Major accident hazard identification, analysis and judgment system and method based on AI large model

Abstract

The present invention provides a system and method for identifying, analyzing, and determining major accident hazards based on a large AI model, wherein the system includes: an information acquisition subsystem for acquiring information to be identified in a detection area; a model training subsystem for training an AI large model for identifying accident hazards; a knowledge graph construction subsystem for constructing an industry-specific knowledge graph; and a hidden danger identification subsystem for performing identification, analysis, and determination of major accident hazards based on the AI ​​large model for identifying accident hazards and the industry-specific knowledge graph and the information to be identified. The system and method for identifying, analyzing, and determining major accident hazards based on a large AI model of the present invention trains an AI large model for identifying accident hazards and constructs an industry-specific knowledge graph; after acquiring the information to be identified, the model automatically extracts effective trigger words through natural language processing technology combined with the industry-specific knowledge graph, optimizes hidden danger inference rules, and automatically performs hidden danger inference, thereby improving the accuracy of hidden danger identification.

Description

Technical Field

[0001] The present invention relates to the field of AI large model technology, and in particular to a system and method for identifying, analyzing, and determining major accident hazards based on AI large models. Background Art

[0002] Currently, the criteria for determining major accident hazards are simplistic and fail to effectively cover complex safety risks. Furthermore, the methods used are relatively primitive, relying primarily on manual judgment by experts, and are unable to effectively address increasingly complex safety hazards. Inconsistent inspection methods and standards, as well as uneven personnel quality, lead to significant blind spots and errors in manual inspections, impacting the accuracy and effectiveness of hazard detection.

[0003] In view of this, there is an urgent need for major accident hazard identification, analysis and judgment systems and methods based on large AI models to at least address the above-mentioned deficiencies. Summary of the Invention

[0004] One of the purposes of the present invention is to provide a major accident hazard identification, analysis and judgment system and method based on an AI big model, train the accident hazard identification AI big model, and at the same time, build an industry knowledge graph based on historical accident data, expert experience and safety laws, regulations and standards in the detection area; after the accident hazard identification AI big model obtains the information to be identified, it automatically extracts effective trigger words through natural language processing technology combined with the industry knowledge graph, optimizes the hazard inference rules and automatically performs hazard inference, thereby avoiding manual investigation errors and improving the accuracy of hazard identification.

[0005] The major accident hazard identification, analysis, and determination system based on an AI large model provided by an embodiment of the present invention includes:

[0006] The information to be identified acquisition subsystem is used to acquire the information to be identified in the detection area;

[0007] Model training subsystem, used to train large AI models for accident hazard identification;

[0008] Knowledge graph construction subsystem, used to build industry knowledge graphs;

[0009] The hidden danger identification subsystem is used to identify, analyze and judge major accident hazards based on the accident hazard identification AI large model and industry knowledge graph and the information to be identified.

[0010] Preferably, the knowledge graph construction subsystem constructs an industry domain knowledge graph, including:

[0011] Obtain historical accident data, expert experience, and safety laws, regulations, and standards;

[0012] Based on knowledge graph construction technology, according to historical accident data, expert experience and safety laws, regulations and standards, the key factors and correlations of major accident hazards are identified and mapped to obtain industry knowledge graphs.

[0013] Preferably, the hidden danger identification subsystem is based on the accident hidden danger identification AI large model and industry domain knowledge graph, and performs major accident hidden danger identification, analysis and judgment according to the information to be identified, including:

[0014] Based on natural language processing technology and the information to be identified, effective trigger words are extracted from the industry knowledge graph;

[0015] Optimize the hidden danger reasoning rules of the accident hidden danger identification AI model based on the graph features of effective trigger words in the industry knowledge graph.

[0016] Preferably, the hidden danger identification subsystem is based on the accident hidden danger identification AI large model and industry domain knowledge graph, and performs major accident hidden danger identification, analysis, and judgment based on the information to be identified, and also includes:

[0017] Perform image recognition analysis on the information to be identified to obtain identification image features;

[0018] Match the identified image features with the suspicious hidden danger feature library obtained from the industry knowledge graph. If a match exists, obtain the hidden danger description corresponding to the matching suspicious hidden danger feature;

[0019] Continue to search the industry knowledge graph for the criteria for determining the corresponding hidden danger type based on the hidden danger description;

[0020] Identify, analyze and determine major accident hazards based on hazard descriptions and determination criteria.

[0021] Preferably, the to-be-identified information acquisition subsystem acquires the to-be-identified information of the detection area, further comprising:

[0022] Get the region type label of the detection region;

[0023] Retrieve accident events based on area type labels;

[0024] According to the accident event, obtain the basis for determining the target to be identified;

[0025] Obtain regional operation information of the detection area;

[0026] According to the regional operation information and the basis for determining the target to be identified, the information to be identified in the detection area is obtained.

[0027] Preferably, the subsystem for obtaining information to be identified obtains the basis for determining the target to be identified based on the accident event, including:

[0028] Analyze the accident event and obtain the first regional task and the first manual analysis factor that are executed when the accident event occurs; the first manual analysis factor includes: the type of accident event, the analysis object of the subsequent manual analysis leading to the accident event, and the analysis value of the analysis object;

[0029] Aggregating the first manual analysis factors of the first regional tasks of the same task type to obtain a second manual analysis factor;

[0030] Divide the second artificial analysis factors containing the same accident event type into the same analysis group;

[0031] Get the frequency of occurrence of the analysis object in the analysis group;

[0032] Traverse the analysis objects in descending order of frequency of occurrence, and take the currently traversed analysis object as the target analysis object;

[0033] Get the number of analysis values ​​of the target analysis object;

[0034] If the number of analysis values ​​is greater than or equal to the preset number threshold, the standard deviation of the analysis value of the target analysis object is directly obtained;

[0035] If the number of analysis values ​​is less than a preset number threshold, determining a reference analysis object that co-occurs with the target analysis object in the same analysis group;

[0036] Based on the transfer learning algorithm and the collaborative relationship between the analysis values ​​of the reference analysis object and the target analysis object, an expanded analysis value of the target analysis object is obtained, and then the standard deviation of the analysis value of the target analysis object is obtained based on the analysis value of the target analysis object and the expanded analysis value;

[0037] If the standard deviation is less than or equal to the preset standard deviation threshold, the first regional task, the accident event type, the target analysis object, and the mean of the analysis values ​​corresponding to the target analysis object are associated and used as the basis for determining the target to be identified.

[0038] Preferably, the to-be-identified information acquisition subsystem acquires the to-be-identified information of the detection area according to the regional operation information and the basis for determining the to-be-identified target, including:

[0039] Analyze regional operation information and obtain second-region tasks;

[0040] Match the second area task with the first area task in the basis for determining the target to be identified. If they are consistent, the corresponding first area task is used as the third area task;

[0041] Based on the target recognition technology, the target analysis object corresponding to the third area task of the detection area is detected to obtain the information to be recognized.

[0042] Preferably, the subsystem for acquiring information to be identified detects the target analysis object corresponding to the third area task of the detection area based on target recognition technology to obtain the information to be identified, including:

[0043] Determine the regional location of the fixed target analysis object corresponding to the third regional task;

[0044] Obtaining a first observation condition of a first analysis value mean of a fixed target analysis object and associating it with a regional position;

[0045] Planning the mobile identification device to move to a regional position closest to the mobile identification device in the execution area of ​​the third regional task, and marking the regional position as the initial position;

[0046] Plan the main observation line based on the distribution of the initial position and regional position in the execution area;

[0047] Marking the first observation condition associated with the regional position on the observation main line to generate first pilotage information;

[0048] Controlling the mobile identification device to move and shoot based on the first pilotage information;

[0049] updating the remaining pilotage information of the current area location according to the captured image of the current area location, and controlling the movement and capturing of the mobile identification device based on the updated pilotage information;

[0050] Repeat the steps of: updating the remaining navigation information of the current area according to the captured image of the current area; and controlling the movement and capturing of the mobile identification device based on the updated navigation information;

[0051] When all the area locations are photographed, the acquisition of the information to be identified is completed.

[0052] Preferably, updating the remaining pilotage information of the current area location according to the captured image of the current area location includes:

[0053] Detecting, based on the captured image, a target analysis object corresponding to a third area task in the detection area to obtain a first detection target;

[0054] Acquire a first detection target that appears in the same captured image and has not acquired information to be identified, and use the first detection target as a second detection target;

[0055] If the acquisition is successful, the second observation condition is obtained according to the second analysis value mean corresponding to the second detection target; the second observation condition includes: observation point position and observation posture;

[0056] Calculate the real-time distance between the observation point of the mobile identification device and the second detection target based on the captured image;

[0057] Plan the observation order of the second detection target according to the order of real-time distance from small to large;

[0058] Determine the second pilotage information based on the observation sequence and observation posture;

[0059] The second pilotage information is placed in front of the remaining pilotage information to obtain updated pilotage information.

[0060] The method for identifying, analyzing, and determining major accident hazards based on an AI large model provided by an embodiment of the present invention includes:

[0061] Step 1: Obtain the information to be identified in the detection area;

[0062] Step 2: Train the AI ​​model for identifying potential accidents;

[0063] Step 3: Build industry domain knowledge graph;

[0064] Step 4: Based on the accident hazard identification AI big model and industry knowledge graph, major accident hazards are identified, analyzed, and determined according to the information to be identified.

[0065] The beneficial effects of the present invention are:

[0066] The present invention trains an AI big model for accident hazard identification, and at the same time, combines historical accident data, expert experience, and safety laws, regulations, and standards in the detection area to construct an industry knowledge graph; after the AI ​​big model for accident hazard identification obtains the information to be identified, it automatically extracts effective trigger words through natural language processing technology combined with the industry knowledge graph, optimizes hazard inference rules, and automatically performs hazard inference, avoiding manual investigation errors and improving the accuracy of hazard identification.

[0067] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the present invention. The purpose and other advantages of the present invention can be achieved and obtained through the structures specifically pointed out in this application document.

[0068] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0070] Figure 1 Schematic diagram of a major accident hazard identification, analysis, and determination system based on an AI large model in an embodiment of the present invention;

[0071] Figure 2Schematic diagram of a method for identifying, analyzing, and determining major accident hazards based on an AI large model in an embodiment of the present invention. DETAILED DESCRIPTION

[0072] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0073] The embodiment of the present invention provides a major accident hazard identification, analysis and judgment system based on AI large model, such as Figure 1 Shown, including:

[0074] The to-be-identified information acquisition subsystem 1 is used to acquire the to-be-identified information of the detection area;

[0075] The information to be identified acquisition subsystem acquires the information to be identified in the detection area, including:

[0076] Obtain information to be identified through mobile terminals, inspection robots, and drones;

[0077] Model training subsystem 2, used to train the large AI model for accident hazard identification;

[0078] Knowledge graph construction subsystem 3, used to construct industry knowledge graphs;

[0079] The knowledge graph construction subsystem constructs an industry domain knowledge graph, including:

[0080] Obtain historical accident data, expert experience, and safety laws, regulations, and standards;

[0081] Based on knowledge graph construction technology, we identify and map the key factors and correlations of major accident hazards based on historical accident data, expert experience, and safety laws, regulations, and standards, and obtain industry-specific knowledge graphs.

[0082] Hidden danger identification subsystem 4 is used to identify, analyze, and determine major accident hazards based on the information to be identified, using the accident hazard identification AI large model and industry knowledge graph;

[0083] The hidden danger identification subsystem is based on the accident hidden danger identification AI large model and industry domain knowledge graph, and performs major accident hidden danger identification, analysis, and judgment based on the information to be identified, including:

[0084] Based on natural language processing technology and the information to be identified, effective trigger words are extracted from the industry knowledge graph;

[0085] Optimize the hidden danger reasoning rules of the accident hidden danger identification AI model based on the graph features of effective trigger words in the industry knowledge graph.

[0086] In this embodiment, the detection area is an area where accident potential hazards need to be checked, such as a construction area of ​​a substation or a construction site.

[0087] In this embodiment, the information to be identified is relevant information used to determine whether there is a potential accident hazard, such as video data and image data of the detection area. The information to be identified can be obtained by photographing the detection area using a device equipped with a camera, such as a mobile vehicle or an inspection robot.

[0088] In this embodiment, the accident hazard identification AI model is: an AI model obtained by learning the artificial reasoning and analysis records of accident hazards based on deep learning technology, and is used to analyze the accident hazard risk situation in the detection area based on the input analysis basis (such as: hazard analysis images, etc.).

[0089] In this embodiment, the historical accident data is a record of accidents that have occurred in the detection area in the past;

[0090] In this embodiment, the effective trigger words are: first, the information to be identified is described in words to obtain a description word, and then the description word is matched with the entity description words and relationship description words in the industry knowledge graph to obtain a consistent matching entity description word or relationship description word;

[0091] In this embodiment, the graph features of the effective trigger words in the industry domain knowledge graph are: entities or relationships related to the entities or relationships in the industry domain knowledge graph corresponding to the effective trigger words.

[0092] The working principle and beneficial effects of the above technical solution are:

[0093] The present invention trains an AI big model for accident hazard identification, and at the same time, combines historical accident data, expert experience, and safety laws, regulations, and standards in the detection area to construct an industry knowledge graph; after the AI ​​big model for accident hazard identification obtains the information to be identified, it automatically extracts effective trigger words through natural language processing technology combined with the industry knowledge graph, optimizes hazard inference rules, and automatically performs hazard inference, avoiding manual investigation errors and improving the accuracy of hazard identification.

[0094] In one embodiment, the hazard identification subsystem performs analysis and determination of major accident hazards based on the information to be identified, based on the accident hazard identification AI large model and industry domain knowledge graph, and also includes:

[0095] Perform image recognition analysis on the information to be identified to obtain identification image features; wherein the identification image features include: texture, color, etc. of the image in the information to be identified;

[0096] The identified image features are matched with a library of suspected hidden danger features obtained from the industry knowledge graph. If a match is found, the hidden danger description corresponding to the matched suspected hidden danger features is obtained. The suspected hidden danger feature library includes: image features extracted from accident images that may cause hidden dangers, such as image features of worn safety ropes; hidden danger descriptions are, for example, "the safety rope is worn and may break."

[0097] Continue to search the industry knowledge graph for the criteria for determining the corresponding hazard type based on the hazard description. The hazard type corresponding to the hazard description is: the type of hazard described in the hazard description. For example, if the hazard description is "the safety rope may break," the corresponding hazard type is: safety rope break. The determination criteria are: the overall basis for determining the hazard type, such as the degree of wear and tear and the number of load-bearing objects that determine the risk of a safety rope break.

[0098] Identify, analyze and determine major accident hazards based on hazard descriptions and determination criteria.

[0099] The working principle and beneficial effects of the above technical solution are:

[0100] The present invention performs image recognition analysis on the identification information to obtain identification image features; connects to the suspicious hidden danger feature library obtained from the industry field knowledge graph, matches the identification image features with the suspicious hidden danger features in the suspicious hidden danger feature library, and determines the hidden danger description corresponding to the matched suspicious hidden danger features; after obtaining the hidden danger description, the hidden danger type can be determined, and then the judgment standard corresponding to the hidden danger type can be searched from the industry field knowledge graph, and then the hidden danger description and the judgment standard are compared. If the standard is met, the suspicious hidden danger is confirmed and output, otherwise it is determined that there is no hidden danger, the hidden danger reasoning rules are optimized and the hidden danger reasoning is automatically performed, which is more intelligent.

[0101] In one embodiment, the to-be-identified information acquisition subsystem acquires the to-be-identified information of the detection area, including:

[0102] Obtain the information to be identified through mobile terminals, inspection robots and drones.

[0103] The working principle and beneficial effects of the above technical solution are:

[0104] The present invention introduces multiple service modes to provide a major accident hidden danger identification function, thereby improving the comprehensiveness of the identification.

[0105] In one embodiment, the to-be-identified information acquisition subsystem acquires the to-be-identified information of the detection area, further comprising:

[0106] Get the region type label of the detection area; the region type label is: an ID that identifies the category of the detection area, such as "substation", "automobile manufacturing plant", etc.

[0107] Retrieve accident events based on the area type label. Accident events are historical anomalies and accidents that occurred within the area of ​​this type, such as workers being electrocuted or transformer equipment being burned.

[0108] According to the accident event, obtain the basis for determining the target to be identified;

[0109] Among them, according to the accident event, the basis for determining the target to be identified is obtained, including:

[0110] Analyze the accident event to obtain the first regional task and the first manual analysis factor that were executed when the accident event occurred. The first manual analysis factor includes: the type of accident event, the analysis object that caused the accident event in the manual subsequent analysis, and the analysis value of the analysis object. The first regional task is the task that was being executed in the area where the accident event occurred, such as "substation equipment maintenance". The accident event type is: the type of accident event, such as "electric shock". The analysis object is: the analysis target in the area corresponding to the first regional task when manually attributing the accident event post-event, such as: staff members, equipment under maintenance. The analysis value of the analysis object is: the state of the analysis object that caused the accident when manually attributing the accident event post-event, such as: improper operation of the staff (for example, not wearing insulating gloves) or the state of the internal components of the equipment under maintenance that caused leakage in the equipment under maintenance (for example, damaged insulation at the joint of line A and line B).

[0111] Aggregating the first manual analysis factors of the first regional tasks of the same task type to obtain a second manual analysis factor;

[0112] Divide the second artificial analysis factors containing the same accident event type into the same analysis group;

[0113] Obtaining the frequency of occurrence of the analysis object in the analysis group; the frequency of occurrence is: the number of second artificial analysis factors present in the analysis object in the same analysis group divided by the number of all second artificial analysis factors in the analysis group;

[0114] Traverse the analysis objects in descending order of frequency of occurrence, and take the currently traversed analysis object as the target analysis object;

[0115] Get the number of analysis values ​​of the target analysis object;

[0116] If the number of analysis values ​​is greater than or equal to a preset number threshold, the standard deviation of the analysis value of the target analysis object is directly obtained; wherein the preset number threshold is manually set in advance;

[0117] If the number of analysis values ​​is less than a preset number threshold, determining reference analysis objects that co-occur with the target analysis object in the same analysis group; the reference analysis objects that co-occur with the target analysis object are: other analysis objects that co-occur with the target analysis object in the second manual analysis factor in the same analysis group;

[0118] Based on the transfer learning algorithm and the synergistic relationship between the analysis values ​​of the reference analysis object and the target analysis object, the expanded analysis value of the target analysis object is obtained, and then the standard deviation of the analysis value of the target analysis object is obtained based on the analysis value of the target analysis object and the expanded analysis value; the synergistic relationship of the analysis values ​​is: the correlation between the analysis value of the reference analysis object and the analysis value of the target analysis object, such as: the relationship between the steps of manually adding insulating oil and the content of insulating oil leaked in the environment; the expanded analysis value is: using the transfer learning algorithm to learn the synergistic relationship between the analysis values ​​of the reference analysis object and the target analysis object and applying it to a second artificial analysis factor that does not contain the target analysis object but contains the reference analysis object, to predict the state of the accident caused by the target analysis object omitted from the analysis in the corresponding second artificial analysis factor;

[0119] If the standard deviation is less than or equal to the preset standard deviation threshold, the first regional task, the accident event type, the target analysis object, and the mean of the analysis values ​​corresponding to the target analysis object are associated and used as the basis for determining the target to be identified; the preset standard deviation threshold is set manually;

[0120] Obtain regional operation information of the detection area; regional operation information includes: the current execution task type and execution location of the detection area;

[0121] Obtain the information to be identified in the detection area based on the regional operation information and the basis for determining the target to be identified;

[0122] Among them, according to the regional operation information and the basis for determining the target to be identified, the information to be identified in the detection area is obtained, including:

[0123] Analyze regional operation information and obtain the second regional task; the second regional task is to detect the tasks currently being performed in the area, such as transformer inspection;

[0124] Match the second area task with the first area task in the basis for determining the target to be identified. If they are consistent, the corresponding first area task is used as the third area task;

[0125] Based on the target recognition technology, the target analysis object corresponding to the third area task of the detection area is detected to obtain the information to be recognized.

[0126] The working principle and beneficial effects of the above technical solution are:

[0127] Different types of detection areas have different types of accident events. The present invention first searches for historical accident events in the same area type based on the area type label, and extracts the basis for determining the target to be identified from the accident events. The specific extraction process is as follows:

[0128] The first regional task and the first manual analysis factor in the accident event are analyzed and obtained, the first manual analysis factors of the same task type are aggregated, and the second manual analysis factors containing the same accident event type are divided into the same analysis group; however, not all manual analysis factors are credible, therefore, the analysis results of different analysts on the same type of analysis task can be obtained and compared. Here, the standard deviation is used to characterize the difference in the analysis results of different analysts. However, when the analysis result data of different analysts on the same type of analysis task is insufficient (that is, the number of analysis values ​​is less than the number threshold), the standard deviation is not accurate enough to characterize the difference in the analysis results of different analysts. Considering the omission of analysis by analysts in the analysis data of the accident event, the transfer learning algorithm and the existing data are used to obtain the expanded analysis value of the target analysis object, and then the standard deviation of the analysis value of the target analysis object is obtained based on the analysis value of the target analysis object and the expanded analysis value;

[0129] The standard deviation is compared with the introduced standard deviation threshold. If the standard deviation is less than or equal to the preset standard deviation threshold, the first area task, accident event type, target analysis object and the mean of the analysis value corresponding to the target analysis object are associated and used as the basis for determining the target to be identified.

[0130] After determining the basis for determining the target to be identified, the second regional task is determined based on the regional operation information, and the second regional task is matched with the first regional task in the basis for determining the target to be identified. If they are consistent, the target analysis object corresponding to the first regional task is used as the target to be identified; the target to be identified in the detection area is detected based on the target recognition technology for subsequent acquisition of the information to be identified, thereby improving the recognition efficiency.

[0131] In one embodiment, the to-be-identified information acquisition subsystem detects the target analysis object corresponding to the third area task of the detection area based on target recognition technology to obtain the to-be-identified information, including:

[0132] Determine the regional position of the fixed target analysis object corresponding to the third regional task; the fixed target analysis object is: the target analysis object corresponding to the third regional task with a fixed position;

[0133] Obtain the first observation condition for the first analysis value mean of the fixed target analysis object and associate it with the regional location. The first analysis value mean is the analysis value mean corresponding to the fixed target analysis object, such as transformer oil leakage. The first observation condition is the condition under which the first analysis value mean can be obtained, such as shooting towards the oil tank seal ring and shooting distance less than 50 cm.

[0134] Planning the mobile identification device to move to a regional position closest to the mobile identification device in the execution area of ​​the third regional task, and marking the regional position as the initial position;

[0135] Plan the main observation line based on the distribution of the initial position and regional positions in the execution area. When planning the main observation line, based on route planning technology, use the initial position as the starting point of route planning and plan the shortest route passing through each regional position.

[0136] Marking the first observation condition associated with the regional position on the observation main line to generate first pilotage information;

[0137] Controlling the mobile identification device to move and shoot based on the first pilotage information;

[0138] updating the remaining pilotage information of the current area location according to the captured image of the current area location, and controlling the movement and capturing of the mobile identification device based on the updated pilotage information;

[0139] Repeat the steps of: updating the remaining navigation information of the current area according to the captured image of the current area; and controlling the movement and capturing of the mobile identification device based on the updated navigation information;

[0140] When all the area positions are photographed, the acquisition of the information to be identified is completed;

[0141] The remaining pilotage information of the current area is updated based on the captured image of the current area, including:

[0142] Detecting the target analysis object corresponding to the third area task in the detection area according to the captured image to obtain a first detection target; this is achieved based on target detection technology;

[0143] Acquire a first detection target that appears in the same captured image and has not acquired information to be identified, and use the first detection target as a second detection target;

[0144] If the acquisition is successful, the second observation condition is obtained based on the second analysis value mean corresponding to the second detection target; the second observation condition includes: the observation point and the observation posture; the second analysis value mean is: the analysis value mean corresponding to the target analysis object at a non-fixed position, such as: the irregular operation behavior of the maintenance personnel;

[0145] Calculate the real-time distance between the observation point of the mobile identification device and the second detection target based on the captured image;

[0146] Plan the observation order of the second detection target according to the order of real-time distance from small to large;

[0147] Determine the second pilotage information based on the observation sequence and observation posture;

[0148] The second pilotage information is placed in front of the remaining pilotage information to obtain updated pilotage information.

[0149] The working principle and beneficial effects of the above technical solution are:

[0150] The target analysis objects are divided into movable objects (such as substation staff) and fixed objects. The observation position of the fixed object is fixed. Therefore, based on the regional position of the fixed target analysis object corresponding to the third regional task, the initial position and the regional position are determined according to the regional position in the execution area and the main observation line is planned. The first observation condition of the first analysis value mean of the fixed target analysis object is marked on the observation main line, and the first navigation information is obtained. The mobile identification device is controlled to move and shoot based on the first navigation information. When moving and shooting, the movable object (second detection target) is obtained according to the captured image of the current regional position. According to the second analysis value mean of the movable object, the second observation condition (observation point position and observation posture of the movable object) is determined. ); using the depth camera and the captured image, the real-time distance between the observation point of the mobile identification device and the second detection target is calculated, and the observation order of the second detection target is planned in the order of the real-time distance from small to large, based on the observation order and the observation posture, the second navigation information is determined, the second navigation information is placed in front of the remaining navigation information, and the updated navigation information is obtained, and the system operation of updating the remaining navigation information of the current area position according to the captured image of the current area position as described in the information acquisition subsystem to be identified is repeated, and the system operation of controlling the movement and shooting of the mobile identification device is controlled based on the updated navigation information, so that in the process of analyzing the fixed target analysis object, the movable object is captured and analyzed based on the necessary captured images, and the navigation information is dynamically determined, which greatly improves the analysis and judgment efficiency.

[0151] The embodiment of the present invention provides a method for identifying, analyzing and determining major accident hazards based on an AI large model, such as Figure 2 Shown, including:

[0152] Step 1: Obtain the information to be identified in the detection area;

[0153] Step 2: Train the AI ​​model for identifying potential accidents;

[0154] Step 3: Build industry domain knowledge graph;

[0155] Step 4: Based on the accident hazard identification AI big model and industry knowledge graph, major accident hazards are identified, analyzed, and determined according to the information to be identified.

[0156] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A major accident hazard identification, analysis, and judgment system based on a large AI model, characterized by: include: The information to be identified acquisition subsystem is used to acquire the information to be identified in the detection area; Model training subsystem, used to train large AI models for accident hazard identification; Knowledge graph construction subsystem, used to build industry knowledge graphs; The hidden danger identification subsystem is used to identify, analyze, and determine major accident hazards based on the information to be identified, using the accident hazard identification AI large model and industry knowledge graph; The to-be-identified information acquisition subsystem acquires the to-be-identified information of the detection area, and also includes: Get the region type label of the detection region; Retrieve accident events based on area type labels; According to the accident event, obtain the basis for determining the target to be identified; Obtain regional operation information of the detection area; Based on the regional operation information and the basis for determining the target to be identified, obtain the information to be identified in the detection area, including: Analyze regional operation information and obtain second-region tasks; Match the second area task with the first area task in the basis for determining the target to be identified. If they are consistent, the corresponding first area task is used as the third area task; Based on the target recognition technology, the target analysis object corresponding to the third area task in the detection area is detected to obtain the information to be recognized, including: Determine the regional location of the fixed target analysis object corresponding to the third regional task; Obtaining a first observation condition of a first analysis value mean of a fixed target analysis object and associating it with a regional position; Planning the mobile identification device to move to a regional position closest to the mobile identification device in the execution area of ​​the third regional task, and marking the regional position as the initial position; Plan the main observation line based on the distribution of the initial position and regional position in the execution area; Marking the first observation condition associated with the regional position on the observation main line to generate first pilotage information; Controlling the mobile identification device to move and shoot based on the first pilotage information; updating the remaining pilotage information of the current area location according to the captured image of the current area location, and controlling the movement and capturing of the mobile identification device based on the updated pilotage information; Repeat the steps of: updating the remaining navigation information of the current area according to the captured image of the current area; and controlling the movement and capturing of the mobile identification device based on the updated navigation information; When all the area positions are photographed, the acquisition of the information to be identified is completed; The remaining pilotage information of the current area is updated based on the captured image of the current area, including: Detecting, based on the captured image, a target analysis object corresponding to a third area task in the detection area to obtain a first detection target; Acquire a first detection target that appears in the same captured image and has not acquired information to be identified, and use the first detection target as a second detection target; If the acquisition is successful, the second observation condition is obtained according to the second analysis value mean corresponding to the second detection target; the second observation condition includes: observation point position and observation posture; Calculate the real-time distance between the observation point of the mobile identification device and the second detection target based on the captured image; Plan the observation order of the second detection target according to the order of real-time distance from small to large; Determine the second pilotage information based on the observation sequence and observation posture; The second pilotage information is placed in front of the remaining pilotage information to obtain updated pilotage information.

2. The major accident hazard identification, analysis and determination system based on the AI ​​large model according to claim 1 is characterized in that: The knowledge graph construction subsystem builds industry-specific knowledge graphs, including: Obtain historical accident data, expert experience, and safety laws, regulations, and standards; Based on knowledge graph construction technology, according to historical accident data, expert experience and safety laws, regulations and standards, the key factors and correlations of major accident hazards are identified and mapped to obtain industry knowledge graphs.

3. The major accident hazard identification, analysis and determination system based on the AI ​​large model according to claim 1 is characterized in that: The hidden danger identification subsystem is based on the accident hidden danger identification AI large model and industry domain knowledge graph, and conducts major accident hidden danger identification, analysis, and judgment based on the information to be identified, including: Based on natural language processing technology and the information to be identified, effective trigger words are extracted from the industry knowledge graph; Optimize the hidden danger reasoning rules of the accident hidden danger identification AI model based on the graph features of effective trigger words in the industry knowledge graph.

4. The major accident hazard identification, analysis and determination system based on the AI ​​large model according to claim 1 is characterized in that: The hazard identification subsystem, based on the accident hazard identification AI large model and industry knowledge graph, identifies, analyzes, and determines major accident hazards based on the information to be identified. It also includes: Perform image recognition analysis on the information to be identified to obtain identification image features; Match the identified image features with the suspicious hidden danger feature library obtained from the industry knowledge graph. If a match exists, obtain the hidden danger description corresponding to the matching suspicious hidden danger feature; Continue to search the industry knowledge graph for the criteria for determining the corresponding hidden danger type based on the hidden danger description; Identify, analyze and determine major accident hazards based on hazard descriptions and determination criteria.

5. The major accident hazard identification, analysis and determination system based on the AI ​​large model according to claim 1 is characterized in that: The information acquisition subsystem for obtaining information to be identified obtains the information to be identified in the detection area, including: Obtain the information to be identified through mobile terminals, inspection robots and drones.

6. A method for identifying, analyzing, and determining major accident hazards based on a large AI model, characterized by: include: Step 1: Obtain the information to be identified in the detection area; Step 2: Train the AI ​​model for identifying potential accident hazards; Step 3: Build industry domain knowledge graph; Step 4: Based on the accident hazard identification AI large model and industry knowledge graph, major accident hazards are identified, analyzed, and determined according to the information to be identified; The step 1: obtaining the information to be identified in the detection area includes: Get the region type label of the detection region; Retrieve accident events based on area type labels; According to the accident event, obtain the basis for determining the target to be identified; Obtain regional operation information of the detection area; Based on the regional operation information and the basis for determining the target to be identified, obtain the information to be identified in the detection area, including: Analyze regional operation information and obtain second-region tasks; Match the second area task with the first area task in the basis for determining the target to be identified. If they are consistent, the corresponding first area task is used as the third area task; Based on the target recognition technology, the target analysis object corresponding to the third area task in the detection area is detected to obtain the information to be recognized, including: Determine the regional location of the fixed target analysis object corresponding to the third regional task; Obtaining a first observation condition of a first analysis value mean of a fixed target analysis object and associating it with a regional position; Planning the mobile identification device to move to a regional position closest to the mobile identification device in the execution area of ​​the third regional task, and marking the regional position as the initial position; Plan the main observation line based on the distribution of the initial position and regional position in the execution area; Marking the first observation condition associated with the regional position on the observation main line to generate first pilotage information; Controlling the mobile identification device to move and shoot based on the first pilotage information; updating the remaining pilotage information of the current area location according to the captured image of the current area location, and controlling the movement and capturing of the mobile identification device based on the updated pilotage information; Repeat the steps of: updating the remaining navigation information of the current area according to the captured image of the current area; and controlling the movement and capturing of the mobile identification device based on the updated navigation information; When all the area positions are photographed, the acquisition of the information to be identified is completed; The remaining pilotage information of the current area is updated based on the captured image of the current area, including: Detecting, based on the captured image, a target analysis object corresponding to a third area task in the detection area to obtain a first detection target; Acquire a first detection target that appears in the same captured image and has not acquired information to be identified, and use the first detection target as a second detection target; If the acquisition is successful, the second observation condition is obtained according to the second analysis value mean corresponding to the second detection target; the second observation condition includes: observation point position and observation posture; Calculate the real-time distance between the observation point of the mobile identification device and the second detection target based on the captured image; Plan the observation order of the second detection target according to the order of real-time distance from small to large; Determine the second pilotage information based on the observation sequence and observation posture; The second pilotage information is placed in front of the remaining pilotage information to obtain updated pilotage information.

Images