 in the candidate set of TopN1.

[0084] In one embodiment, the first stage uses the DTW algorithm, and N2 is set to 500, and the second stage uses Hausdorff, and N3 is set to 10, thereby obtaining the final adjoint results as shown in the following table;

[0085]

[0086] It can be clearly seen from the data in the table that the method of the present application significantly improves the recall rate of the accompanying results.

[0087] When the positive samples are known or the accompanying relationship files calculated daily are obtained, the TrajCL algorithm is used to determine the similarity between trajectories;

[0088] TrajCL is a trajectory similarity measurement method based on contrastive learning. It can process sparse trajectory data and learn the spatial and structural patterns of trajectories at the same time. It adopts the trajectory modeling strategy of contrastive learning. As an efficient similarity measurement method, it can better cope with the characteristics and distribution of heterogeneous data, solve the limitations of trajectory sparsity, and obtain more accurate accompanying results.

[0089] First, it uses a trajectory enhancement method to generate diverse trajectory variants with different features for each training sample. Secondly, the internal DualSTB encoder can use the enhanced trajectory embedding to capture the spatial distance correlation between trajectories, that is, to encode the structure and spatial features of the trajectory into its learned embedding. In addition, it can measure the coarse-grained and fine-grained position information of the trajectory, and calculate the similarity of two trajectories as the L1 distance between the embeddings. TrajCL's learning strategy combines contrastive learning and dual-feature attention mechanisms. By maximizing the similarity of positive sample pairs and minimizing the difference of negative sample pairs, it learns the similarity between trajectories, improves the performance and generalization ability of the model, and the analysis effect of trajectory data, so that it can more accurately capture the similarity between trajectories.

[0090] In the case of a certain number of positive samples, TrajCL is used to train the trajectory embedding, and the L1 distance of the trajectory embedding is calculated to obtain the most similar accompanying result of the target number. In an exemplary embodiment, the relevant parameters of the TrajCL algorithm are set as follows, and the Adam optimizer is used for training, the maximum training cycle is set to 30, and when the loss does not improve, 5 iterations are set for early stopping. The initial learning rate is set to 0.001, the embedding latitude is set to 32, the number of encoder layers is set to 2, and the default trajectory enhancement method, i.e., trajectory truncation, is used. The trained trajectory vector is measured using the L1 distance to take into account the distance between each point in the trajectory and summarize their differences into a single metric value.

[0091]

[0092] Among them, v 1i and v 2i are the values ​​of trajectory vectors v1 and v2 in the i-th dimension, respectively, and n is the embedding dimension. By calculating the L1 distance, we can get the degree of difference between the two trajectory vectors. The smaller the L1 distance, the more similar the two trajectory vectors are; and the larger the L1 distance, the greater the difference between the two trajectory vectors.

[0093] In a preferred embodiment, as the number of accumulated positive samples increases, the parameters in the TrajCL algorithm are trained and updated using the accumulated positive samples, and a heuristic trajectory similarity algorithm is used for verification and optimization to achieve a technical closed loop, so that the parameters of the TrajCL algorithm can be dynamically variable and the feature vector can contain richer and more accurate information, thereby calculating a more reliable most similar result to meet different application scenarios.

[0094] The present invention is applicable to different data sources and can be applied to trajectory data of different scales to determine the accompanying relationships in various tasks. By combining multiple data sources, the present method can improve the accuracy and robustness of accompanying relationship mining. Therefore, the innovation of the present invention lies in that by comprehensively utilizing heterogeneous spatiotemporal information, the limitations of traditional methods are overcome and the accuracy and applicability of accompanying relationship identification are improved.

[0095] Embodiment 2

[0096] This embodiment discloses a heterogeneous spatiotemporal trajectory accompanying relationship recognition system. Figure 2 As shown, including:

[0097] The trajectory table creation module is used to obtain heterogeneous data and create corresponding trajectory tables;

[0098] The initial candidate screening module is used to determine the accompanying tasks, perform spatiotemporal collision and similarity evaluation on the trajectories in the trajectory table corresponding to the accompanying tasks, and generate a candidate set of accompanying results based on the spatiotemporal collision and similarity evaluation results;

[0099] The accompanying recognition module is used to calculate the trajectory similarity based on the accompanying result candidate set, and to identify the accompanying relationship of heterogeneous spatiotemporal trajectories according to the calculation results.

[0100] The system provided in the embodiment of the present invention has the same implementation principle and technical effects as those in the aforementioned method embodiment. For the sake of brief description, for matters not mentioned in the system embodiment, reference can be made to the corresponding contents in the aforementioned method embodiment, which will not be repeated here.

[0101] Embodiment 3

[0102] An embodiment of the present invention also provides a storage medium on which one or more programs readable by a computing device are stored. The one or more programs include instructions. When the instructions are executed by the computing device, the computing device executes a method for identifying heterogeneous spatiotemporal trajectory accompanying relationships in embodiment one.

[0103] In the embodiment of the present invention, the storage medium may be, for example, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination thereof. More specific examples of storage media (a non-exhaustive list) include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, and any suitable combination thereof.

[0104] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0105] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for identifying accompanying relationships of heterogeneous spatiotemporal trajectories, characterized in that: include: Obtain heterogeneous data and create corresponding heterogeneous spatiotemporal trajectory tables; Determine the accompanying tasks, perform spatiotemporal collision and similarity evaluation on the trajectories in the heterogeneous spatiotemporal trajectory table corresponding to the accompanying tasks, and generate a candidate set of accompanying results based on the spatiotemporal collision and similarity evaluation results; The similarity evaluation method is: Determine the straight-line distance on the earth's surface based on the longitude and latitude of the beginning and end of the two tracks. Add the straight-line distance and the spherical distance between the first and last ends to obtain the total distance between the first and last trajectory points; The reciprocal of the total distance between the first and last trajectory points is used as the similarity score of the two trajectories; Based on the accompanying result candidate set, heterogeneous spatiotemporal trajectory accompanying relationship is identified, including trajectory similarity calculation based on the accompanying result candidate set, and identifying the heterogeneous spatiotemporal trajectory accompanying relationship according to the calculation result; The trajectory similarity calculation includes: When the positive samples are unknown, the trajectory is filled by linear interpolation, and the trajectory similarity is determined by a multi-level heuristic similarity algorithm; When the positive samples are known, based on the trajectory similarity profile determined when the positive samples are unknown, the TrajCL method is used to determine the similarity between the trajectories using the feature vectors of the positive and negative sample pairs.

2. A method for identifying heterogeneous spatiotemporal trajectory accompanying relationships according to claim 1, characterized in that: The unique identifier, longitude, latitude and timestamp in the heterogeneous data are extracted to create a corresponding heterogeneous spatiotemporal trajectory table.

3. The method for identifying the accompanying relationship of heterogeneous spatiotemporal trajectories according to claim 1, characterized in that: The process of generating the candidate set of the accompanying result includes: taking the union of the spatiotemporal collision result and the similarity evaluation result to obtain the candidate set of the accompanying result.

4. The method for identifying the accompanying relationship of heterogeneous spatiotemporal trajectories according to claim 1, characterized in that: Multi-level heuristic trajectory similarity algorithms include DTW and Hausdorff.

5. A heterogeneous spatiotemporal trajectory accompanying relationship recognition system, characterized in that: A method for identifying accompanying relationships of heterogeneous spatiotemporal trajectories according to any one of claims 1 to 4 is applied, comprising: The trajectory table creation module is used to obtain heterogeneous data and create corresponding heterogeneous spatiotemporal trajectory tables; The initial candidate screening module is used to determine the accompanying tasks, perform spatiotemporal collision and similarity evaluation on the trajectories in the heterogeneous spatiotemporal trajectory table corresponding to the accompanying tasks, and generate a candidate set of accompanying results based on the spatiotemporal collision and similarity evaluation results; The accompanying recognition module is used to identify the accompanying relationship of heterogeneous spatiotemporal trajectories based on the accompanying result candidate set.

Images