Supervision security policy generation method and system based on large model

Abstract

The invention provides a supervision security policy generation method and system based on a large model, and the method comprises the steps: collecting and processing data, transmitting the data to a large model service module for safety behavior analysis, obtaining a safety reasoning model, generating a supervision security policy instruction, and transmitting the supervision security policy instruction to a scheduling module; the scheduling module generates a scheduling strategy according to the supervision security strategy instruction, and the scheduling module pushes the supervision security strategy instruction to the reasoning module; and the inference module performs safety inference work, analyzes the supervision safety strategy instruction to obtain an inference result, and transmits the inference result to the business system to form an operation safety supervision event. According to the method, comprehensive analysis is carried out through a large model, and limited computing resources are utilized to the greatest extent, so that the overall processing efficiency of the system, the convenience of safety production work and the automation degree are greatly improved; through dynamic scheduling and intelligent analysis, monitoring blind areas and invalid monitoring are effectively reduced, the overall supervision efficiency is improved, and a reliable basis is provided for subsequent safety supervision event decisions.

Description

Technical Field

[0001] The present invention belongs to the technical field of industrial automation and intelligent safety monitoring, and in particular, relates to a method and system for generating a supervisory safety strategy based on a large model. Background Art

[0002] Under the background of the continuous upgrading and improvement of domestic information technology infrastructure, the steel production industry is undergoing a profound transformation from traditional monitoring to intelligent monitoring. The core driving force of this transformation lies in the breakthrough progress in video image recognition technology and artificial intelligence (AI), especially AI machine vision technology. These technological innovations have prompted the digitalization departments of steel mills to widely deploy intelligent detection and monitoring systems based on AI machine vision. The system integrates high-performance monitoring equipment and advanced deep learning algorithms to achieve continuous and accurate monitoring of safe operation behaviors on the production site.

[0003] However, although AI machine vision technology has shown great potential in improving monitoring efficiency and accuracy, its technical implementation is highly dependent on real-time decoding of video data and complex image reasoning and recognition processes, which require extremely large computing resources, especially CPU and GPU resources. At the same time, given the nature of the safe operation behavior supervision system as an auxiliary management tool, its budget allocation is often relatively limited, resulting in a general lack of computing resources available to the system in actual deployment.

[0004] This technical bottleneck is particularly prominent in low-cost, computing-limited hardware environments, limiting the performance of intelligent monitoring systems in detecting production operation behaviors. Specifically, it is manifested in low detection efficiency and slow response speed, making it difficult to effectively support production safety management departments to make timely and accurate emergency decisions, which may increase safety risks and hidden dangers in the production process.

[0005] Patent document "A heterogeneous edge cloud AI system task scheduling framework based on reinforcement learning" (CN116010054A) proposes a heterogeneous edge cloud AI system task scheduling framework based on reinforcement learning to meet the requirements of high throughput and low latency in heterogeneous edge cloud AI reasoning systems. However, its algorithm scheduling strategy is to evaluate the status and reasoning information of edge devices to adjust the reasoning tasks of edge devices. This has low overall supervision efficiency and is prone to invalid monitoring.

[0006] The patent document "Park Security Video Surveillance Method, Device, Equipment and Storage" (CN118097534A) can be compiled into task instructions for the park security management goals, which belongs to data understanding based on large language models and large visual models. However, its algorithm usage scenario is used for multiple different CV algorithm engines without pre-training, relying on large models, and cannot adapt to complex industrial environments and changing operating scenarios. On the one hand, it lacks strategy pre-configuration for specific tasks, resulting in the inability to meet complex security supervision needs. On the other hand, it is difficult to provide efficient monitoring support under conditions of limited computing power. It requires high-performance computing resources and a large amount of storage resources to store model parameters and training data, and lacks the real-time nature of task strategy scheduling.

[0007] Therefore, there is an urgent need for a monitoring and supervision system that can efficiently and real-time detect safe operation behaviors in the production process under limited computing resources and different computing power environments. Summary of the invention

[0008] In view of the defects in the prior art, the object of the present invention is to provide a method and system for generating a supervisory security strategy based on a large model.

[0009] The method for generating a supervisory safety strategy based on a large model provided by the present invention comprises:

[0010] Step S1: collect data, process and transmit to the large model service module;

[0011] Step S2: Import the data into the large model for security behavior analysis to obtain a security reasoning model;

[0012] Step S3: Instruct the large model to generate a supervisory safety policy instruction based on the safety reasoning model and transmit it to the scheduling module;

[0013] Step S4: instructing the scheduling module to generate a scheduling strategy according to the supervisory security policy instruction;

[0014] Step S5: Instruct the scheduling module to schedule computing power according to the scheduling strategy and push the supervision security policy instructions to the reasoning module;

[0015] Step S6: instruct the reasoning module to perform security reasoning and parse the supervisory security policy instructions to obtain reasoning results;

[0016] Step S7: Instruct the reasoning module to transmit the reasoning result to the business system to form an operation safety supervision event.

[0017] Preferably, the data in step S1 are input parameter modules of a security management system, a video surveillance system and a production management system.

[0018] The processing is to weave the goals that the safety management system needs to supervise the factory area into task instructions that can be understood and processed by the large language model, perform image acquisition on the video by the frame extraction module of the video surveillance system to obtain real-time image frame information, and process the existing production task situations in the production management system to generate production task information in the current production management system.

[0019] The task instruction includes the existing safety supervision behavior task information and monitoring strategy description in the safety reasoning model.

[0020] The production task information includes operation time, operation location, and operation task description.

[0021] Preferably, in step S2, the image frame information, production task information and task instructions are taken together as pre-configured inputs, and the safety behavior analysis is performed according to the monitoring strategy description requirements in the task instructions.

[0022] The safety behavior analysis uses a large model to understand the image frame information and evaluate the input data.

[0023] The image frame information is obtained by a frame extraction module of a video surveillance system.

[0024] The production task information is provided by the production management system.

[0025] The safety supervision behavior task information in the task instruction includes environmental data information and fault information.

[0026] The big model is a multimodal visual understanding big model, including a visual big model and a language big model.

[0027] The large model is configured to receive input in both image and text modalities.

[0028] The visual macromodel includes a visual component module capable of converting an image into a format compatible with the language macromodel.

[0029] Preferably, the scheduling module includes a strategy parsing module and an inference task pushing module.

[0030] The policy parsing module parses the supervisory security policy instructions into a specified reasoning task format.

[0031] The inference task push module performs task scheduling work on the inference module, and pushes supervisory security policy instructions to the inference module according to the scheduling strategy scheduling computing power.

[0032] After the large model gathers data in step S3, the corresponding supervisory safety policy instructions are generated from the large model according to the safety reasoning model and transmitted to the scheduling module.

[0033] The supervisory security policy instruction includes a task description related to the video to be monitored, a security reasoning model that needs to be analyzed and processed in the current video image, and a description of the monitoring policy that needs to be executed.

[0034] The monitoring strategy description includes information about the task executed by the reasoning module, task description, task execution time, video information to be monitored and the reasoning model to be used.

[0035] Preferably, in step S6, the reasoning module obtains the supervisory security policy instruction, parses it, and schedules computing power to perform security reasoning work.

[0036] Scheduling computing power includes deploying computing resources on the edge side and the terminal side.

[0037] The reasoning module starts and stops the visual tasks it manages, and pushes the event information generated by reasoning to the business system to form operational safety behavior events.

[0038] The business system includes a production management system and an equipment inspection system.

[0039] The event information includes the camera number, event occurrence time, event occurrence type and event picture.

[0040] A large model-based supervisory security policy generation system provided in accordance with the present invention includes: a video acquisition service module, a data acquisition service module, a business acquisition service module, a large model service module, a scheduling module and an inference module.

[0041] The video acquisition service module, data acquisition service module and business acquisition service module collect data, process it and transmit it to the large model service module;

[0042] Import the data into the big model for security behavior analysis to obtain a security reasoning model;

[0043] The large model service module generates supervisory safety policy instructions based on the safety reasoning model and transmits them to the scheduling module;

[0044] The scheduling module generates scheduling policies based on supervisory security policy instructions;

[0045] The scheduling module dispatches computing power according to the scheduling strategy and pushes supervisory security policy instructions to the inference module;

[0046] The reasoning module performs security reasoning and parses the supervisory security policy instructions to obtain reasoning results;

[0047] The reasoning module transmits the reasoning results to the business system to form operational safety supervision events.

[0048] The video acquisition service module includes a video monitoring system.

[0049] The data collection service module includes a security management system.

[0050] The business collection service module includes a production management system and an equipment inspection system.

[0051] Preferably, the data are input parameter modules of a security management system, a video surveillance system and a production management system.

[0052] The processing is to weave the goals that the safety management system needs to supervise the factory area into task instructions that can be understood and processed by the large language model, perform image acquisition on the video by the frame extraction module of the video surveillance system to obtain real-time image frame information, and process the existing production task situations in the production management system to generate production task information in the current production management system.

[0053] The task instruction includes the existing safety supervision behavior task information and monitoring strategy description in the safety reasoning model.

[0054] The production task information includes operation time, operation location, and operation task description.

[0055] Preferably, the safety behavior analysis is performed by taking the image frame information, the production task information and the task instruction as pre-configured inputs, and performing the safety behavior analysis according to the monitoring strategy description requirements in the task instruction.

[0056] In the safety behavior analysis, the understanding of the image frame information and the input data are evaluated through the large model service module.

[0057] The image frame information is obtained by a frame extraction module of a video surveillance system.

[0058] The production task information is provided by the production management system.

[0059] The safety supervision behavior task information in the task instruction includes environmental data information and fault information.

[0060] The big model in the big model service module is a multimodal visual understanding big model, including a visual big model and a language big model.

[0061] The large model is configured to receive input in both image and text modalities.

[0062] The visual macromodel includes a visual component module capable of converting an image into a format compatible with the language macromodel.

[0063] Preferably, the scheduling module includes a strategy parsing module and an inference task pushing module.

[0064] The policy parsing module parses the supervisory security policy instructions into a specified reasoning task format.

[0065] The inference task push module performs task scheduling work on the inference module, and pushes supervisory security policy instructions to the inference module according to the scheduling strategy scheduling computing power.

[0066] Scheduling computing power includes deploying computing resources on the edge side and the terminal side.

[0067] After the large model service module aggregates the data, it generates corresponding supervisory safety policy instructions based on the safety reasoning model and transmits them to the scheduling module.

[0068] The supervisory security policy instruction includes a task description related to the video to be monitored, a security reasoning model that needs to be analyzed and processed in the current video image, and a description of the monitoring policy that needs to be executed.

[0069] The monitoring strategy description includes information about the task executed by the reasoning module, task description, task execution time, video information to be monitored and the reasoning model to be used.

[0070] Preferably, the reasoning module obtains the supervisory security policy instructions, parses them, and dispatches computing power to perform security reasoning work.

[0071] The reasoning module starts and stops the visual tasks it manages, and pushes the event information generated by reasoning to the business system to form operational safety behavior events.

[0072] The business system includes a production management system and an equipment inspection system.

[0073] The event information includes the camera number, event occurrence time, event occurrence type and event picture.

[0074] Compared with the prior art, the present invention has the following beneficial effects:

[0075] 1. The present invention uses the system to call strategies and dispatch computing resources to perform security reasoning, which provides a reliable basis for subsequent security supervision event decision-making, and greatly improves the convenience and automation of safe production work.

[0076] 2. The present invention uses a large model to conduct a comprehensive analysis of various parameter conditions, effectively reducing the pressure on personnel to handle algorithm scheduling, and providing emergency decision-making plans in a timely manner, thereby reducing the safety hazards of safety incidents within the factory, while ensuring the monitoring quality of key areas and taking into account the coverage of safety supervision throughout the factory.

[0077] 3. When allocating inference scheduling tasks, the present invention dynamically adjusts them according to the current device parameters and business, and reasonably allocates them to the inference module, maximizes the use of limited computing resources, avoids waste or overload of computing power, and thus improves the overall processing efficiency of the system.

[0078] 4. The present invention comprehensively analyzes the reasoning results, and through dynamic scheduling and intelligent analysis, closely integrates the supervision tasks with the actual production, flexibly adjusts the monitoring focus according to the production plan and actual situation, effectively reduces monitoring blind spots and ineffective monitoring, and improves the overall supervision efficiency.

[0079] 5. The present invention decouples the scheduling module and the reasoning module that generate the scheduling strategy, optimizes resource allocation, ensures that high-performance computing resources can be used by their respective services on demand and efficiently, enhances the flexibility and scalability of the system, and achieves a significant improvement in system performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0081] Figure 1 Schematic flow diagram of the method for generating supervisory safety policies based on large models.

[0082] Figure 2 Generate system architecture diagrams for large model-based supervisory safety strategies.

[0083] Figure 3 Generate system module diagrams for large model-based supervisory safety strategies. DETAILED DESCRIPTION

[0084] The present invention is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those of ordinary skill in the art, several changes and improvements can also be made without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

[0085] In complex industrial environments, the biggest technical difficulty is how to ensure the accuracy of task strategies and the real-time response capabilities of the system, especially in high-risk scenarios that require precise monitoring of specific tasks. When faced with complex safety supervision needs, general model processing methods limit the precise control and analysis depth of tasks, especially in industrial production environments, where the adaptability and efficiency are low and specific task strategies and CV algorithm engines need to be pre-configured.

[0086] In view of the current situation of tight computing resources and automated scheduling, a regulatory security strategy generation method based on a large model is provided in the present invention. Figure 1 For example, including:

[0087] Step S1: Collect data and transmit to the large model service module;

[0088] Collect the input parameter modules of the security management system, video surveillance system and production management system, and aggregate the data into the large model service module for processing.

[0089] Specifically, the goals of the safety management system for factory supervision are woven into task instructions that can be understood and processed by the large language model. According to the video description of the video surveillance system, the video of the video surveillance system is captured through the frame extraction module to obtain the current real-time image frame information. The existing production task situations in the production management system are used to generate specific processing task instructions, that is, the production task information in the current production management system.

[0090] The task instruction includes the existing safety supervision behavior task information and monitoring strategy description in the safety reasoning model.

[0091] The production task information includes operation time, operation location, operation task description, etc.

[0092] Step S2: automatically import the data into the large model for security behavior analysis to obtain a security reasoning model;

[0093] Specifically, the input content of the large model consists of a production management system, a frame extraction module, and a SCADA system. The production system provides production tasks and plan information, and the frame extraction module provides camera screenshot image frame information of the on-site video monitoring system. The on-site SCADA system, i.e., the safety management system, provides the factory's safety supervision behavior task information, including environmental data information, fault information, etc.

[0094] The collected image frame information, production task information of the production management system and the monitoring strategy description in the task instruction are taken as the input of the preconfigured multimodal visual understanding large model, and the safety behavior analysis of the image is performed according to the requirements of the monitoring strategy description.

[0095] The big model, i.e., the big model service, includes a visual big model service and a language big model service.

[0096] The multimodal visual understanding model is configured to receive inputs in image and text modalities, fuse image and text data, generate more intelligent security supervision strategies, provide more comprehensive security monitoring capabilities, and improve the accuracy and timeliness of decision-making.

[0097] The multimodal visual understanding large model includes a large language model and a visual component module capable of converting an image into a format compatible with the large language model.

[0098] The content of complex strategy generation tasks is handed over to the big model, and the content of simple visual tasks is handed over to the CV algorithm inference engine, realizing efficient use of resources. Through the big model, the analysis of various parameter conditions effectively reduces the pressure on personnel to handle algorithm scheduling, while ensuring the quality of key area monitoring and taking into account the coverage of safety supervision of the whole plant.

[0099] Step S3: Instruct the large model to generate a supervisory safety policy instruction based on the safety reasoning model and transmit it to the scheduling module;

[0100] Specifically, after the big model module gathers these data, it merges the language big model and the visual big model through specific dialogues to form the content that needs to be provided to the scheduling module. The system makes a policy call, generates the corresponding scheduling strategy from the big model, and outputs regulatory safety policy instructions.

[0101] The algorithm scheduling strategy evaluates the large model's own understanding of the current image and the production data and production tasks input by the background system to obtain the tasks that require reasoning, thereby reducing invalid monitoring and improving overall supervision efficiency.

[0102] The supervisory security policy instruction includes a task description related to the video to be monitored, a security reasoning model that needs to be analyzed and processed in the current video image, and a description of the monitoring policy that needs to be executed.

[0103] The monitoring strategy description includes information about the task executed by the reasoning module, task description, task execution time, video information to be monitored, and the reasoning model to be used.

[0104] Step S4: instructing the scheduling module to generate a scheduling strategy according to the supervisory security policy instruction;

[0105] The scheduling module can obtain the content of the specified text generated by the large model module.

[0106] Through pre-training customized CV algorithm task strategies, it can effectively cope with the higher security and precision requirements of various operating scenarios in complex production environments, ensure that the system is more accurate when processing specific tasks, and improve the efficiency of task execution. Through comprehensive analysis of large models and dynamic scheduling of computing resources for security reasoning, computing resources can be used more flexibly, including deploying computing resources on the edge and end sides to cope with different processing needs.

[0107] Step S5: Instruct the scheduling module to schedule computing power according to the scheduling strategy and push the supervision security policy instructions to the reasoning module;

[0108] The policy parsing module is used to parse the regulatory security policy instructions generated in the large model service into the specified reasoning task format; the reasoning task push module is used to schedule tasks for the reasoning module.

[0109] When allocating inference scheduling tasks, especially when dealing with complex industrial safety supervision tasks, it is necessary to optimize computing resources in real time and schedule inference tasks based on the production plans and work plans generated by system docking, dynamically adjust according to current equipment parameters and business, and reasonably allocate them to inference modules, and send inference tasks to the edge side and the terminal side to fully utilize the computing resources on the edge side and the terminal side, maximize the use of limited computing resources, avoid computing power waste or overload, thereby improving the overall processing efficiency of the system and providing a reliable basis for subsequent safety supervision event decision-making.

[0110] Step S6: instruct the reasoning module to perform security reasoning and parse the supervisory security policy instructions to obtain reasoning results;

[0111] Specifically, it receives information from the scheduling module and then starts and stops the visual tasks it manages. The scheduling module inputs the supervisory safety policy instructions into the reasoning module. After obtaining the supervisory safety policy instructions, the reasoning module parses the instructions and schedules computing power for safety reasoning.

[0112] The scheduling module and reasoning module used to generate the scheduling strategy are decoupled from each other. This separation strategy relies on advanced large model technology to specifically undertake the complex task of generating scheduling strategies, ensuring accurate and efficient strategy formulation. At the same time, the deeply customized algorithm model for the actual on-site scenarios is independently deployed as an inference service, focusing on executing reasoning analysis, so as to achieve rapid response and accurate interpretation of specific environmental changes.

[0113] The in-depth application of big models at the scheduling level has promoted the intelligence and dynamics of resource scheduling, while the introduction of customized reasoning services has further enhanced the system's adaptability to complex and changing on-site environments and decision-making accuracy.

[0114] Step S7: Instruct the reasoning module to transmit the reasoning result to the business system to form an operation safety supervision event.

[0115] The inference module pushes the event information generated by the inference to the business system to form a work safety behavior event. The event information includes the camera number, event time, event type, event picture and other information.

[0116] The business system includes the production management system and the equipment inspection system.

[0117] by Figure 3The safety supervision system, taking the example of a comprehensive analysis of the reasoning results, provides supervisors with various safety supervision events that occur in the factory, and provides emergency decision-making plans in a timely manner, thereby reducing the safety hazards of safety events in the factory. Through dynamic scheduling and intelligent analysis, the deep integration and full utilization of resources are achieved, laying a solid foundation for the efficient and stable operation of the system, closely combining supervision tasks with production reality, flexibly adjusting monitoring priorities according to production plans and actual conditions, reducing ineffective monitoring and monitoring blind spots, and improving overall supervision efficiency.

[0118] The present invention also provides a regulatory security policy generation system based on a big model. The regulatory security policy generation system based on a big model can be implemented by executing the process steps of the regulatory security policy generation method based on a big model, that is, those skilled in the art can understand the regulatory security policy generation method based on a big model as a preferred implementation mode of the regulatory security policy generation system based on a big model.

[0119] Those skilled in the art know that, in addition to realizing the system and its various devices, modules, and units provided by the present invention in a purely computer-readable program code, it is entirely possible to realize the same functions in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, the system and its various devices, modules, and units provided by the present invention can be considered as a hardware component, and the devices, modules, and units included therein for realizing various functions can also be regarded as structures within the hardware component; the devices, modules, and units for realizing various functions can also be regarded as both software modules for realizing the method and structures within the hardware component.

[0120] According to a large model-based supervisory security strategy generation system provided by the present invention, Figure 2 For example, it includes: video acquisition service module, data acquisition service module, business acquisition service module, large model service module, scheduling module and reasoning module.

[0121] The video acquisition service module, data acquisition service module and business acquisition service module collect data, process it and transmit it to the large model service module;

[0122] Import the data into the big model for security behavior analysis to obtain a security reasoning model;

[0123] The large model service module generates supervisory safety policy instructions based on the safety reasoning model and transmits them to the scheduling module;

[0124] The scheduling module generates scheduling policies based on supervisory security policy instructions;

[0125] The scheduling module dispatches computing power according to the scheduling strategy and pushes supervisory security policy instructions to the inference module;

[0126] The reasoning module performs security reasoning and parses the supervisory security policy instructions to obtain reasoning results;

[0127] The reasoning module transmits the reasoning results to the business system to form operational safety supervision events.

[0128] The video acquisition service module includes a video monitoring system.

[0129] The data collection service module includes a security management system.

[0130] The business collection service module includes a production management system and an equipment inspection system.

[0131] In more preferred examples, the data are input parameter modules of a security management system, a video surveillance system, and a production management system.

[0132] The processing is to weave the goals that the safety management system needs to supervise the factory area into task instructions that can be understood and processed by the large language model, perform image acquisition on the video by the frame extraction module of the video surveillance system to obtain real-time image frame information, and process the existing production task situations in the production management system to generate production task information in the current production management system.

[0133] The task instruction includes the existing safety supervision behavior task information and monitoring strategy description in the safety reasoning model.

[0134] The production task information includes operation time, operation location, and operation task description.

[0135] In more preferred examples, the safety behavior analysis is performed by taking the image frame information, production task information and task instructions as pre-configured inputs, and performing the safety behavior analysis according to the monitoring strategy description requirements in the task instructions.

[0136] In the safety behavior analysis, the understanding of the image frame information and the input data are evaluated through the large model service module.

[0137] The image frame information is obtained by a frame extraction module of a video surveillance system.

[0138] The production task information is provided by the production management system.

[0139] The safety supervision behavior task information in the task instruction includes environmental data information and fault information.

[0140] The big model in the big model service module is a multimodal visual understanding big model, including a visual big model and a language big model.

[0141] The large model is configured to receive input in both image and text modalities.

[0142] The visual macromodel includes a visual component module capable of converting an image into a format compatible with the language macromodel.

[0143] In more preferred examples, the scheduling module includes a policy parsing module and an inference task pushing module.

[0144] The policy parsing module parses the supervisory security policy instructions into a specified reasoning task format.

[0145] The inference task push module performs task scheduling work on the inference module, and pushes supervisory security policy instructions to the inference module according to the scheduling strategy scheduling computing power.

[0146] After the large model service module aggregates the data, it generates corresponding supervisory safety policy instructions based on the safety reasoning model and transmits them to the scheduling module.

[0147] The supervisory security policy instruction includes a task description related to the video to be monitored, a security reasoning model that needs to be analyzed and processed in the current video image, and a description of the monitoring policy that needs to be executed.

[0148] The monitoring strategy description includes information about the task executed by the reasoning module, task description, task execution time, video information to be monitored and the reasoning model to be used.

[0149] In more preferred examples, the reasoning module obtains the supervisory security policy instructions, parses them, and schedules computing power to perform security reasoning work.

[0150] The reasoning module starts and stops the visual tasks it manages, and pushes the event information generated by reasoning to the business system to form operational safety behavior events.

[0151] The business system includes a production management system and an equipment inspection system.

[0152] The event information includes the camera number, event occurrence time, event occurrence type and event picture.

[0153] The above describes the specific embodiments of the present invention. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. In the absence of conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. A method for generating a supervisory security strategy based on a large model, characterized in that: include: Step S1: collect data, process and transmit to the large model service module; Step S2: Import the data into the large model for security behavior analysis to obtain a security reasoning model; Step S3: Instruct the large model to generate a supervisory safety policy instruction based on the safety reasoning model and transmit it to the scheduling module; Step S4: instructing the scheduling module to generate a scheduling strategy according to the supervisory security policy instruction; Step S5: Instruct the scheduling module to schedule computing power according to the scheduling strategy and push the supervision security policy instructions to the reasoning module; Step S6: instruct the reasoning module to perform security reasoning and parse the supervisory security policy instructions to obtain reasoning results; Step S7: Instruct the reasoning module to transmit the reasoning result to the business system to form an operation safety supervision event.

2. The method for generating a supervisory security strategy based on a large model according to claim 1, characterized in that: The data in step S1 is the input parameter module of the security management system, video monitoring system and production management system; The processing is to weave the goals that the safety management system needs to supervise the factory area into task instructions that can be understood and processed by the large language model, collect images from the video by the frame extraction module of the video monitoring system to obtain real-time image frame information, and process the existing production task situations in the production management system to generate production task information in the current production management system; The task instructions include existing safety supervision behavior task information and monitoring strategy description in the safety reasoning model; The production task information includes operation time, operation location, and operation task description.

3. The method for generating a supervisory security strategy based on a large model according to claim 1, characterized in that: In step S2, the image frame information, production task information and task instructions are used as pre-configured inputs, and safety behavior analysis is performed according to the monitoring strategy description requirements in the task instructions; The safety behavior analysis uses a large model to understand the image frame information and evaluate the input data; The image frame information is obtained by a frame extraction module of a video surveillance system; The production task information is provided by the production management system; The safety supervision behavior task information in the task instruction includes environmental data information and fault information; The large model is a multimodal visual understanding large model, including a visual large model and a language large model; The large model is configured to receive input in image and text modalities; The visual macromodel includes a visual component module capable of converting an image into a format compatible with the language macromodel.

4. The method for generating a supervisory security strategy based on a large model according to claim 1, characterized in that: The scheduling module includes a strategy parsing module and an inference task pushing module; The policy parsing module parses the supervisory security policy instructions into a specified reasoning task format; The inference task push module performs task scheduling on the inference module, and pushes supervision security policy instructions to the inference module according to the scheduling strategy scheduling computing power; The scheduling of computing power includes deploying computing resources on the edge side and the terminal side; After the large model gathers data in step S3, the corresponding supervisory safety policy instructions are generated from the large model according to the safety reasoning model and transmitted to the scheduling module; The supervisory security policy instruction includes a task description related to the video to be monitored, a security reasoning model that needs to be analyzed and processed in the current video image, and a description of the monitoring policy that needs to be executed; The monitoring strategy description includes information about the task executed by the reasoning module, task description, task execution time, video information to be monitored and the reasoning model to be used.

5. The method for generating a supervisory security strategy based on a large model according to claim 1, characterized in that: In step S6, the reasoning module obtains the supervisory security policy instruction, parses it, and dispatches computing power to perform security reasoning work; The reasoning module starts and stops the visual tasks it manages, and pushes the event information generated by reasoning to the business system to form operational safety behavior events; The business system includes a production management system and an equipment inspection system; The event information includes the camera number, event occurrence time, event occurrence type and event picture.

6. A regulatory safety strategy generation system based on a large model, characterized in that: include: Video acquisition service module, data acquisition service module, business acquisition service module, large model service module, scheduling module and reasoning module; The video acquisition service module, data acquisition service module and business acquisition service module collect data, process it and transmit it to the large model service module; Import the data into the big model for security behavior analysis to obtain a security reasoning model; The large model service module generates supervisory safety policy instructions based on the safety reasoning model and transmits them to the scheduling module; The scheduling module generates scheduling policies based on supervisory security policy instructions; The scheduling module dispatches computing power according to the scheduling strategy and pushes supervisory security policy instructions to the inference module; The reasoning module performs security reasoning and parses the supervisory security policy instructions to obtain reasoning results; The reasoning module transmits the reasoning results to the business system to form an operation safety supervision event; The video acquisition service module includes a video monitoring system; The data acquisition service module includes a security management system; The business collection service module includes a production management system and an equipment inspection system.

7. The large model-based supervisory security strategy generation system according to claim 6, characterized in that: The data is the input parameter module of the security management system, video monitoring system and production management system; The processing is to weave the goals that the safety management system needs to supervise the factory area into task instructions that can be understood and processed by the large language model, collect images from the video by the frame extraction module of the video monitoring system to obtain real-time image frame information, and process the existing production task situations in the production management system to generate production task information in the current production management system; The task instructions include existing safety supervision behavior task information and monitoring strategy description in the safety reasoning model; The production task information includes operation time, operation location, and operation task description.

8. The large model-based supervisory security strategy generation system according to claim 6, characterized in that: The safety behavior analysis is to use the image frame information, production task information and task instructions as pre-configured inputs, and perform the safety behavior analysis according to the monitoring strategy description requirements in the task instructions; In the safety behavior analysis, the understanding of the image frame information and the input data are evaluated by the large model service module; The image frame information is obtained by a frame extraction module of a video surveillance system; The production task information is provided by the production management system; The safety supervision behavior task information in the task instruction includes environmental data information and fault information; The big model in the big model service module is a multimodal visual understanding big model, including a visual big model and a language big model; The large model is configured to receive input in image and text modalities; The visual macromodel includes a visual component module capable of converting an image into a format compatible with the language macromodel.

9. The large model-based supervisory security strategy generation system according to claim 6, characterized in that: The scheduling module includes a strategy parsing module and an inference task pushing module; The policy parsing module parses the supervisory security policy instructions into a specified reasoning task format; The inference task push module performs task scheduling on the inference module, and pushes supervision security policy instructions to the inference module according to the scheduling strategy scheduling computing power; The scheduling of computing power includes deploying computing resources on the edge side and the terminal side; After the large model service module gathers data, it generates corresponding supervisory safety policy instructions based on the safety reasoning model and transmits them to the scheduling module; The supervisory security policy instruction includes a task description related to the video to be monitored, a security reasoning model that needs to be analyzed and processed in the current video image, and a description of the monitoring policy that needs to be executed; The monitoring strategy description includes information about the task executed by the reasoning module, task description, task execution time, video information to be monitored and the reasoning model to be used.

10. The large model-based supervisory security strategy generation system according to claim 6, characterized in that: The reasoning module obtains the supervisory security policy instructions, parses them, and dispatches computing power to perform security reasoning work; The reasoning module starts and stops the visual tasks it manages, and pushes the event information generated by reasoning to the business system to form operational safety behavior events; The business system includes a production management system and an equipment inspection system; The event information includes the camera number, event occurrence time, event occurrence type and event picture.

Images