Anomaly Analysis Method and Device for 5G High-Precision Positioning in Open-Pit Mines
By acquiring and associating the location and identity data of monitored objects in open-pit mines, and using graph deep learning models to analyze open-pit mine anomalies, the problem of low reliability in anomaly analysis in existing technologies is solved, and higher-precision anomaly identification and processing are achieved.
Patent Information
- Application Number
- CN202410493317.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-04-23
- Publication Date
- 2026-11-10
- Estimated Expiration
- 2044-04-23
AI Technical Summary
The reliability of open-pit mine anomaly analysis in existing technologies is relatively low, mainly because simple comparison of location data with preset locations fails to effectively capture complex behavioral anomalies.
By acquiring object representation data of all monitored objects within the open-pit mine area, including location data and identity data, correlation mining is performed to form a vector of monitored objects to be processed. This vector is then analyzed using a graph deep learning model to determine the abnormal state of the target monitored object and issue a prompt message.
It improves the reliability of anomaly analysis in open-pit mines. By combining correlation mining and graph deep learning models, it achieves higher-precision anomaly analysis and can more accurately identify and handle potential behavioral anomalies.
Smart Images

Figure CN118429886B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of coal mining technology, and more specifically, to an anomaly analysis method, an anomaly analysis device, a computer-readable storage medium, and a computer program product for 5G high-precision positioning in open-pit mines. Background Technology
[0002] 5G high-precision positioning equipment is a device that utilizes 5G technology for positioning, providing more accurate location information. Compared to traditional GPS or base station positioning, 5G high-precision positioning equipment leverages the characteristics of 5G networks, such as multiple antennas, high bandwidth, and low latency, to offer higher positioning accuracy and faster response speeds. By receiving information such as time, frequency, and interference from 5G signals, 5G high-precision positioning equipment can calculate the user's location and provide more accurate positioning results. It has broad application potential in fields such as intelligent transportation, logistics tracking, autonomous driving, and mine monitoring.
[0003] Open-pit mining is a method of extraction commonly used to obtain underground resources such as coal and metal ores. However, managing equipment and personnel during open-pit mining presents greater challenges. Therefore, positioning devices are typically installed on equipment and personnel to collect relevant location data. This data is then used to analyze anomalies in equipment and personnel behavior. However, current technologies simply compare the positioning data with preset locations, such as a simple determination of whether a restricted area has been entered. Consequently, the reliability of anomaly analysis is relatively low. Summary of the Invention
[0004] The main objective of this application is to provide an anomaly analysis method, anomaly analysis device, computer-readable storage medium, and computer program product for 5G high-precision positioning in open-pit mines, so as to at least solve the problem of relatively low reliability of anomaly analysis in open-pit mines in the prior art.
[0005] To achieve the objective, according to one aspect of this application, an anomaly analysis method for 5G high-precision positioning in open-pit mines is provided, comprising: acquiring object representation data corresponding to all monitored objects within a target open-pit mine area, wherein any monitored object belongs to a person or a device within the target open-pit mine area, and the object representation data includes at least positioning data, the positioning data including the location information of the monitored object at at least one monitoring time point; performing correlation mining on the positioning data of the target monitored object and the positioning data of related monitored objects to obtain a vector of monitored objects to be processed corresponding to the target monitored object, wherein the related monitored objects and the target monitored object have an action correlation within the target open-pit mine area, and the vector of monitored objects to be processed includes the action semantic features of the target monitored object and the target monitored object; analyzing the vector of monitored objects to be processed to obtain target object monitoring data corresponding to the target monitored object, the target object monitoring data being used to reflect the type or probability of action anomalies of the target monitored object within the target open-pit mine area; determining the abnormal state of the target monitored object based on the target object monitoring data and issuing a prompt message for anomaly handling.
[0006] Optionally, the location data of the target monitoring object and the location data of related monitoring objects are correlated and mined to obtain a vector of monitoring objects to be processed corresponding to the target monitoring object. This includes: performing preliminary correlation processing on the location data of the target monitoring object and the location data of related monitoring objects to obtain an intermediate first object vector and an intermediate second object vector. The intermediate first object vector is semantic information obtained by preliminary vectorization of the location data of the target monitoring object, and the intermediate second object vector is semantic information obtained by preliminary vectorization of the location data of the related monitoring objects. The semantic information is the behavioral state information of the target monitoring object in the target open-pit mine area. A corresponding category mapping parameter is determined based on the target action-related categories. The category mapping parameter is used to give different emphases to different semantic information according to different target action-related categories. The target action-related categories refer to the types of causes that form the action correlation. A linear mapping is performed on the intermediate first object vector according to the category mapping parameter to form an initial first object vector of the target monitoring object, and a linear mapping is performed on the intermediate second object vector according to the category mapping parameter to form an initial second object vector of the related monitoring objects. The initial first object vector carries a portion of the related monitoring objects. The semantic information includes a portion of the semantic information of the target monitored object carried in the initial second object vector; a first probability parameter is determined based on the initial first object vector, and a second probability parameter is determined based on the initial second object vector, wherein the first probability parameter is the category probability of the target action-related category corresponding to the target monitored object, and the second probability parameter is the category probability of the target action-related category corresponding to the relevant monitored object; a first related category vector is obtained based on the initial first object vector, the initial second object vector, and the second probability parameter, and a second related category vector is obtained based on the initial second object vector, the initial first object vector, and the first probability parameter, wherein the first related category vector is the category probability of the target monitored object for the target action-related category, and the second related category vector is the category probability of the relevant monitored object for the target action-related category; the second related category vector is propagated to the first related category vector to form at least one propagated related category vector, wherein the propagated related category vector is the probability of the target monitored object relative to the target action-related category, and one propagated related category vector corresponds to one propagated related category vector; the propagated related category vectors are aggregated to obtain the monitoring object vector to be processed.
[0007] Optionally, the object representation data further includes identity data, which is identification information related to the monitored object. Preliminary association processing is performed based on the location data of the target monitored object and the location data of related monitored objects to obtain intermediate first object vectors and intermediate second object vectors. This includes: obtaining a first location dimension vector and a first identity dimension vector of the target monitored object based on the location data and identity data corresponding to the target monitored object. The first location dimension vector is used to represent the action semantic features of the target monitored object, and the first identity dimension vector is used to represent the identity semantic features of the target monitored object; aggregating the first location dimension vector and the first identity dimension vector to obtain the intermediate first object vector; mining a second location dimension vector and a second identity dimension vector of the related monitored objects based on the location data and identity data corresponding to the related monitored objects. The second location dimension vector is used to represent the action semantic features of the related monitored objects, and the second identity dimension vector is used to represent the identity semantic features of the related monitored objects; aggregating the second location dimension vector and the second identity dimension vector to obtain the intermediate second object vector.
[0008] Optionally, obtaining a first relevant category vector based on the initial first object vector, the initial second object vector, and the second possibility parameter, and obtaining a second relevant category vector based on the initial second object vector, the initial first object vector, and the first possibility parameter, includes: multiplying the initial second object vector and the second possibility parameter to obtain a pending first object vector corresponding to the target monitored object; superimposing the initial first object vector and the pending first object vector to obtain a phased first object vector, wherein the phased first object vector represents the action semantic feature of the target monitored object relative to the target action-related category; obtaining the first relevant category vector based on the phased first object vector; performing an operation on the initial first object vector and the first possibility parameter to obtain a pending second object vector corresponding to the relevant monitored object; superimposing the initial second object vector and the pending second object vector to obtain a phased second object vector; and obtaining a second relevant category vector of the relevant monitored object relative to the target action-related category based on the phased second object vector.
[0009] Optionally, determining a first probability parameter based on the initial first object vector and a second probability parameter based on the initial second object vector includes: analyzing and obtaining an intermediate third related category vector based on the initial first object vector, the reference probability parameter, and the initial second object vector, wherein the reference probability parameter is a pre-configured parameter; analyzing and obtaining an intermediate fourth related category vector based on the initial second object vector, the reference probability parameter, and the initial first object vector; performing exponential operation and normalization on the fusion result of the initial first object vector and the intermediate third related category vector to obtain the first probability parameter; and performing exponential operation and normalization on the fusion result of the initial second object vector and the intermediate fourth related category vector to obtain the second probability parameter.
[0010] Optionally, propagating the second relevant category vector to the first relevant category vector to form at least one propagated relevant category vector includes: linearly mapping the first relevant category vector according to the category mapping parameters to obtain a first mapped object vector corresponding to the target monitoring object; linearly mapping the second relevant category vector according to the category mapping parameters to obtain a second mapped object vector corresponding to the relevant monitoring object; obtaining an intermediate fifth relevant category vector based on the first mapped object vector, the reference probability parameter, and the second mapped object vector, wherein the intermediate fifth relevant category vector is a new intermediate relevant category vector that the target monitoring object has relative to the target action relevant category; fusing the intermediate fifth relevant category vector and the second mapped object vector to obtain a fusion result, and sequentially performing exponential operation and normalization on the fusion result. The process involves processing the vector to obtain corresponding stage-specific probability parameters. Based on the first mapping object vector, the second mapping object vector, and the stage-specific probability parameters, an intermediate sixth related category vector is obtained. This intermediate sixth related category vector is a new intermediate related category vector that the target monitoring object possesses relative to the target action related category. The second related category vector is propagated to the first related category vector according to a vector propagation rule to form a fused related category vector. This vector propagation rule is a vector propagation rule in a graph deep learning model, which includes at least two graph distribution members corresponding to each of the monitoring objects. The at least two monitoring objects include the target monitoring object and the related monitoring objects. Based on the vector propagation rule, the fused related category vector, and the intermediate sixth related category vector, the propagated related category vector is formed.
[0011] Optionally, the analysis of the vector of the monitoring object to be processed to obtain the target object monitoring data corresponding to the target monitoring object includes: using a graph deep learning model to perform object analysis on the vector of the monitoring object to be processed to obtain a target probability parameter distribution, wherein the target probability parameter distribution is the probability that the target monitoring object belongs to each cause category in the target action related categories; and determining the action anomaly category corresponding to the maximum value in the target probability parameter distribution as the target object monitoring data.
[0012] According to another aspect of this application, an anomaly analysis device for 5G high-precision positioning in open-pit mines is provided. The device includes: an acquisition unit, configured to acquire object representation data corresponding to all monitored objects within a target open-pit mine area, wherein any monitored object belongs to a person or a device within the target open-pit mine area, and the object representation data includes at least positioning data, which includes the location information of the monitored object at at least one monitoring time point; an association unit, configured to perform association mining on the positioning data of the target monitored object and the positioning data of related monitored objects to obtain a vector of monitored objects to be processed corresponding to the target monitored object, wherein the related monitored objects and the target monitored object have an action correlation within the target open-pit mine area, and the vector of monitored objects to be processed includes the action semantic features of the target monitored object and the target monitored object; an analysis unit, configured to analyze the vector of monitored objects to be processed to obtain target object monitoring data corresponding to the target monitored object, wherein the target object monitoring data reflects the type or probability of action anomalies of the target monitored object within the target open-pit mine area; and a determination unit, configured to determine the abnormal state of the target monitored object based on the target object monitoring data and control the issuance of prompt information for anomaly handling.
[0013] According to another aspect of this application, a computer-readable storage medium is provided, the computer-readable storage medium including a stored program, wherein, when the program is executed, it controls the device on which the computer-readable storage medium is located to perform any of the methods described.
[0014] According to another aspect of this application, a computer program product is provided, including computer instructions that, when executed by a processor, implement any of the methods described.
[0015] By applying the technical solution of this application, the application first acquires the object representation data of each monitored object within the open-pit mine area, then performs correlation mining on the location data of the target monitored object and related monitored objects to obtain the vector of the monitored object to be processed corresponding to the target monitored object. Based on the vector of the monitored object to be processed, the monitoring data of the target monitored object corresponding to the target monitored object is analyzed. The abnormal state of the target monitored object can be determined according to the status of the monitoring data of the target monitored object. Related monitored objects are personnel or equipment in the target open-pit mine area that are related to the target monitored object. The vector of the monitored object to be processed carries not only the action semantic features of the target monitored object, but also the action semantic features of related monitored objects. This vector has an auxiliary analytical function for the action semantic features of the target monitored object, thus achieving higher accuracy analysis and improving the reliability of anomaly analysis in open-pit mines to a certain extent. This application solves the problem of relatively low reliability of anomaly analysis in open-pit mines in the prior art. Attached Figure Description
[0016] Figure 1 A hardware structure block diagram of a mobile terminal for performing anomaly analysis method for 5G high-precision positioning in open-pit mines, according to an embodiment of this application, is shown.
[0017] Figure 2 A flowchart illustrating an anomaly analysis method for 5G high-precision positioning in open-pit mines according to an embodiment of this application is shown.
[0018] Figure 3 A structural block diagram of an anomaly analysis device for 5G high-precision positioning in an open-pit mine, according to an embodiment of this application, is shown.
[0019] The accompanying drawings include the following reference numerals:
[0020] 102. Processor; 104. Memory; 106. Transmission device; 108. Input / output device. Detailed Implementation
[0021] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.
[0022] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.
[0023] It should be noted that the terms "first," "second," etc., used in the specification, claims, and drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be used interchangeably where appropriate for the embodiments of this application described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.
[0024] As described in the background section, existing technologies typically equip equipment and personnel with positioning devices to collect corresponding positioning data, and then perform anomaly analysis on the movement of equipment and personnel based on the positioning data. However, existing technologies simply compare the positioning data with preset locations, such as simply determining whether a restricted area has been entered. To address the issue of relatively low reliability in anomaly analysis, embodiments of this application provide an anomaly analysis method, anomaly analysis device, computer-readable storage medium, and computer program product for 5G high-precision positioning in open-pit mines.
[0025] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.
[0026] The methods and embodiments provided in this application can be executed on a mobile terminal, computer terminal, or similar computing device. Taking running on a mobile terminal as an example, Figure 1 This is a hardware structure block diagram of a mobile terminal for an anomaly analysis method of 5G high-precision positioning in an open-pit mine, according to an embodiment of the present invention. Figure 1 As shown, a mobile terminal may include one or more ( Figure 1 Only one is shown in the diagram. A processor 102 (which may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.) and a memory 104 for storing data are also shown. The mobile terminal may further include a transmission device 106 for communication functions and an input / output device 108. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of a mobile terminal. For example, a mobile terminal may also include components that are more... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.
[0027] The memory 104 can be used to store computer programs, such as application software programs and modules, like the computer program corresponding to the device information display method in this embodiment of the invention. The processor 102 executes various functional applications and data processing, i.e., the implemented method, by running the computer program stored in the memory 104. The memory 104 may include high-speed random access memory and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the mobile terminal via a network. Examples of networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof. The transmission device 106 is used to receive or send data via a network. Specific examples of the network may include a wireless network provided by the mobile terminal's communication provider. In one instance, the transmission device 106 includes a network interface controller (NIC), which can be connected to other network devices via a base station to communicate with the Internet. In one example, the transmission device 106 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.
[0028] This embodiment provides an anomaly analysis method for 5G high-precision positioning in open-pit mines that runs on mobile terminals, computer terminals, or similar computing devices. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.
[0029] Figure 2 This is a flowchart of an anomaly analysis method for 5G high-precision positioning in open-pit mines according to an embodiment of this application. Figure 2 As shown, the method includes the following steps:
[0030] Step S201: Obtain object representation data corresponding to all monitored objects within the target open-pit mine area. Any monitored object belongs to a person or a piece of equipment within the target open-pit mine area. The object representation data includes at least location data, which includes the location information of the monitored object at at least one monitoring time point.
[0031] Specifically, location data is collected for each monitored object within the target open-pit mine area. Each monitored object belongs to a person or piece of equipment within the target open-pit mine area. The location data is generated by collecting location information through 5G positioning devices deployed on the monitored objects, and includes the location information of the monitored object at at least one monitoring time point, used to characterize action information. For example, each worker wears a watch equipped with a 5G positioning device, and corresponding 5G positioning devices are also installed on equipment (such as trucks, excavators, etc.). These devices will periodically send location information, including the location information of each monitored object at at least one monitoring time point. Additionally, for example, the monitored personnel could be: mine blasters: specifically responsible for safely conducting blasting operations to ensure effective rock breaking; tunneling workers: responsible for mining new mine shafts or expanding existing mine shafts, using tools such as drills and hammers for underground excavation; maintenance technicians: responsible for maintaining and repairing mining equipment, including excavators, trucks, and conveyor systems. The monitoring targets for equipment can include: open-pit excavators: large excavating equipment used to extract and remove topsoil and rock cover to expose ore; drilling rigs: used to drill holes in the mine for exploration and sampling analysis; ore screening machines: used to separate and classify ore according to particle size and ore grade; and transport trucks: used to transport mining products from the open-pit mine area to processing plants or storage facilities. For a specific monitoring target, the location data is as follows: Time point 1: September 20, 2023, 9:00 AM; Location data: Latitude 41.3021, Longitude -112.1235; Area: Mine entrance.
[0032] Step S202: The location data of the target monitoring object and the location data of related monitoring objects are correlated and mined to obtain the vector of the monitoring object to be processed corresponding to the target monitoring object. The related monitoring objects and the target monitoring object have action correlation within the target open-pit mine area. The vector of the monitoring object to be processed includes the target monitoring object and the action semantic features of the target monitoring object.
[0033] Specifically, after collecting the location data of each monitored object, in order to analyze the action characteristics of the target monitored object, the location data needs to be vectorized to obtain the corresponding vector representation, namely the monitoring object vector to be processed. The monitoring object vector corresponding to the target monitored object carries not only the action semantic features of the target monitored object, but also the action semantic features of related monitored objects. Assuming worker A is the target monitored object, the monitoring objects related to it may be other workers B and C nearby. Then, the monitoring object vector corresponding to the target monitored object (worker A) is extracted through a data mining algorithm. After obtaining the monitoring object vector to be processed for the target monitored object (worker A), the system will perform further analysis based on the monitoring object vector to determine the type or probability of abnormal action of the target monitored object. The types of abnormal action can be: unauthorized monitoring object entering a sensitive area, or the movement trajectory of the monitored object not conforming to the preset trajectory.
[0034] Step S203: Analyze the vector of the monitored object to be processed to obtain the monitoring data of the target object. The monitoring data of the target object is used to reflect the type or probability of abnormal behavior of the target monitored object in the target open-pit mine area.
[0035] Specifically, since the vector of the target monitoring object carries not only the action semantic features of the target monitoring object but also the action semantic features of related monitoring objects, and since both the target monitoring object and related monitoring objects belong to the target open-pit mine area (i.e., they are correlated within the target open-pit mine area), the action semantic features of related monitoring objects can assist in the analysis of the action semantic features of the target monitoring object. Therefore, object analysis based on the vector of the target monitoring object can achieve higher accuracy, thereby improving the reliability of anomaly analysis in open-pit mines and addressing the problem of relatively low reliability in anomaly analysis. Furthermore, by combining the location data of related monitoring objects, the accuracy of the analysis can be further improved; for example, in situations where a large number of personnel enter... In a blasting site, analyzing only the location data of a single person to obtain the target monitoring data might reveal the types of abnormal actions that could lead to entering a dangerous area (or, in other words, the probability of obtaining abnormal actions would be relatively high). However, if the location data of related monitored objects are combined, the semantic features of the vector representation of the monitored objects to be processed will contain information about multiple people entering the dangerous area. In this case, the types of abnormal actions that could lead to entering the dangerous area might not be identified. This is because multiple people entering the dangerous area simultaneously might mean that the danger has been eliminated or that they are on a special mission, meaning it is not considered an accidental entry into the dangerous area. Therefore, it can be considered as an abnormal action (or, in other words, the probability of obtaining abnormal actions would be relatively low). In other words, joint analysis can improve the reliability of anomaly analysis.
[0036] Step S204: Determine the abnormal state of the target monitored object based on the target object monitoring data and issue a prompt message to handle the abnormality.
[0037] Specifically, based on the target object monitoring data, the final type of abnormal behavior of the target object (or, the probability of abnormal behavior is high) is determined, and a prompt message is issued to alert the target object to enter a dangerous area or that danger may occur, so as to handle the abnormality in a timely manner and avoid danger.
[0038] Through this embodiment, firstly, object representation data corresponding to all monitored objects within the target open-pit mine area is acquired. Each monitored object belongs to a person or a piece of equipment within the target open-pit mine area. The object representation data includes at least location data, which includes the location information of the monitored object at at least one monitoring time point. Then, the location data of the target monitored object and the location data of related monitored objects are correlated and mined to obtain the vector of the monitored object to be processed corresponding to the target monitored object. There is an action correlation between the related monitored objects and the target monitored object within the target open-pit mine area. The vector of the monitored object to be processed includes the target monitored object and the action semantic features of the target monitored object. Next, the vector of the monitored object to be processed is analyzed to obtain the target object monitoring data corresponding to the target monitored object. The target object monitoring data is used to reflect the type or probability of action anomalies of the target monitored object within the target open-pit mine area. Finally, the abnormal state of the target monitored object is determined based on the target object monitoring data, and a prompt message is issued for anomaly handling. This application first acquires object representation data for each monitored object within the open-pit mine area. Then, it performs correlation mining on the location data of the target monitored object and related monitored objects to obtain a vector of monitored objects to be processed corresponding to the target monitored object. Based on this vector, it analyzes the monitoring data of the target monitored object. The abnormal state of the target monitored object can be determined based on the status of the monitoring data. Related monitored objects are personnel or equipment in the target open-pit mine area that are related to the target monitored object. The vector of monitored objects to be processed carries not only the action semantic features of the target monitored object but also the action semantic features of related monitored objects. This provides auxiliary analysis for the action semantic features of the target monitored object, thus achieving higher accuracy analysis and improving the reliability of anomaly analysis in open-pit mines to a certain extent. This application solves the problem of relatively low reliability in existing technologies for anomaly analysis in open-pit mines.
[0039] To enable those skilled in the art to better understand the technical solution of this application, the implementation process of the anomaly analysis method for 5G high-precision positioning in open-pit mines will be described in detail below with reference to specific embodiments.
[0040] To improve the accuracy of inferring the behavioral state of the target monitored object, in one optional implementation, step S202 includes:
[0041] Step S2021: Perform preliminary association processing based on the location data of the target monitoring object and the location data of related monitoring objects to obtain intermediate first object vector and intermediate second object vector. The intermediate first object vector is the semantic information obtained by preliminary vectorization of the location data of the target monitoring object, and the intermediate second object vector is the semantic information obtained by preliminary vectorization of the location data of related monitoring objects. The semantic information is the behavioral status information of the target monitoring object in the target open-pit mine area.
[0042] Specifically, the location data of the target monitoring object and the location data of related monitoring objects are first vectorized to obtain intermediate object vectors (not final) of the target monitoring object and related monitoring objects, namely the first intermediate object vector and the second intermediate object vector.
[0043] Step S2022: Determine the corresponding category mapping parameters based on the target action related categories. The category mapping parameters are used to give different emphases to different semantic information according to different target action related categories. Target action related categories refer to the types of reasons that form action relatedness.
[0044] Specifically, different mappings may be required for different types of target actions, so that different information may be given different emphases or attention. Therefore, it is necessary to determine the category mapping parameters corresponding to the types of target actions in order to perform more accurate mappings.
[0045] Step S2023: Perform linear mapping on the intermediate first object vector according to the category mapping parameters to form the initial first object vector of the target monitoring object, and perform linear mapping on the intermediate second object vector according to the category mapping parameters to form the initial second object vector of the related monitoring object. The initial first object vector carries some semantic information of the related monitoring object, and the initial second object vector carries some semantic information of the target monitoring object.
[0046] Specifically, the category mapping parameters may include a first parameter and a second parameter. The first parameter can be used as a weight to weight the intermediate first object vector, and the second parameter can be used as a bias to superimpose the weighted results to obtain the initial first object vector. Similarly, for the initial second object vector, the first parameter can be used as a weight to weight the intermediate second object vector, and the second parameter can be used as a bias to superimpose the weighted results to obtain the initial second object vector.
[0047] Step S2024: Determine a first probability parameter based on the initial first object vector, and determine a second probability parameter based on the initial second object vector. The first probability parameter is the category probability of the target action-related categories corresponding to the target monitored object, and the second probability parameter is the category probability of the target action-related categories corresponding to the relevant monitored object.
[0048] Specifically, a first probability parameter can be determined by processing the abnormal action type or probability of the target monitored object represented in the initial first object vector, and a second probability parameter can be determined by processing the abnormal action type or probability of the relevant monitored object represented in the initial second object vector.
[0049] Step S2025: Based on the initial first object vector, the initial second object vector, and the second probability parameter, a first related category vector is obtained, and based on the initial second object vector, the initial first object vector, and the first probability parameter, a second related category vector is obtained. The first related category vector represents the category probability of the target monitored object for the target action related category, and the second related category vector represents the category probability of the related monitored object for the target action related category.
[0050] Specifically, a correlation mining process is first performed between the initial first object vector and the initial second object vector. When performing correlation mining on the initial first object vector based on the information in the initial second object vector, the probability that the monitored object belongs to the target action-related category is considered. The higher the probability, the higher the importance of the correlation mining. Similarly, when performing correlation mining on the initial second object vector based on the information in the initial first object vector, the probability that the monitored object belongs to the target action-related category is considered. The higher the probability, the higher the importance of the correlation mining.
[0051] Step S2026: Propagate the second related category vector to the first related category vector to form at least one propagated related category vector. The propagated related category vector represents the probability that the target monitored object has relative to the target action related category. One target action related category corresponds to one propagated related category vector.
[0052] Specifically, after completing the mutual correlation mining based on the aforementioned steps, the vectors of both the target monitoring object and related monitoring objects will carry some semantic information about each other. At this point, by propagating the second correlation category vector of the related monitoring object to the first correlation category vector of the target monitoring object, the resulting propagated correlation category vector can fully represent the semantic features of both the target and related monitoring objects, effectively improving its semantic representation capability. The second and first correlation category vectors can be directly superimposed. When there is only one type of target action correlation, such as the action correlation formed by the intersection of movement trajectories or the action correlation formed by job type, a propagated correlation category vector can be formed. When there are multiple types of target action correlation, such as the action correlation formed by the intersection of movement trajectories and the action correlation formed by job type, a corresponding propagated correlation category vector can be obtained for each type of target action correlation.
[0053] Step S2027: Aggregate the propagation-related type vectors to obtain the vector of the monitored object to be processed.
[0054] Specifically, when there is only one type of target action correlation, such as the action correlation formed by the intersection of movement trajectories or the action correlation formed by job type, the propagation correlation type vector can be directly used as the vector of the target monitoring object to be processed. When there are multiple types of target action correlation, the propagation correlation type vectors corresponding to various target action correlation types are aggregated, such as by superposition, averaging, or concatenation, to obtain the vector of the target monitoring object to be processed. In this way, the correlation of the monitoring object in different dimensions can be fully considered, so that the final vector of the target monitoring object to be processed has higher accuracy in representing the semantic features of the monitoring object. By obtaining the target action correlation types corresponding to the action correlation between the target monitoring object and related monitoring objects, the category probability of the related monitoring object for that target action correlation type can be obtained. The propagation correlation type vector of the target monitoring object for that target action correlation type can be generated through this category probability. This propagation correlation type vector includes the category representation feature of the related monitoring object for that target action correlation type obtained through feature transfer (i.e., the second correlation type vector).
[0055] To facilitate further semantic and association analysis, in one optional implementation, step S2021 includes:
[0056] Step S20211: Based on the location data and identity data corresponding to the target monitoring object, obtain the first location dimension vector and the first identity dimension vector of the target monitoring object. The first location dimension vector is used to represent the action semantic features of the target monitoring object, and the first identity dimension vector is used to represent the identity semantic features of the target monitoring object.
[0057] Specifically, vectorizing the location data of the monitored target object yields the first location dimension vector, and vectorizing the identity data yields the first identity dimension vector. The identity data can refer to the job type of the monitored target object, such as the work content of personnel or equipment. For example, location data could be "September 20, 2023, 9:00 AM; Location data: Latitude 41.3021, Longitude -112.1235; Region: Mine entrance," and identity data could be "Mine Engineer: Responsible for planning and managing mine activities, including designing and supervising mining operations." The location data can be vectorized (e.g., through a word embedding network) to obtain a first location dimension vector, such as [0.78, 0.36, 0.92, 0.15, 0.62, 0.81, 0.92, 0.45, 0.73, 0.28]; the identity data can be vectorized (e.g., through a word embedding network) to obtain a first identity dimension vector, such as [0.12, 0.25, 0.33, 0.18, 0.35, 0.21, 0.41, 0.29, 0.18, 0.26, 0.15, 0.32, 0.42, 0.27, 0.19, 0.34].
[0058] Step S20212: Aggregate the first positioning dimension vector and the first identity dimension vector to obtain the intermediate first object vector.
[0059] Specifically, the first positioning dimension vector and the first identity dimension vector can be superimposed or concatenated to obtain the intermediate first object vector of the target monitoring object.
[0060] Step S20213: Based on the location data and identity data corresponding to the relevant monitored objects, the second location dimension vector and the second identity dimension vector of the relevant monitored objects are mined. The second location dimension vector is used to characterize the action semantic features of the relevant monitored objects, and the second identity dimension vector is used to characterize the identity semantic features of the relevant monitored objects.
[0061] Specifically, vectorizing the location data of the relevant monitored objects yields a second location dimension vector, and vectorizing the identity data yields a second identity dimension vector. Here, identity data can refer to the job type of the relevant monitored objects, such as the work content of personnel or equipment.
[0062] Step S20214: Aggregate the second positioning dimension vector and the second identity dimension vector to obtain the intermediate second object vector.
[0063] Specifically, the second positioning dimension vector and the second identity dimension vector can be superimposed or concatenated to obtain the intermediate second object vector of the relevant monitored object.
[0064] To make the performance of the target monitoring object and related monitoring objects more accurate across different target action-related categories, in one optional implementation, step S2024 includes:
[0065] Step S20241: Based on the initial first object vector, the reference probability parameter, and the initial second object vector, the intermediate third related category vector is obtained through analysis. The reference probability parameter is a pre-configured parameter.
[0066] Specifically, the reference probability parameter can be initial, original, or in a default state, and is the same as the number of target action related categories. When there is only one target action related category, the reference probability parameter can be equal to 1. When there are two target action related categories (such as action correlation based on the intersection of movement trajectories and action correlation based on job type), the reference probability parameter can be equal to 1 / 2, which is an average probability parameter. Based on the initial first object vector, the reference probability parameter, and the initial second object vector, the intermediate related category vector of the target monitored object relative to the target action related categories is analyzed. For example, multiplying the reference probability parameter and the initial second object vector, and then adding it to the initial first object vector, yields the intermediate related category vector, i.e., the intermediate third related category vector.
[0067] Step S20242: Based on the initial second object vector, the reference probability parameter, and the initial first object vector, the intermediate fourth related category vector is obtained through analysis.
[0068] Specifically, the reference probability parameter is multiplied by the initial first object vector, and then added to the initial second object vector to obtain the intermediate related category vector, which is the intermediate fourth related category vector.
[0069] Step S20243: Perform exponential operation and normalization on the fusion result of the initial first object vector and the intermediate third related category vector to obtain the first probability parameter.
[0070] Specifically, the fusion result of the initial first object vector and the intermediate third related category vector is subjected to exponential operation and normalization to obtain the first probability parameter that the target monitored object belongs to the target action related category.
[0071] Step S20244: Perform exponential operation and normalization on the fusion result of the initial second object vector and the intermediate fourth related category vector to obtain the second possibility parameter.
[0072] Specifically, the fusion result of the initial second object vector and the intermediate fourth related category vector is subjected to exponential operation and normalization to obtain the second probability parameter of the relevant monitored object belonging to the target action related category. The first probability parameter and the second probability parameter are used to represent the probability of belonging to the corresponding target action related category.
[0073] To more accurately understand the relevant category and likelihood of the target monitored object, in one optional implementation, step S2025 includes:
[0074] Step S20251: Multiply the initial second object vector and the second possibility parameter to obtain the undetermined first object vector corresponding to the target monitoring object.
[0075] Specifically, the initial second object vector is multiplied by the second possibility parameter to output the corresponding undetermined first object vector. That is, the initial second object vector is weighted based on the second possibility parameter to apply the corresponding matching degree or importance to the initial second object vector.
[0076] Step S20252: Perform a superposition operation on the initial first object vector and the undetermined first object vector to obtain the staged first object vector. The staged first object vector is the action semantic feature of the target monitoring object relative to the target action related category.
[0077] Specifically, the initial first object vector and the undetermined first object vector are superimposed. Based on the output data of the superposition operation, the phased first object vector of the target monitoring object relative to the target action related category is obtained. For example, the output data of the superposition operation can be used as the phased first object vector.
[0078] Step S20253: Based on the first object vector of the stage, obtain the first related category vector.
[0079] Specifically, a pre-configured vector propagation rule is first determined. This vector propagation rule is a type of vector propagation rule in graph deep learning models. Based on this vector propagation rule, the intermediate second object vector is propagated to the intermediate first object vector to form the corresponding object propagation vector. For example, the intermediate second object vector and the intermediate first object vector are added together (superposition operation) to obtain the object propagation vector. The object propagation vector is then added to the stage-specific first object vector to obtain the first relevant category vector of the target monitoring object relative to the target action-related category.
[0080] Step S20254: Perform calculations on the initial first object vector and the first probability parameter to obtain the undetermined second object vector corresponding to the relevant monitored object.
[0081] Specifically, the initial first object vector and the first possibility parameter are multiplied to output the corresponding undetermined second object vector. That is, the initial first object vector is weighted based on the first possibility parameter to apply the corresponding matching degree or importance to the initial first object vector.
[0082] Step S20255: Perform a superposition operation on the initial second object vector and the undetermined second object vector to obtain a staged second object vector.
[0083] Specifically, the initial second object vector and the undetermined second object vector are superimposed. Based on the output data of the superposition operation, a phased second object vector of the target monitoring object relative to the target action-related category is obtained. For example, the output data of the superposition operation can be used as the phased second object vector.
[0084] Step S20256: Based on the phased second object vector, obtain the second related category vector of the relevant monitored object relative to the target action related category.
[0085] Specifically, based on this vector propagation rule, the intermediate first object vector is propagated to the intermediate second object vector to form the corresponding object propagation vector. For example, the intermediate second object vector and the intermediate first object vector are added together (superposition operation) to obtain the object propagation vector; the object propagation vector is added together with the stage second object vector to obtain the second related category vector of the target monitoring object relative to the target action related category.
[0086] To quantify the degree of correlation and improve the comprehensiveness and accuracy of the analysis, in one optional implementation, step S2026 includes:
[0087] Step S20261: Perform linear mapping on the first related category vector according to the category mapping parameters to obtain the first mapped object vector corresponding to the target monitoring object, and perform linear mapping on the second related category vector according to the category mapping parameters to obtain the second mapped object vector corresponding to the related monitoring object.
[0088] Specifically, based on the category mapping parameters corresponding to the target action-related categories and the first related category vector, the corresponding first mapping object vector is obtained; the category mapping parameters are linearly mapped to the first related category vector to obtain the first mapping object vector.
[0089] Step S20262: Based on the first mapping object vector, the reference probability parameter, and the second mapping object vector, the intermediate fifth related category vector is obtained. The intermediate fifth related category vector is a new intermediate related category vector that the target monitoring object has relative to the target action related category.
[0090] Specifically, the second mapping object vector can be weighted (i.e. multiplied) based on the reference probability parameter. Then, the weighted result is superimposed with the first mapping object vector to obtain a new intermediate related category vector, namely the intermediate fifth related category vector. Compared with the initial first object vector, the intermediate first related category vector, and the intermediate third related category vector, the accuracy is higher and it can more accurately reflect the abnormal state of the target monitoring object. Each calculation of the category vector of the target monitoring object can improve the accuracy. Each target action related category has an intermediate related category vector.
[0091] Step S20263: The intermediate fifth related category vector and the second mapping object vector are fused to obtain the fusion result, and the fusion result is then subjected to exponential operation and normalization to obtain the corresponding stage probability parameters.
[0092] Specifically, the transpose of the second mapping object vector can be multiplied with the intermediate fifth related category vector, and then the result of the multiplication can be exponentially calculated to obtain the first exponential value. The transpose of the second mapping object vector can also be multiplied with the new intermediate related category vectors of each of the other target action related categories, and then the result of the multiplication can be exponentially calculated to obtain each second exponential value. Then, the ratio of the first exponential value to (the sum of the first exponential value and each second exponential value) can be calculated to obtain the stage-specific probability parameter. Here, "stage" can refer to a step in the cyclic or iterative processing. The current stage-specific probability parameter can be used as the final probability parameter, or it can be used as a reference probability parameter for cyclic iteration to obtain the final probability parameter.
[0093] Step S20264: Based on the first mapping object vector, the second mapping object vector, and the stage possibility parameter, the intermediate sixth related category vector is obtained. The intermediate sixth related category vector is a new intermediate related category vector that the target monitoring object has relative to the target action related category.
[0094] Specifically, the first mapped object vector can be weighted (i.e. multiplied) based on the reference probability parameter. Then, the weighted result is superimposed on the second mapped object vector to obtain a new intermediate related category vector, namely the intermediate sixth related category vector. Compared with the initial second object vector, the intermediate second related category vector and the intermediate fourth related category vector, the accuracy is higher and it can more accurately reflect the abnormal state of the relevant monitored objects.
[0095] Step S20265: According to the vector propagation rule, the second related category vector is propagated to the first related category vector to form a fused related category vector. The vector propagation rule is a vector propagation rule in the graph deep learning model. The graph deep learning model includes graph distribution members corresponding to at least two monitoring objects. The at least two monitoring objects include the target monitoring object and related monitoring objects.
[0096] Specifically, according to the vector propagation rule, the second related category vector and the first related category vector can be added together to obtain the fused related category vector. When there are multiple second related category vectors, the multiple second related category vectors can be directly added together with the first related category vector to obtain the fused related category vector. Alternatively, the average of the multiple second related category vectors can be calculated first, and then the average vector can be added together with the first related category vector to obtain the fused related category vector.
[0097] Step S20266: Based on the vector propagation rules, the relevant category vectors are fused and the intermediate sixth relevant category vector is used to form the propagation relevant category vector.
[0098] Specifically, the fused relevant category vector and the new relevant category vector can be added together to obtain the propagation relevant category vector.
[0099] In order to accurately identify the type of abnormal behavior of the target monitored object, in one optional implementation, step S203 includes:
[0100] Step S2031: Use a graph deep learning model to perform object analysis on the vector of the monitored object to be processed, and obtain the target probability parameter distribution. The target probability parameter distribution is the probability of the monitored object belonging to each cause category in the target action related categories.
[0101] Specifically, using a graph deep learning model, the vectors of the monitored objects to be processed are analyzed to obtain the corresponding target probability parameter distribution. This distribution contains probability parameters corresponding to various types of behavioral anomalies, which are considered candidate types. During the training of the graph deep learning model, the data used includes actual behavioral anomaly types as labels (identifiers, supervision data). By comparing the estimated behavioral anomaly types with the actual types, the graph deep learning model can learn the mapping relationship between various monitored object vectors and various behavioral anomaly types. The graph deep learning model can include a multilayer perceptron for calculating propagation-related category vectors, thereby outputting multiple target probability parameters, i.e., obtaining the target probability parameter distribution. Furthermore, the graph deep learning model can be a graph convolutional neural network (GCN) or a graph attention network (GAT).
[0102] Step S2032: The type of action anomaly corresponding to the maximum value in the distribution of target probability parameters is determined as the target object monitoring data.
[0103] Specifically, in the distribution of target probability parameters, the target probability parameter with the maximum value is determined; based on the type of action anomaly corresponding to the target probability parameter with the maximum value, the target object monitoring data corresponding to the target monitoring object is determined. For example, the target probability parameter for the type of action anomaly entering the dangerous area is 0.7, and the target probability parameter for the type of action anomaly not entering the dangerous area is 0.2. In this way, the type of action anomaly entering the dangerous area can be used as the target object monitoring data.
[0104] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.
[0105] This application also provides an anomaly analysis device for 5G high-precision positioning in open-pit mines. It should be noted that this anomaly analysis device can be used to execute the anomaly analysis method for 5G high-precision positioning in open-pit mines provided in this application. This device is used to implement the embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.
[0106] The following describes the anomaly analysis device for 5G high-precision positioning in open-pit mines provided in the embodiments of this application.
[0107] Figure 3 This is a structural block diagram of an anomaly analysis device for 5G high-precision positioning in open-pit mines according to an embodiment of this application.
[0108] like Figure 3 As shown, the device includes:
[0109] The acquisition unit 10 is used to acquire object representation data corresponding to all monitored objects in the target open-pit mine area. Any monitored object belongs to a person or a piece of equipment in the target open-pit mine area. The object representation data includes at least location data, which includes the location information of the monitored object at at least one monitoring time point.
[0110] Specifically, location data is collected for each monitored object within the target open-pit mine area. Each monitored object belongs to a person or piece of equipment within the target open-pit mine area. The location data is generated by collecting location information through 5G positioning devices deployed on the monitored objects, and includes the location information of the monitored object at at least one monitoring time point, used to characterize action information. For example, each worker wears a watch equipped with a 5G positioning device, and corresponding 5G positioning devices are also installed on equipment (such as trucks, excavators, etc.). These devices will periodically send location information, including the location information of each monitored object at at least one monitoring time point. Additionally, for example, the monitored personnel could be: mine blasters: specifically responsible for safely conducting blasting operations to ensure effective rock breaking; tunneling workers: responsible for mining new mine shafts or expanding existing mine shafts, using tools such as drills and hammers for underground excavation; maintenance technicians: responsible for maintaining and repairing mining equipment, including excavators, trucks, and conveyor systems. The monitoring targets for equipment can include: open-pit excavators: large excavating equipment used to extract and remove topsoil and rock cover to expose ore; drilling rigs: used to drill holes in the mine for exploration and sampling analysis; ore screening machines: used to separate and classify ore according to particle size and ore grade; and transport trucks: used to transport mining products from the open-pit mine area to processing plants or storage facilities. For a specific monitoring target, the location data is as follows: Time point 1: September 20, 2023, 9:00 AM; Location data: Latitude 41.3021, Longitude -112.1235; Area: Mine entrance.
[0111] The association unit 20 is used to perform association mining on the location data of the target monitoring object and the location data of related monitoring objects to obtain the vector of the monitoring object to be processed corresponding to the target monitoring object. The related monitoring objects and the target monitoring object have action correlation within the target open-pit mine area. The vector of the monitoring object to be processed includes the target monitoring object and the action semantic features of the target monitoring object.
[0112] Specifically, after collecting the location data of each monitored object, in order to analyze the action characteristics of the target monitored object, the location data needs to be vectorized to obtain the corresponding vector representation, namely the monitoring object vector to be processed. The monitoring object vector corresponding to the target monitored object carries not only the action semantic features of the target monitored object, but also the action semantic features of related monitored objects. Assuming worker A is the target monitored object, the monitoring objects related to it may be other workers B and C nearby. Then, the monitoring object vector corresponding to the target monitored object (worker A) is extracted through a data mining algorithm. After obtaining the monitoring object vector to be processed for the target monitored object (worker A), the system will perform further analysis based on the monitoring object vector to determine the type or probability of abnormal action of the target monitored object. The types of abnormal action can be: unauthorized monitoring object entering a sensitive area, or the movement trajectory of the monitored object not conforming to the preset trajectory.
[0113] Analysis unit 30 is used to analyze the vector of the monitored object to be processed and obtain the target object monitoring data corresponding to the target monitored object. The target object monitoring data is used to reflect the type or probability of abnormal behavior of the target monitored object in the target open-pit mine area.
[0114] Specifically, since the vector of the target monitoring object carries not only the action semantic features of the target monitoring object but also the action semantic features of related monitoring objects, and since both the target monitoring object and related monitoring objects belong to the target open-pit mine area (i.e., they are correlated within the target open-pit mine area), the action semantic features of related monitoring objects can assist in the analysis of the action semantic features of the target monitoring object. Therefore, object analysis based on the vector of the target monitoring object can achieve higher accuracy, thereby improving the reliability of anomaly analysis in open-pit mines and addressing the problem of relatively low reliability in anomaly analysis. Furthermore, by combining the location data of related monitoring objects, the accuracy of the analysis can be further improved; for example, in situations where a large number of personnel enter... In a blasting site, analyzing only the location data of a single person to obtain the target monitoring data might reveal the types of abnormal actions that could lead to entering a dangerous area (or, in other words, the probability of obtaining abnormal actions would be relatively high). However, if the location data of related monitored objects are combined, the semantic features of the vector representation of the monitored objects to be processed will contain information about multiple people entering the dangerous area. In this case, the types of abnormal actions that could lead to entering the dangerous area might not be identified. This is because multiple people entering the dangerous area simultaneously might mean that the danger has been eliminated or that they are on a special mission, meaning it is not considered an accidental entry into the dangerous area. Therefore, it can be considered as an abnormal action (or, in other words, the probability of obtaining abnormal actions would be relatively low). In other words, joint analysis can improve the reliability of anomaly analysis.
[0115] The determination unit 40 is used to determine the abnormal state of the target monitored object based on the target object monitoring data and issue a prompt message to handle the abnormality.
[0116] Specifically, based on the target object monitoring data, the final type of abnormal behavior of the target object (or, the probability of abnormal behavior is high) is determined, and a prompt message is issued to alert the target object to enter a dangerous area or that danger may occur, so as to handle the abnormality in a timely manner and avoid danger.
[0117] In this embodiment, object representation data for each monitored object within the open-pit mine area is first acquired. Then, correlation mining is performed on the location data of the target monitored object and related monitored objects to obtain the vector of the monitored object to be processed corresponding to the target monitored object. Based on the vector of the monitored object to be processed, the monitoring data of the target monitored object is analyzed. The abnormal state of the target monitored object can be determined according to the status of the monitoring data of the target monitored object. Related monitored objects are personnel or equipment in the target open-pit mine area that are related to the target monitored object. The vector of the monitored object to be processed carries not only the action semantic features of the target monitored object but also the action semantic features of related monitored objects. This vector has an auxiliary analytical function for the action semantic features of the target monitored object, thus achieving higher accuracy analysis and improving the reliability of anomaly analysis in open-pit mines to a certain extent. This application solves the problem of relatively low reliability of anomaly analysis in open-pit mines in the prior art.
[0118] To improve the accuracy of inferring the behavioral state of the target monitored object, in one optional implementation, the association unit includes:
[0119] The association module performs preliminary association processing based on the location data of the target monitoring object and the location data of related monitoring objects to obtain an intermediate first object vector and an intermediate second object vector. The intermediate first object vector is the semantic information obtained by preliminary vectorization of the location data of the target monitoring object, and the intermediate second object vector is the semantic information obtained by preliminary vectorization of the location data of related monitoring objects. The semantic information is the behavioral status information of the target monitoring object in the target open-pit mine area.
[0120] Specifically, the location data of the target monitoring object and the location data of related monitoring objects are first vectorized to obtain intermediate object vectors (not final) of the target monitoring object and related monitoring objects, namely the first intermediate object vector and the second intermediate object vector.
[0121] The first determination module determines the corresponding category mapping parameters based on the target action related categories. The category mapping parameters are used to give different emphases to different semantic information according to different target action related categories. Target action related categories refer to the types of reasons that form action relatedness.
[0122] Specifically, different mappings may be required for different types of target actions, so that different information may be given different emphases or attention. Therefore, it is necessary to determine the category mapping parameters corresponding to the types of target actions in order to perform more accurate mappings.
[0123] The mapping module performs a linear mapping on the intermediate first object vector according to the category mapping parameters to form the initial first object vector of the target monitoring object, and performs a linear mapping on the intermediate second object vector according to the category mapping parameters to form the initial second object vector of the relevant monitoring object. The initial first object vector carries some semantic information of the relevant monitoring object, and the initial second object vector carries some semantic information of the target monitoring object.
[0124] Specifically, the category mapping parameters may include a first parameter and a second parameter. The first parameter can be used as a weight to weight the intermediate first object vector, and the second parameter can be used as a bias to superimpose the weighted results to obtain the initial first object vector. Similarly, for the initial second object vector, the first parameter can be used as a weight to weight the intermediate second object vector, and the second parameter can be used as a bias to superimpose the weighted results to obtain the initial second object vector.
[0125] The second determination module determines a first probability parameter based on the initial first object vector and a second probability parameter based on the initial second object vector. The first probability parameter is the category probability of the target action-related categories corresponding to the target monitored object, and the second probability parameter is the category probability of the target action-related categories corresponding to the relevant monitored object.
[0126] Specifically, a first probability parameter can be determined by processing the abnormal action type or probability of the target monitored object represented in the initial first object vector, and a second probability parameter can be determined by processing the abnormal action type or probability of the relevant monitored object represented in the initial second object vector.
[0127] The calculation module obtains a first related category vector based on the initial first object vector, the initial second object vector, and the second probability parameter, and obtains a second related category vector based on the initial second object vector, the initial first object vector, and the first probability parameter. The first related category vector represents the category probability of the target monitored object for the target action related category, and the second related category vector represents the category probability of the related monitored object for the target action related category.
[0128] Specifically, the association mining is first performed between the initial first object vector and the initial second object vector. When performing association mining on the initial first object vector based on the information in the initial second object vector, the possibility that the relevant monitored object belongs to the target action related category is considered. The higher the possibility, the higher the importance in association mining. The same applies to the target monitored object.
[0129] The propagation module propagates the second related category vector to the first related category vector to form at least one propagated related category vector. The propagated related category vector represents the probability that the target monitored object has relative to the target action related category. One target action related category corresponds to one propagated related category vector.
[0130] Specifically, when there is only one type of target action related to a given action, such as the action relatedness formed by the intersection of movement trajectories or the action relatedness formed by job type, a propagation relatedness type vector can be formed. When there are multiple types of target action related to a given action, such as the action relatedness formed by the intersection of movement trajectories and the action relatedness formed by job type, a corresponding propagation relatedness type vector can be obtained for each type of target action related to a given action.
[0131] The processing module aggregates the propagation-related type vectors to obtain the vector of the monitored object to be processed.
[0132] Specifically, when there is only one type of target action related to the target, the propagation related type vector can be directly used as the vector of the target monitoring object to be processed. When there are multiple types of target action related to the target, the propagation related type vectors corresponding to various target action related types are aggregated, such as by superposition, averaging, or splicing, to obtain the vector of the target monitoring object to be processed.
[0133] To facilitate further semantic and association analysis, in one optional implementation, the first association module includes:
[0134] The first calculation submodule obtains the first location dimension vector and the first identity dimension vector of the target monitoring object based on the location data and identity data corresponding to the target monitoring object. The first location dimension vector is used to represent the action semantic features of the target monitoring object, and the first identity dimension vector is used to represent the identity semantic features of the target monitoring object.
[0135] Specifically, vectorizing the location data of the target monitored object yields the first location dimension vector, and vectorizing the identity data yields the first identity dimension vector. Here, identity data can refer to the job type of the target monitored object, such as the work content of personnel or equipment.
[0136] The first aggregation submodule aggregates the first positioning dimension vector and the first identity dimension vector to obtain the intermediate first object vector.
[0137] Specifically, the first positioning dimension vector and the first identity dimension vector can be superimposed or concatenated to obtain the intermediate first object vector of the target monitoring object.
[0138] The mining submodule mines the second location dimension vector and the second identity dimension vector of the relevant monitored objects based on the location data and identity data of the relevant monitored objects. The second location dimension vector is used to represent the action semantic features of the relevant monitored objects, and the second identity dimension vector is used to represent the identity semantic features of the relevant monitored objects.
[0139] Specifically, vectorizing the location data of the relevant monitored objects yields a second location dimension vector, and vectorizing the identity data yields a second identity dimension vector. Here, identity data can refer to the job type of the relevant monitored objects, such as the work content of personnel or equipment.
[0140] The second aggregation submodule aggregates the second positioning dimension vector and the second identity dimension vector to obtain the intermediate second object vector.
[0141] Specifically, the second positioning dimension vector and the second identity dimension vector can be superimposed or concatenated to obtain the intermediate second object vector of the relevant monitored object.
[0142] To make the performance of the target monitoring object and related monitoring objects more accurate across different target action-related categories, in one optional implementation, the second determining module includes:
[0143] The first analysis submodule analyzes and obtains the intermediate third related category vector based on the initial first object vector, the reference probability parameter, and the initial second object vector. The reference probability parameter is a pre-configured parameter.
[0144] Specifically, when there is only one type of target action related to a given objective, the reference probability parameter can be equal to 1. When there are two types of target action related to a given objective, the reference probability parameter can be equal to 1 / 2, which is an average probability parameter. Multiplying the reference probability parameter by the initial second object vector and then adding it to the initial first object vector yields the intermediate related category vector, i.e., the intermediate third related category vector.
[0145] The second analysis submodule analyzes and obtains the intermediate fourth related category vector based on the initial second object vector, the reference probability parameter, and the initial first object vector.
[0146] Specifically, the reference probability parameter is multiplied by the initial first object vector, and then added to the initial second object vector to obtain the intermediate related category vector, which is the intermediate fourth related category vector.
[0147] The first processing submodule performs exponential operations and normalization on the fusion result of the initial first object vector and the intermediate third related category vector to obtain the first probability parameter.
[0148] Specifically, the fusion result of the initial first object vector and the intermediate third related category vector is subjected to exponential operation and normalization to obtain the first probability parameter that the target monitored object belongs to the target action related category.
[0149] The second processing submodule performs exponential operations and normalization on the fusion result of the initial second object vector and the intermediate fourth related category vector to obtain the second probability parameter.
[0150] Specifically, the fusion result of the initial second object vector and the intermediate fourth related category vector is subjected to exponential operation and normalization to obtain the second probability parameter of the relevant monitored object belonging to the target action related category. The first probability parameter and the second probability parameter are used to represent the probability of belonging to the corresponding target action related category.
[0151] To more accurately understand the relevant category and likelihood of the target monitored object, in one optional implementation, the calculation module includes:
[0152] The first calculation submodule performs a multiplication operation on the initial second object vector and the second possibility parameter to obtain the undetermined first object vector corresponding to the target monitoring object.
[0153] Specifically, the initial second object vector is multiplied by the second possibility parameter to output the corresponding undetermined first object vector. That is, the initial second object vector is weighted based on the second possibility parameter to apply the corresponding matching degree or importance to the initial second object vector.
[0154] The second calculation submodule performs a superposition operation on the initial first object vector and the undetermined first object vector to obtain the staged first object vector. The staged first object vector represents the action semantic features of the target monitoring object relative to the target action-related categories.
[0155] Specifically, the initial first object vector and the undetermined first object vector are superimposed. Based on the output data of the superposition operation, the phased first object vector of the target monitoring object relative to the target action related category is obtained. For example, the output data of the superposition operation can be used as the phased first object vector.
[0156] The third calculation submodule, based on the phased first object vector, obtains the first related category vector.
[0157] Specifically, a pre-configured vector propagation rule is first determined. This vector propagation rule is a type of vector propagation rule in graph deep learning models. Based on this vector propagation rule, the intermediate second object vector is propagated to the intermediate first object vector to form the corresponding object propagation vector. For example, the intermediate second object vector and the intermediate first object vector are added together (superposition operation) to obtain the object propagation vector. The object propagation vector is then added to the stage-specific first object vector to obtain the first relevant category vector of the target monitoring object relative to the target action-related category.
[0158] The fourth calculation submodule performs calculations on the initial first object vector and the first probability parameter to obtain the undetermined second object vector corresponding to the relevant monitored object.
[0159] Specifically, the initial first object vector and the first possibility parameter are multiplied to output the corresponding undetermined second object vector. That is, the initial first object vector is weighted based on the first possibility parameter to apply the corresponding matching degree or importance to the initial first object vector.
[0160] The fifth calculation submodule performs a superposition operation on the initial second object vector and the undetermined second object vector to obtain a staged second object vector.
[0161] Specifically, the initial second object vector and the undetermined second object vector are superimposed. Based on the output data of the superposition operation, a phased second object vector of the target monitoring object relative to the target action-related category is obtained. For example, the output data of the superposition operation can be used as the phased second object vector.
[0162] The sixth calculation submodule, based on the phased second object vector, obtains the second related category vector of the relevant monitored object relative to the target action related category.
[0163] Specifically, based on this vector propagation rule, the intermediate first object vector is propagated to the intermediate second object vector to form the corresponding object propagation vector. For example, the intermediate second object vector and the intermediate first object vector are added together (superposition operation) to obtain the object propagation vector; the object propagation vector is added together with the stage second object vector to obtain the second related category vector of the target monitoring object relative to the target action related category.
[0164] To quantify the degree of correlation and improve the comprehensiveness and accuracy of the analysis, in one optional implementation, the processing module includes:
[0165] The mapping submodule performs linear mapping on the first relevant category vector according to the category mapping parameters to obtain the first mapped object vector corresponding to the target monitoring object, and performs linear mapping on the second relevant category vector according to the category mapping parameters to obtain the second mapped object vector corresponding to the relevant monitoring object.
[0166] Specifically, based on the category mapping parameters corresponding to the target action-related categories and the first related category vector, the corresponding first mapping object vector is obtained; the category mapping parameters are linearly mapped to the first related category vector to obtain the first mapping object vector.
[0167] The seventh calculation submodule obtains the intermediate fifth related category vector based on the first mapping object vector, the reference probability parameter, and the second mapping object vector. The intermediate fifth related category vector is a new intermediate related category vector that the target monitoring object has relative to the target action related category.
[0168] Specifically, the second mapping object vector can be weighted (i.e. multiplied) based on the reference probability parameter. Then, the weighted result is superimposed with the first mapping object vector to obtain a new intermediate related category vector, namely the intermediate fifth related category vector. Compared with the initial first object vector, the intermediate first related category vector, and the intermediate third related category vector, the accuracy is higher and it can more accurately reflect the abnormal state of the target monitoring object. Each calculation of the category vector of the target monitoring object can improve the accuracy. Each target action related category has an intermediate related category vector.
[0169] The third processing submodule fuses the intermediate fifth related category vector and the second mapping object vector to obtain the fusion result, and then performs exponential operation and normalization on the fusion result to obtain the corresponding stage possibility parameters.
[0170] Specifically, the transpose of the second mapping object vector can be multiplied with the intermediate fifth related category vector, and then the result of the multiplication can be exponentially calculated to obtain the first exponential value. The transpose of the second mapping object vector can also be multiplied with the new intermediate related category vectors of each of the other target action related categories, and then the result of the multiplication can be exponentially calculated to obtain each second exponential value. Then, the ratio of the first exponential value to (the sum of the first exponential value and each second exponential value) can be calculated to obtain the stage-specific probability parameter. Here, "stage" can refer to a step in the cyclic or iterative processing. The current stage-specific probability parameter can be used as the final probability parameter, or it can be used as a reference probability parameter for cyclic iteration to obtain the final probability parameter.
[0171] The eighth calculation submodule obtains the intermediate sixth related category vector based on the first mapping object vector, the second mapping object vector, and the stage possibility parameter. The intermediate sixth related category vector is a new intermediate related category vector that the target monitoring object has relative to the target action related category.
[0172] Specifically, the first mapped object vector can be weighted (i.e. multiplied) based on the reference probability parameter. Then, the weighted result is superimposed on the second mapped object vector to obtain a new intermediate related category vector, namely the intermediate sixth related category vector. Compared with the initial second object vector, the intermediate second related category vector and the intermediate fourth related category vector, the accuracy is higher and it can more accurately reflect the abnormal state of the relevant monitored objects.
[0173] The propagation submodule propagates the second related category vector to the first related category vector according to the vector propagation rule, forming a fused related category vector. The vector propagation rule is a vector propagation rule in the graph deep learning model. The graph deep learning model includes graph distribution members corresponding to at least two monitoring objects; the at least two monitoring objects include the target monitoring object and related monitoring objects.
[0174] Specifically, according to the vector propagation rule, the second related category vector and the first related category vector can be added together to obtain the fused related category vector. When there are multiple second related category vectors, the multiple second related category vectors can be directly added together with the first related category vector to obtain the fused related category vector. Alternatively, the average of the multiple second related category vectors can be calculated first, and then the average vector can be added together with the first related category vector to obtain the fused related category vector.
[0175] A submodule is formed, which, based on the vector propagation rules, integrates the relevant category vectors and the intermediate sixth relevant category vector, to form a propagation-related category vector.
[0176] Specifically, the fused relevant category vector and the new relevant category vector can be added together to obtain the propagation relevant category vector.
[0177] In order to accurately identify the types of abnormal behavior of the target monitored object, in one optional implementation, the step analysis unit includes:
[0178] The analysis module uses a graph deep learning model to perform object analysis on the vector of the monitored object to be processed, and obtains the target probability parameter distribution. The target probability parameter distribution is the probability that the monitored object belongs to each cause category in the target action related categories.
[0179] Specifically, graph deep learning models may include multilayer perceptrons for calculating propagation-related class vectors, thereby outputting multiple target probability parameters, i.e., obtaining the target probability parameter distribution.
[0180] The third determination module identifies the type of action anomaly corresponding to the maximum value in the distribution of target probability parameters as the target object monitoring data.
[0181] Specifically, in the distribution of target probability parameters, the target probability parameter with the maximum value is determined; based on the type of action anomaly corresponding to the target probability parameter with the maximum value, the target object monitoring data corresponding to the target monitoring object is determined. For example, the target probability parameter for the type of action anomaly entering the dangerous area is 0.7, and the target probability parameter for the type of action anomaly not entering the dangerous area is 0.2. In this way, the type of action anomaly entering the dangerous area can be used as the target object monitoring data.
[0182] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.
[0183] The anomaly analysis device for 5G high-precision positioning in open-pit mines includes a processor and a memory. Acquisition units, correlation units, analysis units, and determination units are all stored as program units in the memory. The processor executes these program units to achieve their respective functions. Modules can all reside in the same processor; alternatively, modules can be located in different processors in any combination.
[0184] The processor contains a kernel, which retrieves the corresponding program units from memory. One or more kernels can be configured, and adjusting kernel parameters can address the relatively low reliability of current open-pit mine anomaly analysis technologies.
[0185] The memory may include non-permanent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.
[0186] This invention provides a computer-readable storage medium including a stored program, wherein, when the program is running, it controls the device where the computer-readable storage medium is located to perform anomaly analysis method for 5G high-precision positioning in open-pit mines.
[0187] This invention also provides a 5G high-precision positioning system for open-pit mines. The 5G high-precision positioning system for open-pit mines includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements at least the steps of an anomaly analysis method for 5G high-precision positioning of open-pit mines.
[0188] This application also provides a computer program product that, when executed on a data processing device, is suitable for performing steps of an anomaly analysis method for initializing 5G high-precision positioning in open-pit mines.
[0189] It will be apparent to those skilled in the art that the modules or steps of this invention can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. They can be implemented using computer-executable program code, and thus can be stored in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, this invention is not limited to any particular combination of hardware and software.
[0190] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0191] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0192] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0193] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0194] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.
[0195] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.
[0196] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.
[0197] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.
[0198] As can be seen from the above description, the embodiments of this application achieve the following technical effects:
[0199] 1) The anomaly analysis method for 5G high-precision positioning in open-pit mines disclosed in this application first acquires object representation data for each monitored object within the open-pit mine area. Then, it performs correlation mining on the positioning data of the target monitored object and related monitored objects to obtain the vector of the monitored object to be processed corresponding to the target monitored object. Based on the vector of the monitored object to be processed, the monitoring data of the target monitored object corresponding to the target monitored object is analyzed. The abnormal state of the target monitored object can be determined based on the status of the target monitored object's monitoring data. Related monitored objects are personnel or equipment in the target open-pit mine area that are related to the target monitored object. The vector of the monitored object to be processed carries not only the action semantic features of the target monitored object but also the action semantic features of related monitored objects, which has an auxiliary analytical effect on the action semantic features of the target monitored object. Therefore, higher precision analysis can be achieved, thereby improving the reliability of anomaly analysis in open-pit mines to a certain extent. This application solves the problem of relatively low reliability in existing technologies for anomaly analysis in open-pit mines.
[0200] 2) The anomaly analysis device for 5G high-precision positioning in open-pit mines disclosed in this application first acquires object representation data for each monitored object within the open-pit mine area. Then, it performs correlation mining on the positioning data of the target monitored object and related monitored objects to obtain the vector of the monitored object to be processed corresponding to the target monitored object. Based on the vector of the monitored object to be processed, the monitoring data of the target monitored object corresponding to the target monitored object is analyzed. The abnormal state of the target monitored object can be determined based on the status of the target monitored object's monitoring data. Related monitored objects are personnel or equipment in the target open-pit mine area that are related to the target monitored object. The vector of the monitored object to be processed carries not only the action semantic features of the target monitored object but also the action semantic features of related monitored objects, which has an auxiliary analytical effect on the action semantic features of the target monitored object. Therefore, higher precision analysis can be achieved, thereby improving the reliability of anomaly analysis in open-pit mines to a certain extent. This application solves the problem of relatively low reliability in existing technologies for anomaly analysis in open-pit mines.
[0201] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.
Claims
1. An anomaly analysis method for 5G high-precision positioning in open-pit mines, characterized in that, include: Obtain object representation data corresponding to all monitored objects within the target open-pit mine area. Any one of the monitored objects belongs to a person or a piece of equipment within the target open-pit mine area. The object representation data includes at least location data, which includes the location information of the monitored object at at least one monitoring time point. The location data of the target monitoring object and the location data of related monitoring objects are correlated and mined to obtain the vector of monitoring objects to be processed corresponding to the target monitoring object. The related monitoring objects and the target monitoring object have an action correlation in the target open-pit mine area. The vector of monitoring objects to be processed includes the action semantic features of the target monitoring object and the target monitoring object. The related monitoring objects are personnel or equipment that are related to the target monitoring object in the target open-pit mine area. The vector of the monitored object to be processed is analyzed to obtain the target object monitoring data corresponding to the target monitored object. The target object monitoring data is used to reflect the type or probability of abnormal behavior of the target monitored object in the target open-pit mine area. Based on the monitoring data of the target object, determine the abnormal state of the target monitored object and issue a prompt message to handle the abnormality. Analyzing the vector of the monitoring objects to be processed to obtain the target object monitoring data corresponding to the target monitoring object includes: using a graph deep learning model to perform object analysis on the vector of the monitoring objects to be processed to obtain the target probability parameter distribution, wherein the target probability parameter distribution is the probability that the target monitoring object belongs to each cause category in the target action related categories; The type of action anomaly corresponding to the maximum value in the distribution of the target probability parameters is determined as the monitoring data of the target object.
2. The method according to claim 1, characterized in that, The location data of the target monitoring object and the location data of related monitoring objects are correlated and analyzed to obtain the vector of monitoring objects to be processed corresponding to the target monitoring object, including: Based on the location data of the target monitoring object and the location data of related monitoring objects, preliminary association processing is performed to obtain an intermediate first object vector and an intermediate second object vector. The intermediate first object vector is semantic information obtained by preliminary vectorization of the location data of the target monitoring object, and the intermediate second object vector is semantic information obtained by preliminary vectorization of the location data of the related monitoring objects. The semantic information is the behavioral status information of the target monitoring object in the target open-pit mine area. The corresponding category mapping parameter is determined based on the target action related category. The category mapping parameter is used to give different emphases to different semantic information according to different target action related categories. The target action related category refers to the type of cause that forms the action relatedness. The intermediate first object vector is linearly mapped according to the category mapping parameter to form the initial first object vector of the target monitoring object, and the intermediate second object vector is linearly mapped according to the category mapping parameter to form the initial second object vector of the related monitoring object. The initial first object vector carries part of the semantic information of the related monitoring object, and the initial second object vector carries part of the semantic information of the target monitoring object. A first probability parameter is determined based on the initial first object vector, and a second probability parameter is determined based on the initial second object vector. The first probability parameter is the category probability of the target action-related category corresponding to the target monitoring object, and the second probability parameter is the category probability of the target action-related category corresponding to the related monitoring object. Based on the initial first object vector, the initial second object vector, and the second probability parameter, a first related category vector is obtained, and based on the initial second object vector, the initial first object vector, and the first probability parameter, a second related category vector is obtained. The first related category vector represents the category probability of the target monitored object for the target action related category, and the second related category vector represents the category probability of the related monitored object for the target action related category. The second related category vector is propagated to the first related category vector to form at least one propagated related category vector. The propagated related category vector represents the probability that the target monitored object has relative to the target action related category. One propagated related category corresponds to one propagated related category. The propagation-related category vectors are aggregated to obtain the vector of the monitoring object to be processed.
3. The method according to claim 2, characterized in that, The object representation data also includes identity data, which is identification information related to the monitored object. Preliminary association processing is performed based on the location data of the target monitored object and the location data of related monitored objects to obtain an intermediate first object vector and an intermediate second object vector, including: Based on the location data and identity data corresponding to the target monitoring object, a first location dimension vector and a first identity dimension vector of the target monitoring object are obtained. The first location dimension vector is used to characterize the action semantic features of the target monitoring object, and the first identity dimension vector is used to characterize the identity semantic features of the target monitoring object. By aggregating the first positioning dimension vector and the first identity dimension vector, the intermediate first object vector is obtained; Based on the location data and identity data corresponding to the relevant monitored objects, a second location dimension vector and a second identity dimension vector of the relevant monitored objects are mined. The second location dimension vector is used to characterize the action semantic features of the relevant monitored objects, and the second identity dimension vector is used to characterize the identity semantic features of the relevant monitored objects. The second positioning dimension vector and the second identity dimension vector are aggregated to obtain the intermediate second object vector.
4. The method according to claim 2, characterized in that, Based on the initial first object vector, the initial second object vector, and the second probability parameter, a first related category vector is obtained, and based on the initial second object vector, the initial first object vector, and the first probability parameter, a second related category vector is obtained, including: Multiply the initial second object vector and the second possibility parameter to obtain the undetermined first object vector corresponding to the target monitoring object; The initial first object vector and the undetermined first object vector are superimposed to obtain a phased first object vector, wherein the phased first object vector is the action semantic feature of the target monitoring object relative to the target action-related category; Based on the aforementioned phased first object vector, the first related category vector is obtained; The initial first object vector and the first probability parameter are calculated to obtain the undetermined second object vector corresponding to the relevant monitored object; The initial second object vector and the undetermined second object vector are superimposed to obtain a staged second object vector. Based on the phased second object vector, a second related category vector is obtained for the relevant monitored object relative to the target action related category.
5. The method according to claim 2, characterized in that, Determining a first possibility parameter based on the initial first object vector, and determining a second possibility parameter based on the initial second object vector, includes: Based on the initial first object vector, the reference probability parameter, and the initial second object vector, an intermediate third related category vector is obtained through analysis. The reference probability parameter is a pre-configured parameter. Based on the initial second object vector, the reference probability parameter, and the initial first object vector, the intermediate fourth related category vector is obtained through analysis. The first probability parameter is obtained by performing exponential operation and normalization on the fusion result of the initial first object vector and the intermediate third related category vector. The fusion result of the initial second object vector and the intermediate fourth related category vector is subjected to exponential operation and normalization to obtain the second probability parameter.
6. The method according to claim 2, characterized in that, Propagating the second relevant category vector to the first relevant category vector to form at least one propagated relevant category vector includes: The first related category vector is linearly mapped according to the category mapping parameters to obtain the first mapped object vector corresponding to the target monitoring object, and the second related category vector is linearly mapped according to the category mapping parameters to obtain the second mapped object vector corresponding to the related monitoring object. Based on the first mapping object vector, the reference probability parameter, and the second mapping object vector, an intermediate fifth related category vector is obtained. The intermediate fifth related category vector is a new intermediate related category vector that the target monitoring object has relative to the target action related category. The intermediate fifth related category vector and the second mapping object vector are fused to obtain a fusion result, and the fusion result is sequentially subjected to exponential operation and normalization to obtain the corresponding stage possibility parameter; Based on the first mapping object vector, the second mapping object vector, and the staged probability parameter, an intermediate sixth related category vector is obtained. The intermediate sixth related category vector is a new intermediate related category vector that the target monitoring object has relative to the target action related category. The second related category vector is propagated to the first related category vector according to the vector propagation rule to form a fused related category vector. The vector propagation rule is a vector propagation rule in a graph deep learning model. The graph deep learning model includes graph distribution members corresponding to at least two of the monitored objects. The at least two monitored objects include the target monitored object and the related monitored object. Based on the vector propagation rule, the fusion-related category vector, and the intermediate sixth related category vector, the propagation-related category vector is formed.
7. An anomaly analysis device for 5G high-precision positioning in open-pit mines, characterized in that, The device includes: The acquisition unit is used to acquire object representation data corresponding to all monitored objects in the target open-pit mine area. Any one of the monitored objects belongs to a person or a piece of equipment in the target open-pit mine area. The object representation data includes at least location data, and the location data includes the location information of the monitored object at at least one monitoring time point. The association unit is used to perform association mining on the location data of the target monitoring object and the location data of related monitoring objects to obtain the vector of monitoring objects to be processed corresponding to the target monitoring object. The related monitoring objects and the target monitoring object have an action correlation in the target open-pit mine area. The vector of monitoring objects to be processed includes the action semantic features of the target monitoring object and the target monitoring object. The related monitoring objects are personnel or equipment that are related to the target monitoring object in the target open-pit mine area. The analysis unit is used to analyze the vector of the monitoring object to be processed to obtain the target object monitoring data corresponding to the target monitoring object. The target object monitoring data is used to reflect the type or probability of abnormal behavior of the target monitoring object in the target open-pit mine area. The determining unit is used to determine the abnormal state of the target monitored object based on the target object monitoring data and issue a prompt message to perform abnormal handling. The analysis unit includes: an analysis module, used to perform object analysis on the vector of the monitored object to be processed using a graph deep learning model to obtain a target probability parameter distribution, wherein the target probability parameter distribution is the probability that the monitored object belongs to each cause category in the target action-related categories; and a third determination module, used to determine the action anomaly category corresponding to the maximum value in the target probability parameter distribution as the monitoring data of the target object.
8. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein, when the program is executed, it controls the device on which the computer-readable storage medium is located to perform the method according to any one of claims 1 to 6.
9. A computer program product comprising computer instructions, characterized in that, When executed by a processor, the computer instructions implement the method described in any one of claims 1 to 6.
Citation Information
Patent Citations
Intrusion detection method, system, equipment and medium
CN115346157A
Information output method and system based on 5G communication navigation
CN117807393A
