Safety instrument network acquisition and big data analysis early warning system

Abstract

This invention discloses a safety instrument network acquisition and big data analysis and early warning system, belonging to the field of industrial safety monitoring technology. The system includes: a safety instrument network acquisition module, a data preprocessing and storage module, a digital twin modeling module, a reinforcement learning decision-making module, a risk early warning and adaptive control module, and a human-machine interaction module. This invention achieves comprehensive, real-time acquisition and unified management of various safety parameters at the production site through the safety instrument network acquisition module. The digital twin modeling module enables real-time synchronous mapping and risk diffusion simulation between the physical factory and the virtual system, allowing for early identification of hidden safety risks and prediction of accident development trends. The reinforcement learning decision-making module generates dynamically optimized control strategies, solving the problem that traditional fixed strategies are difficult to adapt to complex working conditions. The risk early warning and adaptive control module enables the coordinated execution of early warning and control, shortening response time and avoiding interference from human factors.

Description

Technical Field

[0001] This invention relates to the field of industrial safety monitoring technology, specifically a safety instrument network acquisition and big data analysis and early warning system. Background Technology

[0002] Safety instrumented network (SAN) refers to the technical architecture that interconnects various safety instruments in industrial production sites, such as pressure transmitters, temperature sensors, gas detectors, and level gauges, through industrial communication networks, thereby achieving centralized data acquisition and unified management. Its core significance lies in breaking the limitations of traditional isolated operation of safety instruments, eliminating transmission barriers to safety data in production sites, and providing fundamental data support for building a comprehensive and continuous production safety monitoring system. It is an indispensable component of modern industrial safety production control systems. Safety instrumented data acquisition and big data analysis and early warning refer to the real-time acquisition of various parameters related to equipment operation, production processes, and environmental safety in the production site by networked safety instruments. Big data technology is used to clean, fuse, mine, and deeply analyze massive amounts of multi-source data to identify potential safety hazards and issue early warnings. This technology shifts the focus of industrial safety management from traditional post-accident handling to pre-accident prevention, effectively reducing the probability of production accidents, minimizing casualties and property losses, and playing a crucial role in ensuring the stable operation of industrial production.

[0003] However, existing safety instrumented data acquisition (SAD) systems mostly employ fixed threshold warning modes, only issuing alarms when parameters exceed preset thresholds. They cannot identify hidden risks caused by slow parameter drift or multi-parameter coupling in advance. Furthermore, the warning and control processes are disconnected, requiring manual intervention to execute control operations after the warning signal is issued. This results in delayed response and susceptibility to human factors. In addition, existing systems lack the ability to dynamically simulate fault scenarios and optimize strategies. Control strategies are mostly pre-set fixed schemes, making it difficult to adapt to complex and changing production conditions. This can easily lead to over-control or under-control, failing to achieve intelligent closed-loop management of the entire process. Therefore, developing a SAD network data acquisition and big data analysis early warning system is of great significance. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a safety instrument network acquisition and big data analysis early warning system. This system can achieve comprehensive, real-time acquisition and unified management of various safety parameters at the production site by constructing a safety instrument network acquisition module. Through a digital twin modeling module, it can achieve real-time synchronous mapping and risk diffusion simulation between the physical factory and the virtual system, enabling early identification of hidden safety risks and prediction of accident development trends. By using a reinforcement learning decision-making module to generate dynamically optimized control strategies, it solves the problem that traditional fixed strategies are difficult to adapt to complex working conditions. Finally, through a risk early warning and adaptive control module, it achieves the coordinated execution of early warning and control, shortening response time and avoiding interference from human factors.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a safety instrument network acquisition and big data analysis and early warning system, the system comprising: a safety instrument network acquisition module, a data preprocessing and storage module, a digital twin modeling module, a reinforcement learning decision-making module, a risk early warning and adaptive control module, and a human-computer interaction module;

[0006] The safety instrument network acquisition module is used to collect safety parameters from the production site and transmit the collected raw data to the data preprocessing and storage module.

[0007] The data preprocessing and storage module is used to process and store the raw data, and transmit the processed real-time data to the digital twin modeling module and the risk warning and adaptive control module respectively.

[0008] The digital twin modeling module is used to construct a virtual model corresponding to the physical factory and to simulate risk diffusion, and transmit the risk simulation results to the reinforcement learning decision module.

[0009] The reinforcement learning decision-making module is used to generate strategies for handling various security risks, and generates control commands based on risk simulation results and transmits them to the risk warning and adaptive control module.

[0010] The risk warning and adaptive control module is used to identify anomalies in real-time data and generate warning signals, execute received control commands, transmit anomaly signals to the digital twin modeling module, and transmit control execution status data to the human-computer interaction module.

[0011] The human-computer interaction module is used to provide a visual operation interface and record system operation information, receive operation instructions from operators and send them to the corresponding modules.

[0012] Furthermore, the data preprocessing and storage module performs the following operations during data processing and transmission:

[0013] Receive raw data packets sent by the safety instrument network acquisition module, unpack and verify the format of the data packets, and discard data packets that fail the verification.

[0014] The original data that has passed the verification is cleaned to remove noise and abrupt values. The cleaned data is then normalized and converted to a unified format.

[0015] The processed data is written to a distributed time-series database for persistent storage, and the real-time processed data is synchronously sent to the digital twin modeling module and the risk warning and adaptive control module.

[0016] Furthermore, the digital twin modeling module performs the following operations during model running and risk simulation:

[0017] A three-dimensional digital twin model is constructed based on the equipment parameters, layout structure and process flow of the physical factory. Real-time data sent by the data preprocessing and storage module is received, and the operating parameters of the corresponding nodes in the model are updated.

[0018] Upon receiving an abnormal signal from the risk warning and adaptive control module, initiate the risk diffusion simulation program;

[0019] The simulation process of the accident is based on the current model state, generating data on the scope of the accident's impact and its development trend. The risk simulation results are then sent to the reinforcement learning decision module.

[0020] Furthermore, the reinforcement learning decision-making module performs the following operations during policy training and instruction generation:

[0021] A reinforcement learning agent is constructed using a digital twin model as the interaction environment, and the decision network parameters of the agent are initialized.

[0022] Historical fault data is imported to generate various fault training scenarios. Iterative training is performed through the interaction between the agent and the digital twin environment, adjusting the weights of the decision network. During iterative training, the Q-value of the agent is updated using a temporal difference update method, calculated as follows: Where α is the learning rate, determined by the required convergence speed of the reinforcement learning agent, and γ is the discount factor, determined by the degree of influence of future rewards on the current decision. Let t be the environmental state. Let t be the action performed by the agent. To perform the action The instant reward obtained afterward To perform the action The environmental state at the next moment after transition;

[0023] Generate a library of handling strategies for various security risks, receive risk simulation results sent by the digital twin modeling module, match the corresponding control strategies from the strategy library, and generate specific control instructions.

[0024] Furthermore, the safety instrument networking acquisition module supports industrial Ethernet, Modbus, and Profinet communication protocols, and connects pressure transmitters, temperature sensors, combustible gas detectors, and level gauges in the production site to form a network. It acquires equipment operating parameters, production status parameters, and environmental safety parameters according to a preset acquisition frequency, encapsulates the acquired raw data into data packets of a unified format, and transmits them to the data preprocessing and storage module through an industrial Ethernet link.

[0025] Furthermore, the risk warning and adaptive control module incorporates a deep learning-based anomaly detection model. It analyzes the input real-time data stream frame by frame, extracting temporal features and multi-parameter correlation features. The extracted features are then compared with a pre-built normal operating condition feature library to calculate the feature deviation value. The calculation formula is as follows: ,in, The weight of the i-th feature is determined by principal component analysis based on historical normal operating data. For the i-th feature value extracted in real time, The mean of the i-th feature in the normal operating condition feature library is given, and n is the total number of feature dimensions. When the feature deviation exceeds the preset range, an anomaly confirmation signal is generated. At the same time, the risk level is divided according to the degree of feature deviation and a corresponding warning signal is generated.

[0026] Furthermore, the risk warning and adaptive control module receives control commands sent by the reinforcement learning decision module, parses and verifies the legality of the control commands, and after the verification is passed, sends corresponding control commands to the safety instrument system and distributed control system through the industrial control bus to control the field equipment to perform corresponding actions, while feeding back the real-time status data of the control execution to the human-machine interaction module.

[0027] Furthermore, the human-computer interaction module provides a three-dimensional visualization operation interface, which displays the operating status, equipment parameters and process flow of the digital twin model in real time, displays the level and location information of risk warnings, displays the execution progress and results of control commands, supports operators to view historical operating data and historical operation records, supports operators to modify warning threshold parameters, supports operators to issue manual control commands, and automatically records all system operation logs and operation logs.

[0028] Furthermore, the digital twin modeling module receives real-time data sent by the data preprocessing and storage module at preset time intervals, updates the operating parameters of the corresponding equipment nodes and process nodes in the digital twin model, and calibrates the overall operating status of the model to keep the virtual model synchronized with the operating status of the physical factory. When equipment change information is received, the structure and parameters of the corresponding equipment in the digital twin model are automatically updated.

[0029] Furthermore, after each control command is executed, the reinforcement learning decision-making module receives control execution result data from the risk warning and adaptive control module, converts the execution result data into a corresponding reward signal, and inputs it into the reinforcement learning agent. The formula for calculating the immediate reward is: ,in, To control cost weighting, it is determined by the equipment operation and maintenance cost standards at the industrial production site. Safety risk weights are determined based on statistical data of losses caused by historical accidents. Let t be the control cost incurred by executing the control action at time t. Given the remaining safety risk value of the system at time t, the decision network of the agent is iteratively updated, and the corresponding risk handling strategies in the handling strategy library are updated at the same time, so that the strategy library can continuously adapt to changes in production conditions.

[0030] Compared with existing technologies, this safety instrumented network data acquisition and big data analysis early warning system has the following advantages:

[0031] This invention achieves comprehensive, real-time acquisition and unified management of various safety parameters at the production site by constructing a safety instrument network acquisition module. Through a digital twin modeling module, it realizes real-time synchronous mapping and risk diffusion simulation between the physical factory and the virtual system, enabling early identification of hidden safety risks and prediction of accident development trends. A reinforcement learning decision-making module generates dynamically optimized control strategies, solving the problem that traditional fixed strategies are difficult to adapt to complex working conditions. A risk warning and adaptive control module enables the coordinated execution of warnings and controls, shortening response time and avoiding interference from human factors. This effectively solves the problems of delayed warnings, rigid control strategies, and disconnect between warnings and controls in existing technologies, achieving intelligent closed-loop management of safety instrument data throughout the entire process from monitoring, analysis, warning to control, thereby improving the safety and reliability of industrial production.

[0032] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0033] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0034] Figure 1 This is a schematic diagram of a safety instrument network data acquisition and big data analysis early warning system.

[0035] Figure 2 This is a flowchart of a safety instrument network data acquisition and big data analysis early warning system. Detailed Implementation

[0036] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0037] The safety instrument networking data acquisition and big data analysis early warning system provided by this invention is a complete intelligent safety management and control solution for the entire process of industrial production sites. The system uses a safety instrument networking data acquisition module to achieve interconnection and real-time acquisition of production safety parameters from multiple types of safety instruments on-site. A data preprocessing and storage module cleans, normalizes, and persistently stores the raw data. A digital twin modeling module constructs a three-dimensional virtual model of the physical factory and dynamically simulates the spread of safety risks. A reinforcement learning decision-making module trains the intelligent decision-making network and generates dynamic control strategies. A risk early warning and adaptive control module intelligently identifies abnormal production parameters and automatically executes control commands. A human-machine interaction module provides three-dimensional visualization and manual intervention. The six modules work together in accordance with the data flow logic to form a complete closed-loop system of data acquisition, preprocessing, modeling and analysis, intelligent decision-making, early warning and control, and interactive management. The system supports multiple industrial communication protocols to complete instrument networking, uses deep learning technology to achieve multi-parameter coupled anomaly detection, uses the temporal differential update method to complete the iterative optimization of the reinforcement learning agent, and completes the classification of safety risk levels through feature deviation quantification calculation. Ultimately, it realizes accurate early warning of industrial safety risks, intelligent control during the event, and iterative optimization of post-event strategies, fundamentally solving the core technical problems of delayed early warning, rigid control strategies, and disconnect between early warning and control links in traditional safety instrument management systems.

[0038] This embodiment selects a continuous production site in a large chemical enterprise as the application scenario. This site is equipped with various types of safety instruments, including pressure transmitters, temperature sensors, combustible gas detectors, and level gauges. It involves high-temperature, high-pressure, and flammable / explosive production processes, placing extremely high demands on the real-time performance and intelligence of safety monitoring and risk control. The overall system structure is as follows: Figure 1 As shown, the overall workflow is as follows: Figure 2 As shown in the attached diagrams and actual field applications, the specific implementation process of this system will be described in detail below.

[0039] In the initial stage of implementation, this embodiment first completes the on-site deployment of system hardware. Industrial Ethernet switches, industrial control acquisition terminals, and communication gateways are deployed at the chemical production site, while high-performance industrial servers, distributed time-series database storage devices, and digital twin rendering terminals are configured in the central control room. After the hardware deployment is completed, the software initialization of the six core modules is completed sequentially to ensure smooth communication links between modules and data transmission protocol compatibility, laying the foundation for subsequent data acquisition and control execution.

[0040] The safety instrumented network (SAN) data acquisition module serves as the system's data entry point. During the initialization phase, it completes communication protocol configuration and instrument network matching. This module supports three mainstream industrial communication protocols: Industrial Ethernet, Modbus, and Profinet. It adapts the protocol to the communication type of the safety instruments in the field, integrating distributed pressure transmitters, temperature sensors, combustible gas detectors, and level gauges into a unified network system. The module is configured with a fixed acquisition frequency according to the safety management requirements of chemical production, continuously acquiring equipment operating parameters, production status parameters, and environmental safety parameters. The acquired raw data is encapsulated into data packets according to a unified format and stably transmitted to the data preprocessing and storage module via an industrial Ethernet link, achieving seamless and uninterrupted acquisition and transmission of field safety data.

[0041] After receiving the raw data packets transmitted by the safety instrumented network (SAN) acquisition module, the data preprocessing and storage module initiates a standardized data processing flow. First, the data packets are unpacked, and format verification is performed on each packet. Invalid data packets that fail verification are automatically discarded, ensuring that the data entering subsequent processing stages has basic validity. The verified raw data enters the cleaning stage, where the module uses algorithms to remove random noise and abnormal fluctuations, eliminating data distortion caused by on-site electromagnetic interference and equipment fluctuations. After cleaning, the data is normalized, converting safety parameters with different ranges and units into a unified numerical range. Simultaneously, a unified data format conversion is performed to ensure that the data is compatible with the input requirements of the digital twin modeling module and the risk warning module.

[0042] The processed data is transmitted and stored via two paths. One path is written to a distributed time-series database for persistent storage, providing data support for subsequent historical data queries, strategy training, and incident tracing. The other path is sent synchronously as real-time data to the digital twin modeling module and the risk warning and adaptive control module, ensuring that the two core analysis modules can obtain the latest on-site operational data for real-time analysis and control. The entire preprocessing process is fully automated without human intervention, ensuring the efficiency and accuracy of data processing and providing a high-quality data foundation for subsequent intelligent analysis by the system.

[0043] The digital twin modeling module is the core component connecting the physical site and virtual analysis. During implementation, it first constructs a 3D digital twin model that perfectly corresponds to the physical plant, based on the equipment parameters, layout structure, and process flow of the chemical production site. The model accurately reproduces the location of on-site equipment, the routing of process pipelines, and the production logic. The module receives real-time data from the data preprocessing and storage module at preset time intervals, automatically updating the operating parameters of corresponding equipment and process nodes in the model. Simultaneously, it calibrates the overall operating status of the model to ensure a high degree of synchronization between the virtual model and the real-time operating status of the physical plant. When on-site equipment undergoes replacement or modification, the module automatically receives the equipment change information and synchronously updates the structure and parameters of the corresponding equipment in the digital twin model, ensuring the model's timeliness and accuracy.

[0044] When the risk warning and adaptive control module detects abnormal production parameters and sends an anomaly signal, the digital twin modeling module immediately initiates the risk diffusion simulation program. Based on the real-time state of the current virtual model and the type and severity of the abnormal parameters, the module simulates the entire process of the occurrence, development, and diffusion of a safety accident, generating key data such as the accident's impact range, diffusion speed, and development trend. The complete risk simulation results are then transmitted to the reinforcement learning decision-making module, providing a visualized and data-driven simulation basis for intelligent decision-making. This simulation process does not require actual resources from the physical production site and can complete full-scenario risk simulations in a virtual environment, effectively avoiding safety hazards associated with simulation testing under actual working conditions.

[0045] The reinforcement learning decision-making module uses a digital twin model as the interaction environment. First, it completes the construction of the reinforcement learning agent and the initialization of the decision network parameters, providing basic model support for intelligent decision-making. Then, it imports historical fault data from the chemical production site to generate various fault training scenarios such as leakage, over-temperature, and over-pressure, allowing the reinforcement learning agent to continuously interact and iterate with the digital twin virtual environment for training.

[0046] During iterative training, the Q-value of the agent is updated using a temporal difference update method. In the specific implementation of this embodiment, the Q-value update formula is as follows: Here, α is the learning rate, which is determined based on the required convergence speed of the reinforcement learning agent. By adjusting the learning rate, the magnitude of parameter updates in each iteration of the agent is controlled. The faster the training convergence speed, the more the learning rate setting fits the control requirements of rapid iteration. γ is the discount factor, which is determined based on the degree of influence of future rewards on the current decision. The higher the weight of the influence of future rewards on the current decision, the more the discount factor is set to emphasize long-term safety gains. Let t represent the environmental state at time t, indicating the real-time operational status of the chemical production site in the digital twin model. Let t be the action performed by the agent at time t, corresponding to the handling of security risks. To perform the action The immediate reward obtained afterward is calculated based on both control costs and security risk values. To perform the action The environmental state at the next moment is then transitioned to, reflecting the changes in on-site operation after the control operation. Through continuous iterative updates, the decision network weights of the intelligent agent are continuously optimized, gradually forming a decision-making capability adapted to complex chemical working conditions.

[0047] After training, the module generates a library of handling strategies for various security risks. When it receives the risk simulation results sent by the digital twin modeling module, it automatically matches the optimal control strategy from the strategy library, generates specific control instructions that can be directly executed, and transmits them to the risk warning and adaptive control module.

[0048] After each control command is executed, the reinforcement learning decision module receives control execution result data from the risk warning and adaptive control module, converts the execution result into a reward signal, and inputs it into the reinforcement learning agent. In the specific implementation of this embodiment, the instantaneous reward calculation formula is as follows: , in the formula To control cost weighting, this parameter is determined based on the equipment operation and maintenance cost standards of industrial production sites. It is quantified by statistically analyzing the average operation and maintenance costs of on-site equipment start-up, shutdown, parameter adjustment, and other operations, ensuring that the weighting is consistent with the actual cost level on-site. As the safety risk weight, this parameter is determined based on the statistical data of losses caused by historical accidents. It is quantified by summarizing data such as personnel injuries, property damage, and production interruptions from historical safety accidents, so as to fully reflect the actual impact of safety risks. The control cost incurred by executing control actions at time t is the sum of all costs, including equipment operation, energy consumption, and maintenance investment. The remaining safety risk value of the system at time t is determined jointly by the anomaly detection results of the risk warning module and the simulation data of the twin model. The module iteratively updates the agent decision-making network based on the reward signal, synchronously updating the corresponding risk handling strategies in the handling strategy library. This allows the strategy library to continuously adapt to the dynamic changes in chemical production conditions, achieving self-optimization of the decision-making strategies.

[0049] The risk warning and adaptive control module is the core of the system to realize the linkage between warning and control. The module has a built-in deep learning-based anomaly detection model, which analyzes the real-time data stream input from the data preprocessing and storage module frame by frame, automatically extracts the temporal features and multi-parameter correlation features of the data, and accurately compares the extracted features with the pre-built normal operating condition feature library. The feature deviation value is used to determine whether the production status is abnormal.

[0050] In the specific implementation of this embodiment, the formula for calculating the feature deviation value is as follows: , in the formula The weight of the i-th feature is determined by principal component analysis of historical normal operating data. Principal component analysis will screen out features that have a significant impact on the judgment of operating conditions and assign corresponding weights. The weight is positively correlated with the importance of the feature to safety management. The i-th feature value extracted in real time represents the real-time characteristic performance of the on-site safety parameters. is the mean of the i-th feature in the normal operating condition feature library, calculated statistically from long-term stable operating condition data at the chemical plant, reflecting the normal operating level of the parameter. n is the total number of feature dimensions, determined based on the type of safety parameters collected on-site and the number of features extracted.

[0051] When the calculated feature deviation value exceeds the preset range, the module immediately generates an anomaly confirmation signal. Simultaneously, it classifies the safety risk level based on the degree of feature deviation, generates a corresponding warning signal, and pushes it to the human-machine interface module. Upon receiving the control command from the reinforcement learning decision module, the module first parses and verifies the command's validity. If the verification is successful, it sends corresponding control commands to the safety instrumented system and distributed control system via the industrial control bus. This controls the field equipment to perform targeted actions such as valve adjustment, shutdown, and pressure relief, achieving seamless linkage between warning and control. During control execution, the module collects control execution status data in real time and synchronously feeds it back to the human-machine interface module, allowing operators in the control room to monitor the control execution progress and results in real time.

[0052] The human-machine interface module provides the system with an intuitive operation and management interface. Equipped with a 3D visualization interface, it displays the real-time operating status, equipment parameters, and process flow of the digital twin model. It simultaneously displays the level, location, and detailed information of risk warnings, clearly presenting the execution progress and results of control commands. Operators can view historical operating data and operation records at any time through the interface, tracing the production process and control operations. The module supports operators modifying warning threshold parameters according to production process adjustment needs, and can also issue manual control commands under special operating conditions. Manual commands are verified by the module and then sent to the corresponding execution module, enabling flexible switching between automatic control and manual intervention. Simultaneously, the module automatically records all system operation logs, storing the log data in real time, providing complete evidence for system operation and maintenance, accident analysis, and compliance management.

[0053] like Figure 2 As shown, this system forms a complete closed-loop operation process in the chemical production site. The safety instrument network acquisition module continuously collects on-site data, which is then preprocessed and transmitted to the analysis and control module. The digital twin module completes virtual mapping and risk simulation, the reinforcement learning module generates intelligent decisions, the early warning and control module executes anomaly detection and automatic control, the human-machine interaction module completes display and management, and the control execution results are fed back to the reinforcement learning module to complete strategy optimization. Abnormal data triggers the twin model simulation again, forming a fully intelligent closed loop of data acquisition, analysis, early warning, control, and optimization, which can complete all-time safety management without continuous human intervention.

[0054] In summary, this embodiment applies the safety instrumented data acquisition and big data analysis early warning system to high-risk production sites in chemical enterprises, fully realizing all the functions of the patented technical solution and achieving significant control effects. The system achieves comprehensive real-time acquisition and unified management of on-site safety parameters through multi-protocol instrument networking, solving the problem of data silos in traditional instruments. The digital twin modeling module achieves real-time synchronization between the physical plant and the virtual system, accurately completing risk diffusion simulation, identifying hidden safety risks such as parameter drift and multi-parameter coupling in advance, and accurately predicting accident development trends. The reinforcement learning decision-making module generates dynamically adaptable control strategies to complex operating conditions through temporal differential updates and reward feedback optimization, completely solving the problems of insufficient or excessive control by traditional fixed strategies. The risk early warning and adaptive control module relies on deep learning anomaly detection and feature deviation calculation to achieve accurate risk classification and early warning. The linkage between early warning and control significantly shortens response time and completely avoids control delays and errors caused by human factors. The system as a whole has realized the transformation of industrial safety monitoring from passive response to proactive prevention. The intelligent closed-loop management of the entire process has effectively reduced the probability of production safety accidents, improved the efficiency and reliability of safety control at chemical production sites, and provided intelligent and efficient technical support for the safe production of industrial enterprises.

[0055] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A safety instrumented networking collection and big data analysis early warning system, characterized in that, The system includes: a safety instrument networking acquisition module, a data preprocessing and storage module, a digital twin modeling module, a reinforcement learning decision-making module, a risk warning and adaptive control module, and a human-computer interaction module; The safety instrument network acquisition module is used to collect safety parameters from the production site and transmit the collected raw data to the data preprocessing and storage module. The data preprocessing and storage module is used to process and store the raw data, and transmit the processed real-time data to the digital twin modeling module and the risk warning and adaptive control module respectively. The digital twin modeling module is used to construct a virtual model corresponding to the physical factory and to simulate risk diffusion, and transmit the risk simulation results to the reinforcement learning decision module. The reinforcement learning decision-making module is used to generate strategies for handling various security risks, and generates control commands based on risk simulation results and transmits them to the risk warning and adaptive control module. The risk warning and adaptive control module is used to identify anomalies in real-time data and generate warning signals, execute received control commands, transmit anomaly signals to the digital twin modeling module, and transmit control execution status data to the human-computer interaction module. The human-computer interaction module is used to provide a visual operation interface and record system operation information, receive operation instructions from operators and send them to the corresponding modules.

2. The safety instrumented networking collection and big data analysis early warning system according to claim 1, characterized in that, The data preprocessing and storage module performs the following operations during data processing and transmission: Receive raw data packets sent by the safety instrument network acquisition module, unpack and verify the format of the data packets, and discard data packets that fail the verification. The original data that has passed the verification is cleaned to remove noise and abrupt values. The cleaned data is then normalized and converted to a unified format. The processed data is written to a distributed time-series database for persistent storage, and the real-time processed data is synchronously sent to the digital twin modeling module and the risk warning and adaptive control module.

3. The safety instrumented network collection and big data analysis and early warning system of claim 1, wherein, The digital twin modeling module performs the following operations during model running and risk simulation: A three-dimensional digital twin model is constructed based on the equipment parameters, layout structure and process flow of the physical factory. Real-time data sent by the data preprocessing and storage module is received, and the operating parameters of the corresponding nodes in the model are updated. Upon receiving an abnormal signal from the risk warning and adaptive control module, initiate the risk diffusion simulation program; The simulation process of the accident is based on the current model state, generating data on the scope of the accident's impact and its development trend. The risk simulation results are then sent to the reinforcement learning decision module.

4. The safety instrumented networking collection and big data analysis early warning system according to claim 1, characterized in that, The reinforcement learning decision module performs the following operations during policy training and instruction generation: A reinforcement learning agent is constructed using a digital twin model as the interaction environment, and the decision network parameters of the agent are initialized. Historical fault data is imported to generate various fault training scenarios. Iterative training is performed through the interaction between the agent and the digital twin environment, adjusting the weights of the decision network. During iterative training, the agent's Q-value is updated using a temporal difference update method, calculated as follows: Where α is the learning rate and γ is the discount factor. Let t be the environmental state. Let t be the action performed by the agent. To perform the action The instant reward obtained afterward To perform the action The environmental state at the next moment after transition; Generate a library of handling strategies for various security risks, receive risk simulation results sent by the digital twin modeling module, match the corresponding control strategies from the strategy library, and generate specific control instructions.

5. A safety instrument network data acquisition and big data analysis early warning system according to claim 1, characterized in that, The safety instrument networking acquisition module supports industrial Ethernet, Modbus, and Profinet communication protocols. It connects pressure transmitters, temperature sensors, combustible gas detectors, and level gauges in the production site to form a network. It acquires equipment operating parameters, production status parameters, and environmental safety parameters according to a preset acquisition frequency. The acquired raw data is encapsulated into data packets of a unified format and transmitted to the data preprocessing and storage module through an industrial Ethernet link.

6. The safety instrument network data acquisition and big data analysis early warning system according to claim 1, characterized in that, The risk warning and adaptive control module incorporates a deep learning-based anomaly detection model. It analyzes the input real-time data stream frame by frame, extracting temporal features and multi-parameter correlation features. The extracted features are then compared with a pre-built normal operating condition feature library to calculate the feature deviation value. The calculation formula is as follows: ,in, Let i be the weight of the i-th feature. For the i-th feature value extracted in real time, The mean value of the i-th feature in the normal operating condition feature library is given, and n is the total number of feature dimensions. When the feature deviation exceeds the preset range, an anomaly confirmation signal is generated. At the same time, the risk level is divided according to the degree of feature deviation and a corresponding warning signal is generated.

7. A safety instrument network data acquisition and big data analysis early warning system according to claim 1, characterized in that, The risk warning and adaptive control module receives control commands sent by the reinforcement learning decision module, parses and verifies the legality of the control commands, and sends corresponding control commands to the safety instrument system and distributed control system through the industrial control bus after the verification is passed, controls the field equipment to perform corresponding actions, and feeds back the real-time status data of the control execution to the human-machine interaction module.

8. A safety instrument network data acquisition and big data analysis early warning system according to claim 1, characterized in that, The human-machine interaction module provides a three-dimensional visualization operation interface, which displays the operating status, equipment parameters and process flow of the digital twin model in real time, displays the level and location information of risk warnings, displays the execution progress and results of control commands, supports operators to view historical operating data and historical operation records, supports operators to modify warning threshold parameters, supports operators to issue manual control commands, and automatically records all system operation logs and operation logs.

9. A safety instrument network data acquisition and big data analysis early warning system according to claim 1, characterized in that, The digital twin modeling module receives real-time data sent by the data preprocessing and storage module at preset time intervals, updates the operating parameters of the corresponding equipment nodes and process nodes in the digital twin model, and calibrates the overall operating status of the model to keep the virtual model synchronized with the operating status of the physical factory. When equipment change information is received, the structure and parameters of the corresponding equipment in the digital twin model are automatically updated.

10. A safety instrument network data acquisition and big data analysis early warning system according to claim 1, characterized in that, After each control command is executed, the reinforcement learning decision-making module receives control execution result data from the risk warning and adaptive control modules, converts the execution result data into a corresponding reward signal, and inputs it into the reinforcement learning agent. The formula for calculating the immediate reward is as follows: ,in, To control cost weighting, For safety risk weighting, Let t be the control cost incurred by executing the control action at time t. Given the remaining safety risk value of the system at time t, the decision network of the agent is iteratively updated, and the corresponding risk handling strategies in the handling strategy library are updated simultaneously, so that the strategy library continuously adapts to changes in production conditions.

Images