Specific object accompanying analysis method based on large-scale trajectory data

Abstract

The embodiment of the invention belongs to a specific object accompanying analysis method based on large-scale trajectory data, and relates to the technical field of big data analysis, and the method comprises the steps: obtaining original trajectory data generated by an operator network; processing the original trajectory data to obtain target trajectory data; associating the latitude and longitude coordinates to the target trajectory data; adopting a space grid algorithm to simultaneously generate and associatively store a plurality of grid unit identifiers with different precision levels for each piece of target trajectory data; calculating a grid unit identifier set covering the circular space region by calling a space covering function; determining a potential accompanying object of the target object by using the grid unit identifier set; and calculating a time attenuation coefficient, a distance attenuation coefficient and a hit ratio, performing weighted summation to obtain a comprehensive adjoint score, and determining a specific adjoint object based on the comprehensive adjoint score. According to the method, the space-time accompanying group of the specific target object can be efficiently and accurately identified through track analysis of the specific target object.

Description

Technical Field

[0001] This invention belongs to the field of big data analysis technology, specifically relating to a method for specific object adjoint analysis based on large-scale trajectory data. Background Technology

[0002] With the rapid development of mobile internet, the widespread adoption of 5G networks, and the explosive growth of IoT devices, operators generate tens of billions of trajectory data points daily. In the field of trajectory data processing and analysis, traditional methods are gradually revealing their limitations. Early trajectory data processing mainly focused on simple data cleaning and basic statistical analysis, which is insufficient to meet the demands for complex trajectory pattern mining and real-time analysis.

[0003] Traditional trajectory analysis techniques suffer from incomplete data processing. Specifically, the data processing stage lacks a deep understanding of the characteristics of operator trajectory data and targeted processing, resulting in low data quality and consequently affecting the accuracy of subsequent analysis results. Summary of the Invention

[0004] The purpose of this invention is to provide a method for analyzing the trajectory data of hundreds of billions of telecom operators based on large-scale high-performance big data components. This method can efficiently and accurately identify the spatiotemporal accompanying groups of specific target objects through trajectory analysis.

[0005] This invention provides a method for specific object adjoint analysis based on large-scale trajectory data, including: Obtain raw trajectory data generated by the operator's network; the raw trajectory data includes user identifier, timestamp, and base station identifier; The original trajectory data is cleaned, drift-shifted, and compressed to obtain target trajectory data, with each target trajectory data corresponding to a trajectory point. Based on the base station identifier, the corresponding latitude and longitude coordinates are queried from the preset base station knowledge base, and the latitude and longitude coordinates are associated with the corresponding target trajectory data; Based on the associated latitude and longitude coordinates, a spatial grid algorithm is used to generate and associate multiple grid cell identifiers of different precision levels for each target trajectory data. Based on each target trajectory point of the target object, and taking the target trajectory point as the center point, a spatiotemporal analysis range is defined by combining a preset time window and spatial radius; by calling the spatial coverage function, a set of grid cell identifiers covering the circular spatial area is calculated based on the latitude and longitude of the center point and the spatial radius; using the set of grid cell identifiers, potential accompanying objects in which the trajectory point falls within the circular spatial area within the time window are identified; For each potential companion object, a time decay coefficient is calculated based on the time difference between the trajectory point of the potential companion object and the target trajectory point, and a distance decay coefficient is calculated based on the distance between the trajectory point of the potential companion object and the target trajectory point. The hit ratio of the number of trajectory points of the potential companion object falling within the spatiotemporal analysis range to the total number of trajectory points of the potential companion object is calculated within the time window. The time decay coefficient, distance decay coefficient, and hit ratio are weighted and summed to obtain a comprehensive companion score, and a specific companion object is determined based on the comprehensive companion score.

[0006] Optionally, the data drift processing procedure includes: Maintain a sliding time window, within which monitor the base station identifier sequence in continuous raw trajectory data records; When a sequence pattern of switching from the first base station to the second base station and then back to the first base station is detected within a preset time interval, the original trajectory data corresponding to the second base station is removed.

[0007] Optionally, the data compression method includes: The original trajectory data is grouped according to user identifier and base station identifier; for multiple original trajectory data of the same user under the same base station ID and whose timestamps fall within a preset time period, they are merged into one trajectory record; a representative timestamp is generated for the merged trajectory record, which is determined by calculating the average or median value of multiple timestamps of the merged original trajectory data. And / or, Calculate the hash value of multiple original trajectory data separately. The hash value is generated based on key fields in user identifier, timestamp and base station identifier. Merge multiple original trajectory data with the same hash value into one trajectory data record.

[0008] Optionally, the step of calculating the set of grid cell identifiers covering the circular spatial region by calling the spatial coverage function, based on the latitude and longitude of the center point and the spatial radius, includes: The spatial coverage function is invoked. The input parameters of the spatial coverage function include the longitude and latitude of the center point, the spatial radius, the preset minimum precision level, the preset maximum precision level, and the maximum number of elements to be generated. The following steps are then performed through the spatial coverage function: The location of the center point is determined by its longitude and latitude. The circular region is determined based on the center position and the input spatial radius; Generate grid cell identifiers starting from the minimum precision level; Determine whether the number of currently generated grid cell identifiers is less than the maximum number of generated identifiers and whether the current precision level is less than the maximum precision level; If so, increase the precision level, continue generating grid cell identifiers, and repeat the steps of determining whether the number of currently generated grid cell identifiers is less than the maximum number of generated identifiers and whether the current precision level is less than the maximum precision level. If not, return the currently generated set of grid cell identifiers, ensuring that the number of grid cell identifiers in the returned set does not exceed the maximum number of generated identifiers.

[0009] Optionally, calculating the distance attenuation coefficient based on the distance between the trajectory points of the potential accompanying object and the target trajectory point includes: Calculate the target distance between the trajectory points of the potential accompanying object and the target trajectory point, and calculate the ratio of the target distance to the spatial radius; wherein the target distance is less than the spatial radius; By subtracting the ratio from the potential associated trajectory point, the distance attenuation coefficient can be obtained. The calculation process for the target distance includes: The spherical distance calculation function is invoked, and the input parameters of the spherical distance calculation function include the longitude and latitude of the trajectory point of the potential accompanying object, as well as the longitude and latitude of the target trajectory point; The spherical distance calculation function is based on the Haversine algorithm and performs the following calculation process: The longitude and latitude of the trajectory points of the potential accompanying object, as well as the longitude and latitude of the target trajectory point, are converted from angles to radians. The spherical distance between the trajectory points of the potential accompanying object and the target trajectory point is calculated according to the Haversine formula, which takes into account the Earth's radius and the difference in latitude and longitude between the trajectory points of the potential accompanying object and the target trajectory point.

[0010] Optionally, the step of querying the corresponding latitude and longitude coordinates from a pre-set base station knowledge base based on the base station identifier, and associating the latitude and longitude coordinates with the corresponding target trajectory data, includes: A base station knowledge base containing the correspondence between base station identifiers and latitude and longitude coordinates is pre-built; An index is created in the base station knowledge base based on the base station identifier; By querying the base station knowledge base, the base station identifiers in the target trajectory data are mapped to the corresponding latitude and longitude coordinates.

[0011] Optionally, based on the associated latitude and longitude coordinates, a spatial grid algorithm is used to simultaneously generate and associate multiple grid cell identifiers of different precision levels for each target trajectory data, including: For each target trajectory data, the streaming processing engine calculates in real time the grid cell identifiers corresponding to the latitude and longitude coordinates at multiple different precision levels. The grid cell identifiers of each accuracy level obtained from the calculation are added as new fields and associated with the corresponding target trajectory data for storage.

[0012] Optionally, the method is implemented based on a massively parallel processing (MPP) architecture, wherein: The target trajectory data is stored in a distributed manner in a columnar storage format; The vectorized execution engine distributes computational tasks across multiple computing nodes for parallel execution. The raw trajectory data is captured and processed in real time using a stream processing engine.

[0013] Optionally, the real-time ingestion and processing of the raw trajectory data via a streaming engine includes: Real-time reading of the raw trajectory data of the incoming flow; The original trajectory data is processed in real time to obtain the target trajectory data; Grid cell identifiers for real-time calculation of target trajectory data; The target trajectory data is written to the distributed storage system in real time.

[0014] Optionally, the vectorized execution engine-based distribution of computational tasks across multiple computing nodes for parallel execution includes: Receive a companion analysis query request, the request including target trajectory points and search parameters; The spatial query task is decomposed into multiple sub-tasks based on data distribution characteristics; Each compute node performs the following operations in parallel: Retrieve trajectory data of potential accompanying objects based on grid cell identifier index; Perform multi-dimensional scoring calculations on the search results; The intermediate results of each computing node are summarized and combined into a final comprehensive accompanying score.

[0015] The specific object adjoint analysis method based on large-scale trajectory data provided in this invention solves the problem of inaccurate handling of abnormal base station handover situations in a short period of time by traditional methods through cleaning and drift processing of the original trajectory data, effectively improving the quality of trajectory data. Data compression ensures the continuity and rationality of trajectory information, providing a high-quality data foundation for subsequent analysis. Furthermore, the use of a multi-level spatial grid index instead of traditional GeoHash encoding effectively avoids boundary effect problems. Circular area coverage calculations do not require extensive rectangular approximation calculations, ensuring both calculation accuracy and significantly improving query efficiency. By pre-calculating and storing grid cell identifiers of multiple precision levels, complex real-time geometric calculations are transformed into efficient index matching queries, achieving sub-second response times for billions of data points, meeting the needs of application scenarios with extremely high real-time requirements.

[0016] Furthermore, an innovative multi-dimensional scoring model was constructed that integrates time decay coefficient, distance decay coefficient, and hit ratio. This model not only considers the closeness of a single spatiotemporal contact, but also effectively distinguishes between accidental encounters and continuous accompanying behaviors through the hit ratio, greatly reducing false alarms and false negatives. This makes the identification of accompanying relationships more scientific and accurate, providing reliable analytical conclusions for key applications such as public safety and risk analysis. Attached Figure Description

[0017] Figure 1 A flowchart illustrating a specific object adjoint analysis method based on large-scale trajectory data, provided in an embodiment of the present invention; Figure 2 A flowchart illustrating the steps performed by the spatial coverage function provided in this embodiment of the invention. Detailed Implementation

[0018] The present invention will be described in detail below through embodiments.

[0019] With the rapid development of mobile internet, the widespread adoption of 5G networks, and the explosive growth of IoT devices, operators generate tens of billions of trajectory data points daily. In the field of trajectory data processing and analysis, traditional methods are gradually revealing their limitations. Early trajectory data processing mainly focused on simple data cleaning and basic statistical analysis, which is insufficient to meet the demands for complex trajectory pattern mining and real-time analysis.

[0020] Traditional trajectory analysis techniques suffer from incomplete data processing. Specifically, the data processing stage lacks a deep understanding of the characteristics of operator trajectory data and targeted processing, resulting in low data quality and consequently affecting the accuracy of subsequent analysis results.

[0021] Furthermore, existing technologies suffer from limitations in storage and computing architecture and insufficient spatial indexing efficiency. Specifically, batch processing systems based on the Hadoop ecosystem cannot meet the demands of sub-second real-time analysis, and traditional databases experience a precipitous performance drop when performing petabyte-level data join queries. Moreover, traditional GeoHash encoding suffers from boundary effects in region coverage calculations, requiring extensive rectangular approximation calculations for circular region coverage.

[0022] To address the aforementioned technical problems in existing technologies, embodiments of the present invention provide a method for specific object adjoint analysis based on large-scale trajectory data, such as... Figure 1 As shown, the method includes the following steps: S110, acquires raw trajectory data generated by the operator's network.

[0023] The raw trajectory data includes user identifier, timestamp, and base station identifier.

[0024] In this embodiment, raw trajectory data is obtained from the operator's network. This data typically includes the following core fields: user identifier, which can be the International Mobile Subscriber Identity (IMSI) or the Mobile Subscriber International Integrated Services Digital Network Number (MSISDN); timestamp (accurate to milliseconds); and base station identifier (base station ID). The raw trajectory data is continuously generated in a streaming manner, with the daily data volume reaching tens of billions.

[0025] S120 performs data cleaning, data drift processing, and data compression on the original trajectory data to obtain the target trajectory data.

[0026] Each target trajectory data corresponds to a trajectory point.

[0027] Specifically, data cleaning can include the following three aspects: (i) For user identification, the system will strictly check the IMSI (International Mobile Subscriber Identity) and MSISDN (Mobile Subscriber International Integrated Services Digital Network Number) fields. If any field is found to be empty or the format does not meet the specifications (such as incorrect length, containing illegal characters, etc.), the trajectory data record will be determined as invalid and deleted.

[0028] (ii) Regarding timestamp checks: The system will compare the timestamp with the current system time and a reasonable time range. If the timestamp shows a future time, or is significantly earlier than a reasonable business start time (e.g., an extremely early time; a time threshold is set based on business experience, and times earlier than this threshold are considered extremely early times), the trajectory data record will be deleted.

[0029] Third, for key fields such as base station ID, if trajectory data records are found to be empty or do not conform to the base station ID encoding rules, they will also be deleted to ensure data integrity and availability.

[0030] As one implementation of this invention, the data drift processing procedure may include the following steps: Step a1: Maintain a sliding time window and monitor the base station identifier sequence in continuous raw trajectory data records within the sliding time window.

[0031] Step a2: When a sequence pattern of switching from the first base station to the second base station and then back to the first base station is detected within a preset time interval, the original trajectory data corresponding to the second base station is removed.

[0032] Specifically, the system reads continuous trajectory data records line by line. For each trajectory data record, it checks the base station ID in the preceding and following records. When it detects that the trajectory data moves from base station A to base station B and back to base station A within a short period of time (a time interval threshold is set according to the actual business scenario, such as within 2 minutes), the system automatically marks and removes the trajectory data record corresponding to base station B. This process is achieved by maintaining a sliding window, within which changes in base station IDs are monitored and analyzed in real time.

[0033] The data compression process can be summarized as follows: First, the system groups the trajectory data according to base station IDs. For multiple data entries under the same base station ID, its timestamp is checked. If multiple data entries have the same base station ID within a set time period (e.g., 2 minutes), the system merges these data entries. During the merging process, the time value is taken as the median of the data acquisition times, or the average of these acquisition times. For cases where multiple trajectory data entries are duplicated, the system uses a hash algorithm to perform hash calculations on the trajectory data. Data entries with the same hash value and the same base station ID are merged, retaining only one record, thus achieving efficient data compression.

[0034] S130: Based on the base station identifier of the target trajectory data, query the corresponding latitude and longitude coordinates from the pre-set base station knowledge base, and associate the latitude and longitude coordinates with the corresponding target trajectory data.

[0035] As one implementation of this invention, based on the base station identifier, querying the corresponding latitude and longitude coordinates from a pre-set base station knowledge base and associating the latitude and longitude coordinates with the corresponding target trajectory data may include the following steps: A base station knowledge base containing the correspondence between base station identifiers and latitude and longitude coordinates is pre-built.

[0036] An index is created in the base station knowledge base based on the base station identifier.

[0037] By querying the base station knowledge base, the base station identifiers in the target trajectory data are mapped to the corresponding latitude and longitude coordinates.

[0038] Specifically, a base station knowledge base containing a large amount of base station information is pre-built, where each base station ID corresponds to precise latitude and longitude information. When a base station ID in the trajectory data is matched with a record in the base station knowledge base, the system uses an efficient index lookup algorithm to quickly locate the corresponding base station record, extract its latitude and longitude information, and associate it with the target trajectory data. This process accelerates the lookup process by building an index based on the base station ID in the base station knowledge base, ensuring the efficiency and accuracy of latitude and longitude association.

[0039] S140, based on the associated latitude and longitude coordinates, uses a spatial grid algorithm to simultaneously generate and associate multiple grid cell identifiers of different precision levels for each target trajectory data.

[0040] As one implementation of this invention, based on the associated latitude and longitude coordinates, a spatial grid algorithm is used to simultaneously generate and associate multiple grid cell identifiers of different precision levels for each target trajectory data, which may include the following steps: For each target trajectory data, the streaming engine calculates in real time the identifiers of multiple grid cells with different precision levels corresponding to the latitude and longitude coordinates.

[0041] The grid cell identifiers of each accuracy level obtained from the calculation are added as new fields and associated with the corresponding target trajectory data for storage.

[0042] For streaming trajectory data, core fields include object value, object type, latitude and longitude, and collection time. During the governance process, the powerful real-time streaming capabilities of the Flink streaming engine are leveraged to perform real-time analysis and transformation of this data. Flink reads the latitude and longitude information from the trajectory data in real time and uses specialized algorithms to calculate the cellId (grid cell identifier) ​​for five levels (15 to 19) of Google S2.

[0043] The specific calculation process is as follows: Based on the principle of the S2 algorithm, latitude and longitude coordinates are mapped to the S2 space, and corresponding cellIds are generated according to different precision levels. The generated cellIds are written to the cell_id15 to cell_id19 fields respectively, so that subsequent calculations and analyses can be performed under different precision requirements. This process is implemented through Flink's real-time data processing pipeline to ensure the real-time performance and accuracy of the data.

[0044] S150 defines the spatiotemporal analysis range based on each target trajectory point of the target object, using the target trajectory point as the center point, and combining a preset time window and spatial radius. By calling the spatial coverage function, a set of grid cell identifiers covering the circular spatial region is calculated based on the latitude and longitude of the center point and the spatial radius. Using the set of grid cell identifiers, potential accompanying objects within the circular spatial region where the trajectory point falls within the time window are identified.

[0045] It should be noted that, in this invention, the target object refers to the individual designated as the focus of analysis in a specific accompanying analysis task. This individual is characterized by their unique user identifier within the system. The analysis process aims to uncover potential social or behavioral connections based on the target object's historical trajectory data.

[0046] Potential co-occurring objects refer to other individuals identified by the method described in this invention as exhibiting significant co-occurrence patterns with the target object within a specific spatiotemporal analysis range. This determination is not based on simple, one-off spatiotemporal proximity, but rather on a comprehensive consideration of multiple dimensions, including time decay coefficient, distance decay coefficient, and hit ratio, quantified through a comprehensive co-occurrence scoring model. Only individuals whose scores exceed a preset threshold are identified as potential co-occurring objects with analytical value.

[0047] In other words, the target object is the input and center of the analysis, while the potential accompanying object is the analysis result output after a series of complex technical processes.

[0048] Specifically, for each trajectory point of the target object, perform the following operations: (1) Definition of spatiotemporal analysis range. With the target trajectory point as the center, set the time window (e.g., 30 minutes) and spatial radius (e.g., 500 meters) to define the cylindrical spatiotemporal analysis range.

[0049] (2) Spatial Coverage Calculation. Call the custom function s2_circle_coverage, with input parameters including the latitude and longitude of the center point, radius (e.g., 500 meters), minimum precision (e.g., 15), maximum precision (e.g., 18), and maximum number of generated values ​​(e.g., 1000). This function generates a set of all CellIds (grid cell identifiers) covering the circular area based on the S2 algorithm.

[0050] (3) Trajectory Data Retrieval. The generated CellId set is used for efficient querying in the distributed database. Specifically, the set of grid cell identifiers is used to retrieve trajectory data where trajectory points fall within a circular spatial region within the time window; potential accompanying objects are identified based on the user identifiers corresponding to the trajectory data.

[0051] This step can quickly retrieve all potential accompanying object trajectory points within the spatiotemporal analysis range.

[0052] S160, for each potential accompanying object, calculate the time decay coefficient based on the time difference between the trajectory point of the potential accompanying object and the target trajectory point, calculate the distance decay coefficient based on the distance between the trajectory point of the potential accompanying object and the target trajectory point, and calculate the hit ratio of the number of potential accompanying object trajectory points falling within the spatiotemporal analysis range to the total number of potential accompanying object trajectory points within the time window.

[0053] Specifically, for each potential companion, three scoring elements are calculated: 1. Time decay coefficient (T). The closer the time difference is to 0, the higher the weight; it decays linearly over 5 minutes. The calculation formula is T = 1 - |Δt| / 300 (|Δt| ≤ 300 seconds), where Δt is the time difference between the trajectory points of the target object and the potential accompanying object. The smaller the time difference, the higher the coefficient.

[0054] 2. Distance attenuation coefficient (D). The distance is approximately zero, with a higher weighting. It attenuates linearly within 500 meters. The spherical distance d is calculated using the user-defined function `distance_haversine`, and then calculated using the formula D = 1 - d / 500 (d ≤ 500 meters). The closer the distance, the higher the coefficient.

[0055] 3. Hit Ratio (H): This is the percentage of hit points within the spatiotemporal window relative to the total number of hit points within the time window. It calculates the percentage of potential accompanying object trajectory points that hit the spatiotemporal analysis range within the analysis time window, relative to the total number of trajectory points. H = Number of hit points / Total number of points.

[0056] S170, the time decay coefficient, distance decay coefficient and hit ratio are weighted and summed to obtain the comprehensive companion score, and the specific companion object is determined based on the comprehensive companion score.

[0057] The three scoring elements mentioned above are weighted and summed according to preset weights: Score=(T_avg×W t +D_avg×W d +H×W h )×100 Where T_avg and D_avg are the average of the time decay coefficient and distance decay coefficient of all matching points, respectively, and the time weight W t It can be 30%, W d It can be 0.3, W h The weights can be 0.4. It's understood that these three weights can be determined based on the actual situation; this embodiment of the invention does not specifically limit the magnitude of the three weights. A scoring threshold (e.g., 60 points) is set, and potential companion objects with a comprehensive score higher than the threshold are identified as the final specific companion objects.

[0058] The specific object adjoint analysis method based on large-scale trajectory data provided in this invention solves the problem of inaccurate handling of abnormal base station handover situations in a short period of time by traditional methods through cleaning and drift processing of the original trajectory data, effectively improving the quality of trajectory data. Data compression ensures the continuity and rationality of trajectory information, providing a high-quality data foundation for subsequent analysis. Furthermore, the use of a multi-level spatial grid index instead of traditional GeoHash encoding effectively avoids boundary effect problems. Circular area coverage calculations do not require extensive rectangular approximation calculations, ensuring both calculation accuracy and significantly improving query efficiency. By pre-calculating and storing grid cell identifiers of multiple precision levels, complex real-time geometric calculations are transformed into efficient index matching queries, achieving sub-second response times for billions of data points, meeting the needs of application scenarios with extremely high real-time requirements.

[0059] Furthermore, an innovative multi-dimensional scoring model was constructed that integrates time decay coefficient, distance decay coefficient, and hit ratio. This model not only considers the closeness of a single spatiotemporal contact, but also effectively distinguishes between accidental encounters and continuous accompanying behaviors through the hit ratio, greatly reducing false alarms and false negatives. This makes the identification of accompanying relationships more scientific and accurate, providing reliable analytical conclusions for key applications such as public safety and risk analysis.

[0060] As one implementation of this invention, data compression methods may include the following two: The first method involves grouping the original trajectory data according to user identifier and base station identifier; for multiple original trajectory data of the same user under the same base station ID and whose timestamps fall within a preset time period, they are merged into one trajectory record; a representative timestamp is generated for the merged trajectory record, which is determined by calculating the average or median value of multiple timestamps of the merged original trajectory data.

[0061] Specifically, this method primarily handles continuous redundant data caused by frequent signal reporting. The specific implementation process includes: 1. Data Grouping. First, the raw trajectory data is grouped according to user identifier and base station identifier, forming data subsets with user and base station as the dimensions.

[0062] 2. Time window judgment. For multiple data records of the same user under the coverage of the same base station, check whether their timestamps fall within a preset time period (e.g., 2 minutes). This time threshold can be adjusted according to specific business scenarios.

[0063] 3. Record merging. Multiple trajectory data that meet the time window conditions are merged into a single representative trajectory record.

[0064] 4. Timestamp Generation. A representative timestamp is generated for the merged new record. This timestamp is determined by calculating the average or median of the timestamps of the merged records. This method preserves the temporal characteristics of the original data while avoiding the bias caused by simply taking the first and last timestamps.

[0065] The second method involves calculating the hash values ​​of multiple original trajectory data separately. The hash values ​​are generated based on key fields in the user identifier, timestamp, and base station identifier. Multiple original trajectory data with the same hash value are then merged into a single trajectory data record.

[0066] This method primarily addresses the problem of completely duplicated records generated during data transmission and storage, specifically including: 1. Hash value calculation. Select key fields from the trajectory data (including user identifier, timestamp, and base station identifier) ​​as hash input, and use a hash algorithm to generate a unique hash value.

[0067] 2. Duplicate detection. Multiple trajectory data records with identical content are identified by comparing hash values.

[0068] 3. Record merging. Merge multiple records with the same hash value into one, achieving precise deduplication.

[0069] In practice, the two compression methods can be executed sequentially: first, the second method is used for precise deduplication to eliminate completely duplicate records; then, the first method is applied to intelligently merge continuous data. This combined strategy achieves comprehensive coverage of trajectory data redundancy.

[0070] The data compression method used in this implementation method brings significant technological advancements and practical benefits: 1. Significant improvement in data quality.

[0071] The intelligent merging algorithm based on temporal proximity fully considers the temporal continuity and spatial stability of trajectory data, avoiding the loss of effective information caused by simple deduplication. By generating representative timestamps, it compresses the data volume while preserving the spatiotemporal characteristics of the trajectory, providing a higher-quality data foundation for subsequent analysis.

[0072] 2. Storage efficiency has been significantly optimized.

[0073] The combined application of the two compression methods reduces the amount of trajectory data by 60%-80%, effectively lowering storage costs. Tests have shown that with data volumes in the tens of billions, it can save petabytes of storage space while simultaneously reducing I / O overhead.

[0074] 3. Processing performance is significantly improved.

[0075] The reduced data volume after compression directly lowers the computational load of subsequent processing steps, improving the execution efficiency of computationally intensive tasks such as spatial index construction and adjoint analysis by approximately 40%. The intelligent merging algorithm reduces the number of trajectory points that need to be processed, lowers the algorithm complexity, and shortens the overall processing time by more than 35%.

[0076] 4. Enhanced algorithm adaptability.

[0077] The dual compression mechanism employs differentiated processing strategies for different types of redundant data, enhancing the algorithm's adaptability and robustness in practical applications. The configurable time window parameter allows the method to flexibly adapt to trajectory data compression requirements of varying densities and business scenarios.

[0078] 5. Outstanding business value.

[0079] Achieving efficient compression while ensuring data quality enables long-term storage and real-time analysis of large-scale trajectory data, providing technical support for historical data retrospective analysis and real-time monitoring and early warning. The key spatiotemporal information preserved during compression ensures the accuracy and reliability of subsequent advanced applications such as adjoint analysis and trajectory mining.

[0080] In summary, the data compression method provided by this invention achieves efficient compression while ensuring data quality through innovative technical means, providing strong technical support for the storage, processing and analysis of large-scale trajectory data, and has significant practical value and promising prospects for widespread application.

[0081] exist Figure 1 Based on the illustrated embodiment, as one implementation of this invention, by calling a spatial coverage function and calculating the set of grid cell identifiers covering a circular spatial region according to the latitude and longitude of the center point and the spatial radius, the following steps may be included: The spatial coverage function is invoked. Its input parameters include the longitude and latitude of the center point, the spatial radius, the preset minimum precision level, the preset maximum precision level, and the maximum number of grid cells generated. In practical applications, the minimum precision level can be 15 (grid area approximately 0.74 km²). 2 Maximum accuracy level: 18 (grid area approximately 0.09 km²) 2 Maximum number of cells generated: 1000. Spatial radius: 500 meters.

[0082] like Figure 2 As shown, the following steps are performed through the spatial coverage function: S210, the center position is determined based on the longitude and latitude of the center point.

[0083] Specifically, the spatial coverage function first constructs a precise center position of the circle based on the input longitude and latitude values. In the implementation, the longitude and latitude coordinates are converted into point objects in S2 space using the S2LatLng.fromDegrees(lat, lng) method of the S2 library, which serves as the center point of the circular region.

[0084] S220 determines the circular region based on the center position and the input spatial radius.

[0085] Based on the determined center location and the input spatial radius (in meters), a circular query region is constructed. Internally, the function converts the radius value to a spherical distance, taking into account the Earth's curvature to ensure the accuracy of the circular region definition. Specifically, the circular region is represented using the S2Cap.fromAxisHeight() method of S2 or the S2CellUnion covering algorithm.

[0086] S230 generates grid cell identifiers starting from the lowest precision level.

[0087] The S2 algorithm generates grid cell identifiers starting from a preset minimum precision level (e.g., level 15). It divides the Earth's surface into 31 grid levels, with higher levels resulting in smaller grid areas and higher precision. A lower precision level is used initially to ensure rapid generation of the basic cover grid.

[0088] S240: Determine if the number of currently generated mesh cell identifiers is less than the maximum number and the current precision level is less than the maximum precision level. If yes, proceed to S250. If no, proceed to S260.

[0089] S250, increase the precision level, continue to generate grid cell identifiers, and repeat the steps of determining whether the number of currently generated grid cell identifiers is less than the maximum number of generated identifiers and whether the current precision level is less than the maximum precision level.

[0090] S240 and S250 are the core innovations of this implementation method, achieving coverage optimization through dynamic precision adjustment: the system continuously monitors the number of generated grid cell identifiers and compares them with the maximum number of generated ones; if the current number does not exceed the limit and the precision level has not reached the maximum value, the precision level is automatically increased (e.g., from level 15 to level 16) to continue generating finer grid cells at a higher level, enriching the details of regional coverage.

[0091] This process is repeated until the preset termination condition is reached.

[0092] S260 returns the set of currently generated grid cell identifiers, ensuring that the number of grid cell identifiers in the returned set does not exceed the maximum number of generated identifiers.

[0093] When the termination condition is met (reaching the maximum number of generated units or the highest precision level), the spatial coverage function optimizes the generated set of grid cell identifiers to ensure that the returned set of grid cell identifiers meets both the quantity limit and covers the target circular area to the greatest extent.

[0094] The spatial coverage function implementation provided in this method has the following significant advantages compared to traditional spatial query methods: 1. Significantly improved query efficiency.

[0095] By pre-generating and indexing grid cell identifiers, complex geometric spatial calculations are transformed into efficient identifier matching queries, reducing spatial query response time from seconds to milliseconds in traditional methods. The intelligent precision iteration mechanism avoids unnecessary over-refinement calculations, reducing computational complexity by approximately 40% while ensuring coverage quality.

[0096] 2. Highly optimized resource utilization.

[0097] The maximum number of cells generated effectively controls memory and computing resource consumption, preventing system resource exhaustion caused by generating too many grid cells; and the progressive accuracy improvement strategy ensures that usable coverage results can still be obtained in resource-constrained environments, significantly enhancing system robustness.

[0098] 3. Intelligent balancing of coverage quality.

[0099] By setting the minimum / maximum precision level, an intelligent balance between coverage accuracy and computational overhead is achieved, which users can flexibly adjust according to specific scenario requirements. Compared with the traditional fixed precision method, the area coverage integrity is improved by about 25-35% under the same grid number limit.

[0100] 4. Extension of algorithm applicability.

[0101] It supports efficient coverage of spatial ranges of different scales, applicable from tens of meters to tens of kilometers in radius; parametric design enables the algorithm to adapt to diverse application scenarios, including detailed analysis of dense urban areas and overview analysis of large areas.

[0102] 5. Enhanced practicality of the project.

[0103] The function interface is designed to be concise and clear, with six parameters covering the main application requirements, reducing the complexity of use; it integrates seamlessly with mainstream big data platforms (such as Spark and Flink) and supports efficient execution in distributed environments.

[0104] 6. A perfect combination of accuracy and performance.

[0105] Actual testing showed that, in a scenario with billions of trajectory data points, the method of this invention improved query accuracy by approximately 30% compared to the traditional GeoHash method in queries within a 500-meter radius, while reducing query time by approximately 60%, truly achieving the technical goal of being both fast and accurate.

[0106] In summary, the spatial coverage function of this invention, through its innovative intelligent precision control mechanism, has achieved significant technological advancements in spatial query efficiency, resource utilization, and result quality, providing a reliable technical foundation for real-time spatial analysis of large-scale trajectory data.

[0107] Based on the above embodiments, as one implementation of the present invention, the distance attenuation coefficient is calculated based on the distance between the trajectory point of the potential accompanying object and the target trajectory point, including: Calculate the target distance between the trajectory points of the potential accompanying object and the target trajectory point, and calculate the ratio of the target distance to the spatial radius; where the target distance is less than the spatial radius.

[0108] By subtracting the ratio, the distance decay coefficient of the potential associated trajectory point can be obtained.

[0109] The calculation process for the target distance includes: Call the spherical distance calculation function. The input parameters of the spherical distance calculation function include the longitude and latitude of the trajectory points of the potential accompanying object, as well as the longitude and latitude of the target trajectory point.

[0110] The spherical distance calculation function is based on the Haversine algorithm and performs the following calculation process: The longitude and latitude of the trajectory points of potential accompanying objects, as well as the longitude and latitude of the target trajectory points, are converted from angles to radians.

[0111] The spherical distance between the trajectory points of the potential accompanying object and the target trajectory point is calculated using the Haversine formula, which takes into account the Earth's radius as well as the latitude and longitude differences between the trajectory points of the potential accompanying object and the target trajectory point.

[0112] In this implementation, the distance attenuation coefficient is calculated using a scientific quantitative method. Through precise calculation of spherical distance and a linear attenuation model, an accurate assessment of spatial proximity is achieved.

[0113] Distance attenuation coefficient calculation process: Step 1: Calculate the spherical distance.

[0114] Accurate spherical distance calculation is achieved through a custom function, distance_haversine. The specific process is as follows: Coordinate transformation: Converts the input latitude and longitude coordinates from degrees to radians. The conversion formula is: Radius = angle × π / 180.

[0115] This step is a fundamental preparation for trigonometric function calculations, ensuring the accuracy of the results.

[0116] Based on the transformed radian coordinates, the Haversine formula is applied to calculate the spherical distance: the algorithm takes into account the actual shape of the Earth's ellipsoid and avoids the errors in planar distance calculation in large-scale spatial analysis through accurate spherical trigonometric calculation.

[0117] Step 2: Calculate the distance attenuation coefficient.

[0118] First, distance standardization is performed. The calculated actual distance d is compared with the preset spatial radius r: ratio = d / r; this step converts the absolute distance into a relative distance, achieving a unified measurement across different spatial scales.

[0119] Next, the distance attenuation coefficient is calculated using a linear attenuation function: D=1-d / r When d≤r, the range of D is [0,1]; when d>r, the point is not included in the score calculation.

[0120] And set reasonable boundary conditions: when the distance is 0, the attenuation coefficient is 1 (perfect match), and when the distance is equal to the spatial radius, the attenuation coefficient is 0 (critical state).

[0121] The distance attenuation coefficient calculation method provided in this implementation has the following significant advantages compared to traditional distance assessment methods: 1. Significantly improved calculation accuracy.

[0122] Employing the Haversine spherical distance algorithm, which accurately considers the Earth's curvature, it improves accuracy by approximately 30-40% in large-scale spatial analysis compared to traditional Euclidean distance calculations. It is particularly suitable for the characteristics of operator trajectory data, effectively solving the accuracy problem of distance calculation within base station coverage areas.

[0123] 2. Scientific quantification of spatial relationships.

[0124] The linear decay model intuitively reflects the spatial relationship essence of "the closer the distance, the higher the correlation", which is in line with the cognitive law of spatial proximity. By calculating the relative distance ratio, the distance assessment at different spatial scales is made comparable, which enhances the adaptability of the algorithm.

[0125] 3. Optimization and balance of computational efficiency.

[0126] The Haversine algorithm maintains accuracy while having relatively low computational complexity, with a single calculation completed in microseconds. Combined with a spatial radius filtering mechanism, it avoids unnecessary distance calculations, significantly improving overall processing efficiency in environments with billions of data points.

[0127] 4. Flexible and configurable model parameters.

[0128] The spatial radius parameter can be flexibly adjusted according to specific application scenarios. For example, a smaller radius (200-500 meters) can be set in the field of public safety, while a larger radius (1-5 kilometers) can be set in the field of regional pedestrian flow analysis. The linear decay model is simple and effective, and does not require complex parameter tuning, which reduces the difficulty of implementation.

[0129] 5. Perfect integration with the overall scoring system.

[0130] The distance decay coefficient, time decay coefficient, and hit ratio together constitute a multi-dimensional scoring system with reasonable weight allocation (e.g., 30%). The linear decay characteristic makes the scoring results highly interpretable, making it easy for business personnel to understand and apply.

[0131] In summary, the distance attenuation coefficient calculation method of this invention, through accurate spherical distance calculation and a scientific attenuation model, has achieved significant technical progress in terms of calculation accuracy, efficiency, and applicability, providing reliable technical support for spatial relationship analysis of large-scale trajectory data. This solution not only has theoretical innovation value but also demonstrates outstanding practical effects in real-world applications.

[0132] Based on the above embodiments, as one implementation of the present invention, this method is implemented based on a massively parallel processing (MPP) architecture, wherein... The target trajectory data is stored in a distributed manner in a columnar storage format.

[0133] The vectorized execution engine distributes computational tasks across multiple computing nodes for parallel execution.

[0134] Raw trajectory data is captured and processed in real time using a stream processing engine.

[0135] This invention uses an MPP database as the core data processing platform. In this architecture, each node has an independent processor, memory, and storage system, and the nodes are interconnected via a high-speed network. Specific implementation includes: Columnar storage is implemented. Specifically, the storage format adopts an optimized columnar storage layout, storing similar fields of trajectory data (such as user ID, timestamp, latitude and longitude, grid cell identifier, etc.) in different data blocks; each column of data is compressed using data compression algorithms (such as dictionary encoding, trip length encoding) to reduce storage space usage; the data distribution strategy adopts a sharding method based on user identifier hashing to ensure that relevant data is evenly distributed in the cluster and that data locality optimization can be achieved during queries.

[0136] Vectorized execution engine. Specifically, during query execution, a vectorized processing mode is adopted, processing a batch of data records at a time instead of a single record; parallel data computation is achieved by utilizing the SIMD instruction set of modern CPUs (such as AVX2, AVX-512); the execution engine optimizes the query plan, transforming complex adjoint analysis queries into efficient physical execution plans.

[0137] It adopts a stream processing engine architecture, which supports both real-time stream processing and batch historical data analysis; and enables seamless switching between different data processing modes through a unified API interface.

[0138] As one implementation of this invention, the real-time ingestion and processing of raw trajectory data via a stream processing engine may include the following steps: Read the raw trajectory data of the incoming flow in real time.

[0139] Real-time data processing is performed on the raw trajectory data to obtain the target trajectory data.

[0140] Grid cell identifiers for real-time calculation of target trajectory data.

[0141] The target trajectory data is written to the distributed storage system in real time.

[0142] In this implementation, a high-throughput streaming data ingestion pipeline is built based on Apache Flink, supporting real-time processing of millions of trajectory data per second; Kafka can be used as a message middleware to achieve data buffering and backpressure control; data ingestion latency is controlled at the millisecond level to ensure real-time requirements.

[0143] Real-time data processing includes: The system parses the input trajectory data stream in real time, performs format verification and basic cleaning; maintains a sliding time window (default 2 minutes) to detect abnormal base station switching modes in real time; merges consecutive trajectory points of the same user within a short period of time; calculates the multi-level S2 grid cell identifier (levels 15-19) for each trajectory point in parallel, and persists the processed target trajectory data to the distributed storage system using a batch asynchronous writing method.

[0144] As one implementation of this invention, distributing computational tasks across multiple computing nodes for parallel execution based on a vectorized execution engine may include the following steps: Receive a companion analysis query request, which includes the target trajectory points and search parameters.

[0145] The spatial query task is decomposed into multiple sub-tasks based on the data distribution characteristics.

[0146] Each compute node performs the following operations in parallel: Trajectory data of potential accompanying objects are retrieved based on grid cell identifier index.

[0147] Perform multi-dimensional scoring calculations on the search results.

[0148] The intermediate results of each computing node are summarized and combined into a final comprehensive accompanying score.

[0149] Specifically, based on data sharding information, the global query is decomposed into multiple data-local subtasks; a dynamic task scheduling algorithm is adopted to intelligently allocate computing resources according to the cluster load; priority scheduling of query tasks is supported to ensure a fast response for high-priority queries.

[0150] The distributed spatial query execution process includes: It receives query requests in SQL or API form and generates an optimized execution plan; calls the s2_circle_coverage function to generate a set of grid identifiers for the target area; each computing node scans the locally stored grid index data in parallel; performs preliminary data filtering and statistical aggregation locally on the node; and summarizes the intermediate results from each node through distributed aggregation operations to generate the final comprehensive accompanying score.

[0151] The implementation of this invention based on the MPP architecture has the following significant advantages compared to traditional data processing architectures: Columnar storage combined with vectorized execution improves the performance of complex queries by 5-10 times, reducing the response time of multi-billion data association queries from minutes to seconds; the distributed parallel processing architecture supports linear scaling, and performance can be maintained by adding nodes when the data volume grows; and it reduces data movement overhead, improving overall data processing efficiency by about 60%.

[0152] Furthermore, columnar storage reduces I / O overhead, improves storage space utilization by approximately 70%, and can support larger data scales with the same hardware configuration; vectorized execution improves CPU utilization, increasing computing resource utilization efficiency by more than 40%; intelligent task scheduling avoids resource idle waste, and the overall cluster resource utilization rate reaches more than 85%.

[0153] In addition, end-to-end processing latency is controlled within seconds, meeting the business needs of real-time monitoring and early warning; it supports high-concurrency query processing and can still maintain stable performance in scenarios with 100 concurrent users; streaming processing ensures the timeliness of data analysis and provides technical support for rapid decision-making.

[0154] In summary, the implementation scheme based on the MPP architecture of this invention has achieved breakthroughs in performance, efficiency, and reliability through innovative system architecture design and technical optimization, providing a solid technical foundation for large-scale trajectory data analysis and possessing significant value for widespread application.

[0155] In summary, this invention proposes a method for analyzing the accompanying patterns of billions of operator trajectory data points based on large-scale, high-performance big data components. This method can efficiently and accurately identify the spatiotemporal accompanying groups of specific target objects through trajectory analysis. It has broad application prospects, mainly in the following aspects: In the field of public safety, it can be applied to the analysis of key personnel relationship networks. Through in-depth mining of massive trajectory data, it can quickly identify important related personnel with close spatiotemporal connections, significantly improving the efficiency of analysis and judgment; in the field of business analysis, it can provide strong spatiotemporal data support for consumer behavior research and regional population flow analysis, assisting in business decision-making and resource optimization; in the fields of urban planning and traffic management, it can analyze population movement patterns and aggregation characteristics, providing data insights for urban facility layout and traffic management. This invention, through an innovative multi-dimensional scoring model and efficient spatial computing technology, achieves real-time and accurate analysis of ultra-large-scale trajectory data, providing a powerful tool for spatiotemporal relationship mining for various industries.

Claims

1. A method for specific object adjoint analysis based on large-scale trajectory data, characterized in that, include: Acquire raw trajectory data generated by the operator's network; The raw trajectory data includes user identifier, timestamp, and base station identifier; The original trajectory data is cleaned, drift-shifted, and compressed to obtain target trajectory data, with each target trajectory data corresponding to a trajectory point. Based on the base station identifier, the corresponding latitude and longitude coordinates are queried from the preset base station knowledge base, and the latitude and longitude coordinates are associated with the corresponding target trajectory data; Based on the associated latitude and longitude coordinates, a spatial grid algorithm is used to generate and associate multiple grid cell identifiers of different precision levels for each target trajectory data. Based on each target trajectory point of the target object, and taking the target trajectory point as the center point, a spatiotemporal analysis range is defined by combining a preset time window and spatial radius; by calling the spatial coverage function, a set of grid cell identifiers covering the circular spatial area is calculated based on the latitude and longitude of the center point and the spatial radius; using the set of grid cell identifiers, potential accompanying objects in which the trajectory point falls within the circular spatial area within the time window are identified; For each potential companion object, a time decay coefficient is calculated based on the time difference between the trajectory point of the potential companion object and the target trajectory point, and a distance decay coefficient is calculated based on the distance between the trajectory point of the potential companion object and the target trajectory point. The hit ratio of the number of trajectory points of the potential companion object falling within the spatiotemporal analysis range to the total number of trajectory points of the potential companion object is calculated within the time window. The time decay coefficient, distance decay coefficient, and hit ratio are weighted and summed to obtain a comprehensive companion score, and a specific companion object is determined based on the comprehensive companion score.

2. The method according to claim 1, characterized in that, The data drift processing procedure includes: Maintain a sliding time window, within which monitor the base station identifier sequence in continuous raw trajectory data records; When a sequence pattern of switching from the first base station to the second base station and then back to the first base station is detected within a preset time interval, the original trajectory data corresponding to the second base station is removed.

3. The method according to claim 1, characterized in that, The data compression methods include: The original trajectory data is grouped according to user identifier and base station identifier; for multiple original trajectory data of the same user under the same base station ID and whose timestamps fall within a preset time period, they are merged into one trajectory record; a representative timestamp is generated for the merged trajectory record, which is determined by calculating the average or median value of multiple timestamps of the merged original trajectory data. And / or, Calculate the hash value of multiple original trajectory data separately. The hash value is generated based on key fields in user identifier, timestamp and base station identifier. Merge multiple original trajectory data with the same hash value into one trajectory data record.

4. The method according to claim 1, characterized in that, The process of calculating the set of grid cell identifiers covering the circular spatial region by calling the spatial coverage function, based on the latitude and longitude of the center point and the spatial radius, includes: The spatial coverage function is invoked. The input parameters of the spatial coverage function include the longitude and latitude of the center point, the spatial radius, the preset minimum precision level, the preset maximum precision level, and the maximum number of elements to be generated. The following steps are then performed through the spatial coverage function: The location of the center point is determined by its longitude and latitude. The circular region is determined based on the center position and the input spatial radius; Generate grid cell identifiers starting from the minimum precision level; Determine whether the number of currently generated grid cell identifiers is less than the maximum number of generated identifiers and whether the current precision level is less than the maximum precision level; If so, increase the precision level, continue generating grid cell identifiers, and repeat the steps of determining whether the number of currently generated grid cell identifiers is less than the maximum number of generated identifiers and whether the current precision level is less than the maximum precision level. If not, return the currently generated set of grid cell identifiers, ensuring that the number of grid cell identifiers in the returned set does not exceed the maximum number of generated identifiers.

5. The method according to claim 1, characterized in that, Calculating the distance attenuation coefficient based on the distance between the trajectory points of the potential accompanying object and the target trajectory point includes: Calculate the target distance between the trajectory points of the potential accompanying object and the target trajectory point, and calculate the ratio of the target distance to the spatial radius; wherein the target distance is less than the spatial radius; By subtracting the ratio from the potential associated trajectory point, the distance attenuation coefficient can be obtained. The calculation process for the target distance includes: The spherical distance calculation function is invoked, and the input parameters of the spherical distance calculation function include the longitude and latitude of the trajectory point of the potential accompanying object, as well as the longitude and latitude of the target trajectory point; The spherical distance calculation function is based on the Haversine algorithm and performs the following calculation process: The longitude and latitude of the trajectory points of the potential accompanying object, as well as the longitude and latitude of the target trajectory point, are converted from angles to radians. The spherical distance between the trajectory points of the potential accompanying object and the target trajectory point is calculated according to the Haversine formula, which takes into account the Earth's radius and the difference in latitude and longitude between the trajectory points of the potential accompanying object and the target trajectory point.

6. The method according to claim 1, characterized in that, The step of querying the corresponding latitude and longitude coordinates from a pre-set base station knowledge base based on the base station identifier and associating the latitude and longitude coordinates with the corresponding target trajectory data includes: A base station knowledge base containing the correspondence between base station identifiers and latitude and longitude coordinates is pre-built; An index is created in the base station knowledge base based on the base station identifier; By querying the base station knowledge base, the base station identifiers in the target trajectory data are mapped to the corresponding latitude and longitude coordinates.

7. The method according to claim 1, characterized in that, Based on the associated latitude and longitude coordinates, a spatial grid algorithm is used to simultaneously generate and associate multiple grid cell identifiers of different precision levels for each target trajectory data, including: For each target trajectory data, the streaming processing engine calculates in real time the grid cell identifiers corresponding to the latitude and longitude coordinates at multiple different precision levels. The grid cell identifiers of each accuracy level obtained from the calculation are added as new fields and associated with the corresponding target trajectory data for storage.

8. The method according to any one of claims 1 to 7, characterized in that, The method is implemented based on a massively parallel processing (MPP) architecture, wherein: The target trajectory data is stored in a distributed manner in a columnar storage format; The vectorized execution engine distributes computational tasks across multiple computing nodes for parallel execution. The raw trajectory data is captured and processed in real time using a stream processing engine.

9. The method according to claim 8, characterized in that, The process of ingesting and processing the raw trajectory data in real time via a stream processing engine includes: Real-time reading of the raw trajectory data of the incoming flow; The original trajectory data is processed in real time to obtain the target trajectory data; Grid cell identifiers for real-time calculation of target trajectory data; The target trajectory data is written to the distributed storage system in real time.

10. The method according to claim 8, characterized in that, The vectorized execution engine distributes computational tasks across multiple computing nodes for parallel execution, including: Receive a companion analysis query request, the request including target trajectory points and search parameters; The spatial query task is decomposed into multiple sub-tasks based on data distribution characteristics; Each compute node performs the following operations in parallel: Retrieve trajectory data of potential accompanying objects based on grid cell identifier index; Perform multi-dimensional scoring calculations on the search results; The intermediate results of each computing node are summarized and combined into a final comprehensive accompanying score.

Images