Pharmaceutical workshop risk early warning large model identification and processing method

Abstract

The invention belongs to the field of risk early warning, and relates to a pharmaceutical workshop risk early warning large model identification and processing method, which comprises the following steps: constructing a risk data system, and carrying out fusion training on risk data and safety production specifications; acquiring real-time production data, and performing real-time detection of a production risk index through the real-time production data; according to historical production data and a company management system, a large risk processing model is constructed, and a multi-modal generation and processing strategy of data distribution and early warning is carried out.

Description

Technical Field

[0001] This invention belongs to the field of risk warning, and in particular, it is a method for identifying and processing risks in pharmaceutical workshops using a large-scale risk warning model. Background Technology

[0002] Production safety is the cornerstone of enterprise development. Safety production risk management is the foundation for ensuring harmonious development and sustainable operation of an enterprise. It is also the basic work for building harmonious labor relations and maintaining the enterprise's image. Furthermore, it is an important guarantee for promoting enterprise development and ensuring long-term prosperity. The key to safety risk management is how to identify, warn, assess, and respond to safety risks in a timely and accurate manner. Safety production risk management is a systematic and dynamic project involving all employees of the enterprise. All levels, departments, and positions within the enterprise must clearly define their own safety production risk management tasks and requirements. Through continuous improvement, they can shift from passive response to proactive prevention, thereby gradually establishing an enterprise safety production risk management system.

[0003] Existing technologies disclose intelligent identification methods and systems for chemical safety production risks based on knowledge graphs, including: knowledge graph construction and updating; knowledge reasoning and retrieval; and risk identification. Specifically for the chemical industry, a directed graph knowledge base is formed by integrating multi-source chemical safety knowledge, such as chemical processes, equipment operation and experimental data, laws and regulations, operating procedures, process documents, and emergency plans. This creates a knowledge graph of risks and safety hazards during chemical processes and equipment operation, enabling the fusion of multi-source heterogeneous knowledge. Knowledge reasoning is performed based on graph matching and graph neural network methods. Real-time acquired operating parameters and other data are compared with data in the safety knowledge base through the knowledge graph for reasoning and analysis, achieving risk identification and analysis based on graph matching and graph neural networks, thus improving the accuracy of risk identification.

[0004] However, existing technologies for handling chemicals only focus on the degree of hazard of the chemicals themselves, neglecting production equipment, the production environment, and human factors. They lack multi-dimensional early warning systems for chemicals, the environment, and equipment. Furthermore, existing technologies do not classify risks according to their severity, failing to provide differentiated measures based on risk levels. They also lack systematic explanations on how to eliminate risks and how to improve the efficiency of risk mitigation and resolution, hindering risk managers' judgment and management. Therefore, those skilled in the art provide a large-scale model for risk early warning identification and processing in pharmaceutical workshops to address the problems mentioned in the background. Summary of the Invention

[0005] This invention proposes a method for identifying and processing risks in pharmaceutical workshops using a large-scale risk warning model.

[0006] A method for identifying and processing risks in a pharmaceutical workshop using a large-scale risk early warning model includes the following: constructing a risk data system and integrating the risk data with safety production standards for training; Acquire real-time production data and use it to detect production risk indices in real time. Based on historical production data and the company's management system, a large-scale risk management model is constructed to implement multimodal generation and processing strategies for data distribution and early warning.

[0007] Preferably, the construction of the risk data system and the integration and training of risk data with safety production standards include the following: The sources of risk include material risk, equipment risk, environmental risk, and operational risk. In the construction of the risk data system, the data includes safety production standards, historical production data, and corresponding data related to the risk content; the fusion training includes merging risk data with safety production standards to form a structured risk knowledge graph; Knowledge is extracted from the risk knowledge graph using a BERT-based entity recognition model, knowledge is fused using entity linking technology, and knowledge reasoning is performed using graph neural networks, ultimately enabling knowledge application.

[0008] Preferably, the acquisition of real-time production data and the real-time detection of production risk index through real-time production data include the following: the acquisition of real-time production data includes collection by the Internet of Things system and acquisition of special data, wherein the special data includes chemical raw materials, production equipment and production environment; Based on real-time production data, real-time risk prediction is performed through a large model, and a real-time risk index is calculated by combining risk weights and severity, including chemical handling risks, equipment malfunction risks, and environmental exceedance risks. Multimodal early warning data is distributed according to risk type, including text reports, image reports, audio reports, short video reports, and animation reports.

[0009] Preferably, the acquisition of real-time production data includes collection through the Internet of Things system and acquisition of specific data. For chemical raw material management, relevant information of each batch of raw materials is entered through the electronic code of chemicals to establish a material management system and form a dynamic material batch and production batch relationship database. The production speed of batch materials is controlled by uploading and verifying the production and inventory status of each batch of materials in real time.

[0010] Preferably, the acquisition of real-time production data includes collection by the Internet of Things system and acquisition of specialized data. For the monitoring and maintenance of production equipment, various production parameters are monitored in real time. During the production process, production abnormalities of production equipment are detected through intelligent risk warning models and manual inspections, and targeted maintenance solutions are proposed to provide auxiliary decision-making for production equipment management.

[0011] Preferably, the acquisition of real-time production data includes collection by the Internet of Things system and acquisition of specialized data. Among these, the monitoring and continuous improvement of the production environment involves real-time monitoring of the safety status of the production environment, generation of production environment indicator reports, continuous improvement, establishment of an environmental management system, recording production environment data in different regions, locations, and time periods, and realizing real-time monitoring, management, and analysis of production environment data.

[0012] Preferably, the construction of a large-scale risk management model and the multimodal generation and processing strategy for data distribution and early warning include the following: risk identification through the large-scale risk management model; risk assessment through the large-scale risk management model; generation and release of early warnings through risk managers; and elimination and rectification of risks through risk management strategies.

[0013] Preferably, the risk identification achieved through the risk processing big model includes data processing and identification, and data processing includes real-time data filtering, transformation, and cleaning to form standardized data. The identification process includes analyzing and filtering out abnormal data using big data identification algorithms; conducting supervised learning using a Transformer-based deep neural network model combined with historical data and knowledge graphs; and extracting risk features through feature engineering to train a risk inference model library.

[0014] Preferably, the risk assessment using a large-scale risk processing model includes calculating the real-time risk index RI and classifying risks according to the real-time risk index RI. The risk assessment process specifically includes: calculating the real-time risk index RI using the large-scale model and outputting it in real time; making a preliminary assessment by comparing it with a preset threshold; further reviewing and confirming it manually; and generating early warning information based on the final assessment results.

[0015] Preferably, the generation and release of early warnings through risk managers includes multimodal early warnings, including text, images, audio, short videos, and animations; it also includes graded processing of early warnings, implementing different levels of processing schemes according to the severity of the early warning; and it is released through system messages, SMS, security broadcasts, and manual notifications to form a closed loop for emergency response.

[0016] The present invention has the following beneficial effects: 1. This invention improves the efficiency of risk management by using the Internet of Things and specialized data to monitor risks in real time, analyzing environmental and production variables, and responding to emergencies in real time.

[0017] 2. This invention establishes a risk data system that corresponds to relevant safety production regulations and rules, conducts risk assessment, establishes a multi-dimensional risk data system, provides data analysis basis, and conducts graded early warning to reduce production risks.

[0018] 3. This invention integrates multimodal data through a large model to form a unified decision support tool, which automatically judges various situations and provides manual intervention for uncertain situations, thereby improving risk management efficiency and making risk management decisions traceable. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the method for identifying and processing risks in a pharmaceutical workshop using a large-scale risk warning model according to the present invention. Figure 2 This is a schematic diagram illustrating the specific content of the large-scale model identification and processing method for risk early warning in pharmaceutical workshops according to the present invention. Detailed Implementation

[0020] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions in the embodiments of this invention will be clearly described below in conjunction with the examples.

[0021] like Figure 1 , 2 As shown, this invention proposes a method for identifying and processing risks in pharmaceutical workshops using a large-scale risk early warning model, comprising the following steps: Build a risk data system and integrate various risk data with safety production standards for training. Real-time production data is acquired through the IoT system in the pharmaceutical workshop, including equipment operation data, material loading and unloading data, material handling data, temperature and humidity data, and hazardous materials transportation and handling data. Production risk indices are monitored in real time, and large-scale models are used to predict potential production risks, particularly chemical handling risks, and risk warning data is generated in real time. Based on historical production data and the company's management system, a large-scale risk management model is constructed to implement multimodal generation and processing strategies for data distribution and early warning.

[0022] For chemical raw material management, relevant information for each batch of raw materials is entered through electronic chemical codes to establish a material management system, forming a dynamic database of material batches and production batches. By uploading and verifying the production and inventory status of each batch of materials in real time, the production speed of batch materials is controlled.

[0023] The information related to raw materials includes the following: Basic Attributes: This includes the standard name of the chemical raw material, such as "sulfuric acid" or "sodium hydroxide," which is the basis for accurate identification. Specifications include analytical grade, superior grade, etc., for chemical reagents, and concentration for liquid raw materials, such as "36% hydrochloric acid." There are also model numbers; for example, some plastic raw materials have different models due to performance differences, suitable for different product production. Batch Characteristics: Each batch of raw material has a unique batch number for full-process tracking; the production date records the time of raw material production, which is related to determining the shelf life; the expiration date clearly states how long the raw material maintains its quality under specified storage conditions; expired raw materials may affect production or even pose risks. Quality Verification: This involves the quality inspection report number, allowing for quick access to detailed test data; the test items and results, such as purity, impurity content, pH, etc., determine whether the raw material meets standards; the quality qualification status visually indicates whether the batch of raw material can be used. Supplier Details: Provides the supplier's name for easy communication in case of problems; contact information includes address, telephone, and email to ensure smooth communication; qualification certificates such as production and operation licenses ensure the legality and compliance of the raw material source. Safety Points: Describe the hazardous characteristics of the raw materials, such as flammability, explosiveness, toxicity, and corrosiveness, like "ethanol is flammable" and "concentrated sulfuric acid is highly corrosive"; explain safe operating procedures, such as the protective equipment to be worn during operation and environmental requirements; and also provide emergency handling methods, including emergency response measures for accidents such as leaks and fires.

[0024] For the monitoring and maintenance of production equipment, various production parameters are monitored in real time. During the production process, abnormal production problems of production equipment are detected through intelligent risk early warning models and manual inspections, and targeted maintenance solutions are proposed to provide auxiliary decision-making for production equipment management.

[0025] Among these, production anomalies include: 1) Abnormal Equipment Operating Parameters: Abnormal Temperature: For example, the operating temperature of reactors, drying equipment, etc., exceeds the set range. In the pharmaceutical industry, excessively high reactor temperatures may accelerate the reaction, leading to decreased product quality or even safety accidents; excessively low temperatures result in incomplete reactions, affecting production efficiency and product yield. Abnormal Pressure: Commonly seen in compression equipment, high-pressure reaction equipment, etc. For example, unstable air compressor output pressure affects the normal operation of pneumatic equipment; excessively high pressure in high-pressure reaction equipment poses an explosion risk; insufficient pressure cannot meet production process requirements. Abnormal Speed: The speed of motors, stirring equipment, etc., deviates from the rated value. For example, if the stirring equipment speed is too slow, the materials will not mix evenly, affecting product quality; if the speed is too high, it may damage equipment components.

[0026] 2) Equipment Mechanical Failures: Component Wear: During long-term operation, critical components such as bearings, gears, and chains are prone to wear. For example, worn conveyor chains can break, causing production line interruptions; worn gears reduce transmission accuracy, affecting equipment stability. Equipment Jamming or Stuck: This can be caused by foreign objects entering the equipment or component deformation. For instance, in granule packaging machines, foreign objects mixed in with the material can cause the packaging process to stall; deformed components at the joints of robotic arms can cause jamming, affecting automated production. Equipment Vibration and Abnormal Noise: This can be caused by loose or unbalanced internal components. For example, excessive vibration in a centrifuge may indicate rotor imbalance, which not only affects equipment lifespan but also poses safety hazards; abnormal noise from a motor may indicate bearing damage or winding failure.

[0027] 3) Equipment Electrical Faults: Short circuits or open circuits: Short circuits can be caused by aging electrical wiring, damaged insulation, etc., such as a short circuit in the workshop lighting circuit leading to a partial power outage; open circuits can be caused by loose connections, blown fuses, etc., preventing equipment from operating normally, such as a conveyor belt motor failing to start due to a broken circuit. Control System Faults: Faults in control components such as PLCs (Programmable Logic Controllers) and touch screens can cause equipment control commands to fail to execute or execute incorrectly. For example, in an automated production line, a PLC program error can cause equipment operation sequences to become disordered, affecting the production process. Sensor Faults: Faulty temperature, pressure, and flow sensors can cause equipment to acquire incorrect data, thus affecting control decisions. For example, a faulty temperature sensor can cause the reaction vessel temperature to display incorrectly, leading operators to adjust settings based on erroneous data, affecting product quality.

[0028] 4) Abnormal production efficiency: Decreased capacity: The actual output of the equipment is lower than the designed capacity. For example, a filling machine that originally filled 1,000 bottles per hour can now only fill 800 bottles. This may be due to equipment aging, unreasonable process parameters, or poor material supply. Unstable production speed: The production speed of the equipment varies. For example, the printing speed of a printing machine is unstable, affecting the consistency of product quality. This may be due to a failure in the transmission system or an unstable control system.

[0029] For production equipment management, the automation system, sensor monitoring system, and production equipment management are integrated for unified management; an equipment health management ledger system is established to record information such as equipment failures, maintenance, upkeep, and scrapping, and data-driven management is achieved by utilizing equipment asset value management; a workshop environment monitoring system is established to monitor the workshop's environmental conditions, climate, temperature and humidity, and production status; an equipment production operation database is established to statistically analyze the maintenance, operation, performance, failures, and defects of equipment, enabling scientific maintenance and automated maintenance decision-making methods; equipment asset management is established, based on the asset management dimension, to manage equipment ledgers and achieve full lifecycle management of equipment assets; equipment file management is established to manage the calibration, maintenance, and scrapping of equipment, spare parts, tools, and measuring instruments; and asset model management is established, digitizing asset information into three-dimensional models to achieve three-dimensional management of asset models and realize the informatization of asset management.

[0030] For monitoring and continuous improvement of the production environment, the system monitors the safety status of the production environment in real time, generates production environment indicator reports, and continuously improves them. An environmental management system is established to record production environment data in different regions, locations, and time periods, enabling real-time monitoring, management, and analysis of production environment data.

[0031] The implementation method of the large model is as follows: Construct a risk knowledge graph to structure knowledge, data, relationships and patterns related to safe production, and to structure accidents, safety hazards, early warning information, protective measures, production processes, equipment and materials in the production process; Construct a risk reasoning engine to process the data in the risk knowledge graph through automatic reasoning, including knowledge extraction based on BERT entity recognition model, knowledge fusion based on entity linking technology, knowledge storage, knowledge reasoning based on graph neural network (GNN) and knowledge application.

[0032] For risk data fusion, a knowledge extraction model is used to extract data from different data sources and with different structures. For risk data fusion, a knowledge fusion model is used to fuse and organize knowledge from different sources and with different structures, including entity alignment, attribute alignment, and relationship alignment. For risk data reasoning, a knowledge reasoning model is used to reason about knowledge in a logical or structured way. For risk data application, a knowledge application model is used to output the reasoning results, including alerts, decisions, and security knowledge pushes.

[0033] The application method of the large model includes the following steps: risk identification through the large model; risk assessment through the large model; generation and issuance of early warnings through risk managers; and risk elimination and rectification through risk managers.

[0034] The risk identification process is as follows: Real-time collected data is filtered, transformed, and cleaned to standardize the data processing; big data identification algorithms are used to perform anomaly analysis on the standardized data to identify risk data; large-scale model algorithms are used, employing a Transformer-based deep neural network for model training, combined with historical data and safety production-related knowledge graphs to train a risk knowledge graph, using semi-supervised and weakly supervised learning; a risk data knowledge graph is established, representing historical risk data and existing safety production-related knowledge in a knowledge graph format, summarizing known events and patterns; big data feature engineering is used to extract features from the standardized data, generating corresponding feature datasets, combining them with the knowledge graph to train a risk reasoning model, performing reasoning on actual data, and establishing a risk reasoning model library.

[0035] The risk assessment process is as follows: Real-time calculations are performed using the generated data to determine the current real-time risk indicators. The formula for calculating the real-time risk indicators is:

[0036] in: This is a real-time risk index, with a value range of 0-10; For the first The weights for risk categories are set based on the scope of the risk's impact. ; For the first The severity of a risk class ranges from 0 to 1, where 1 indicates that it may lead to a serious accident and 0 indicates that it has no impact. For the first The probability of occurrence of a risk class is 0-1, predicted and output by a large model. Real-time risk indicators are analyzed, and the results are output through the large model and compared with predefined alarm thresholds, such as low risk ≤ 3, 3 < medium risk ≤ 7, and high risk > 7, with a preliminary judgment made on the results. After manual review and judgment, an early warning message is generated. The system automatically generates tasks based on the early warning message and pushes it to risk management personnel in real time. Risk management personnel then manually review and process the message to determine whether an emergency response is necessary. If so, an emergency response is initiated. According to the emergency response plan, the warning is sent to relevant risk managers or employees, and the warning information is disseminated according to the emergency coordination mechanism, forming a closed loop of emergency response information.

[0037] The generation and dissemination of early warning information includes the following steps: generating multimodal early warnings, including but not limited to text reports, image reports, audio reports, short video reports, and animation reports; multi-level classification of early warning information; and implementing early warning information processing strategies based on different early warning levels.

[0038] It also includes risk management, which includes the following steps: Risk identification, establishing a production process model, using big data processing technology, and comprehensively utilizing technologies such as artificial intelligence, the Internet of Things, and video surveillance to construct data on dynamic material identification, equipment operation data, and personnel on-duty status, analyze real-time production data, and generate process risk indicators; Risk assessment, comprehensively utilizing technologies such as artificial intelligence, the Internet of Things, and video surveillance to establish a production environment data model, analyze real-time production environment data, and assess real-time production environment indicators; Through real-time analysis of the above indicators, using a large model to predict risks, and providing preliminary early warning for potential risks; For risk verification or handling, further manual confirmation is required to confirm the handling measures and provide feedback on the handling strategy; Risk handling is carried out according to the risk level, with higher risk levels requiring more severe handling measures, and vice versa; Risk handling strategies are implemented by assigning, stratifying, and classifying personnel according to different risks.

[0039] For the repair and handling of existing problems, different repair and handling strategies should be formulated for different types of problems; for persistent problems, the problem should be included in the risk analysis and continuously observed or monitored until the problem is completely eliminated.

[0040] For management issues, the results should be promptly sent to the relevant management departments and feedback should be provided to the risk management department. For recurring or frequently occurring issues, relevant suggestions should be made in a timely manner to strengthen management or monitoring.

[0041] Regarding the tiered handling of early warning information, after receiving an early warning, risk managers formulate corresponding response plans based on the warning level. After receiving emergency response information, the risk management department, through its emergency response team, formulates corresponding response plans based on the emergency response level and department. For general early warning issues, the personnel responsible for handling the situation promptly send the information and conduct on-site investigations. Based on the on-site investigation, they formulate risk handling plans and report them to the emergency response team. For more serious early warnings, after receiving the emergency response information, the risk management department and the emergency response team immediately organize personnel for emergency handling and formulate emergency plans.

[0042] Example The method for identifying and processing risks in a pharmaceutical workshop using a large-scale risk warning model in this embodiment includes the following steps: 1) Construct a risk data system and integrate various risk data with safety production standards for training. The risks come from: material risks, equipment risks, and environmental risks. 2) Acquire real-time production data. Through the IoT system in the pharmaceutical workshop, acquire equipment operation data, material loading and unloading data, material handling data, temperature and humidity data, and hazardous materials transportation and handling data during the production process. Detect production risk index in real time and predict potential production risks, especially chemical handling risks, through large-scale models. Generate multimodal data from risk warning data in real time and distribute it to different risk managers. For example, distribute equipment operation risk reports to the equipment management department, distribute material risk parameters to the supply line, and generate environmental videos to the management department. 3) Based on historical production data and the company's management system, construct a large-scale risk management model and implement multimodal generation and processing strategies for data distribution and early warning.

[0043] Specifically, the training and application of the risk warning model includes the following steps: Define risk: Risk refers to various potential hazards in production activities. We comprehensively summarize the risk data that may be involved in the production process, including but not limited to production materials, production equipment, production environment, personnel health, and production process, to form a complete set of risk data for large-scale model training. Risk data fusion: Integrate the risk data system with safety production standards to establish a risk management data system, form a risk management model, and perform risk knowledge graph processing on the risk management data system; Risk Data Learning: Using a deep learning framework, a large-scale risk model is constructed. Knowledge extraction, fusion, reasoning, and application are performed on the risk knowledge graph to form a large-scale model that is identifiable, inferable, and usable.

[0044] Furthermore, the characteristics of the production data in the risk warning big data model include the following four points: Data source: including production equipment, production personnel, production environment, production materials, production process, all production links and the data generated by all production links, forming a complete record of production data, constituting the raw data for risk warning; Data format: production data is directly collected and analyzed through sensor data and system data, constituting the data source for risk warning; Data processing: the raw data is preprocessed, including data filtering, data transformation, and data cleaning, to form risk data processing rules; Data storage: the risk data is deeply analyzed through artificial intelligence systems, including but not limited to knowledge graph construction, data standardization processing, and data mining, to form valuable information from the risk data.

[0045] Furthermore, the risk management of the large-scale risk warning model includes the following steps: Establishing a risk management model: Analyzing, evaluating, and handling risk warning information through an artificial intelligence system to ultimately form a risk management plan; Handling warnings: Based on the processing and training of risk data, generating a warning plan library, and generating warning plans for production risks, including but not limited to production environment, equipment safety, personnel safety, and material safety, and disseminating the warnings; Handling risks: Developing different contingency plans based on different warning information, organizing relevant risk managers to handle the risks, evaluating the handling plans, and providing feedback on the results.

[0046] Specifically, the multimodal output of the risk management model for generating early warnings includes the following steps: Early Warning Content: Based on different risks, generate different early warning content outputs, including but not limited to text reports, image reports, audio reports, short video reports, and animation reports; Early Warning Classification: Classify early warnings according to risk classification, including but not limited to accident alarms, alerts, warnings, and incidents; Early Warning Transmission: Based on early warning transmission, transmit the corresponding early warnings to different levels and departments, including but not limited to sending them to workshop risk management personnel, equipment administrators, material management departments, and production management personnel; Based on the early warning transmission method, including the following methods: System message: The early warning information will be directly pushed to relevant personnel through the artificial intelligence system; SMS alerts: Warning information will be sent to relevant personnel via SMS through the SMS system; Safety broadcast: The warning information is broadcast through the broadcast system; Manual notification: The early warning information is delivered manually by personnel. Early warning plan: Based on the transmission of early warning information, corresponding early warning plans shall be formulated, including but not limited to handling measures, handling procedures, emergency plans, and risk management.

[0047] Furthermore, the characteristics of the data processed by the risk warning big data model include the following steps: Data source, including three parts: production environment, production equipment, and production materials, which are acquired in real time through the Internet of Things system; Data processing, using data processing technology to preprocess the production data, including but not limited to data integration, data cleaning, data conversion, data analysis, and data standardization, to form a risk warning data processing library; Data storage, after data processing, forming the storage of the risk warning data processing system, including but not limited to hazardous materials storage, material storage, equipment storage, production environment monitoring, and production parameter storage.

[0048] Furthermore, the multimodal construction of the risk management big data model for risk early warning includes the following steps: Using an artificial intelligence system, risk early warning data is graphed into a risk knowledge graph, risk factors are identified, extracted, and constructed, and combined with safety production standards to form a risk knowledge graph, including but not limited to knowledge graphs of hazardous materials, materials, equipment, environment, and production processes; Using an artificial intelligence system, knowledge extraction, knowledge fusion, knowledge reasoning, and knowledge application are performed on the risk knowledge graph to form predictable and reasonable risk early warning capabilities, including but not limited to establishing hazardous materials knowledge graph models, material risk models, equipment operating status models, and environmental risk models; A multimodal processing system for risk early warning is formed, based on the artificial intelligence system, performing feature construction, big data model reasoning, and early warning processing on the processed risk early warning data to form automated identification, early warning, and processing capabilities, including but not limited to hazardous materials identification, material identification, equipment status identification, environmental identification, and emergency response decision-making.

[0049] Furthermore, for chemical raw material management, relevant information for each batch of raw materials is entered through electronic chemical codes to establish a material management system, forming a dynamic database of material batches and production batches. By uploading and verifying the production and inventory status of each batch of materials in real time, the production speed of batch materials is controlled. For the monitoring and maintenance of production equipment, various production parameters are monitored in real time. During the production process, anomalies in production equipment are detected through intelligent risk warning models and manual inspections, and targeted maintenance plans are proposed to provide auxiliary decision-making for production equipment management. For the management of the production environment, the safety status of the production environment is monitored in real time, production environment indicator reports are generated, and continuous improvement is carried out to form an environmental management system, realizing real-time monitoring, management, and analysis of production environment data.

[0050] Specifically, the construction of the intelligent reasoning system for the risk management big data model includes the following steps: Risk Knowledge Graph Construction: Structured processing of knowledge, data, relationships, and patterns related to safe production; structured processing of accidents, safety hazards, early warning information, protective measures, production processes, equipment, and materials during production; Risk Reasoning Engine: Automatic reasoning through the engine processes the data in the risk knowledge graph, including knowledge extraction based on BERT entity recognition models, knowledge fusion based on entity linking technology, knowledge storage, knowledge reasoning based on graph neural networks (GNNs), and knowledge application; Risk Data Fusion: Using knowledge extraction models to extract data from different data sources and structures; using knowledge fusion models to fuse and organize knowledge from different sources and structures, including entity alignment, attribute alignment, and relationship alignment; Risk Data Reasoning: Using knowledge reasoning models to reason about knowledge in a logical or structured manner; using knowledge application models to output the reasoning results, including alarms, decisions, and safety knowledge push notifications.

[0051] Furthermore, the data distribution and early warning generation of the risk management big data model system includes the following steps: risk identification through an artificial intelligence system; risk assessment through an artificial intelligence system; early warning generation and release through risk managers; and risk elimination and rectification through risk managers.

[0052] Specifically, the risk handling of the large-scale risk management model system includes the following steps: Risk Identification: Establishing a production process model, employing big data processing technology, and comprehensively utilizing technologies such as artificial intelligence, the Internet of Things, and video surveillance to construct data on dynamic material identification, equipment operation data, and personnel on-duty status; analyzing real-time production data to generate process risk indicators; Risk Assessment: Comprehensively utilizing technologies such as artificial intelligence, the Internet of Things, and video surveillance to establish a production environment data model; analyzing real-time production environment data; assessing real-time production environment indicators; through real-time analysis of the above indicators, using the large-scale model to predict risks; and providing preliminary early warning for potential risks; for the verification or handling of risks, further investigation is required. The process involves: 1) One-step manual confirmation to confirm handling measures and provide feedback on the handling strategy; 2) Risk management: Risks are managed according to their level, with higher risk levels requiring more severe measures, and vice versa; 3) Risk management strategies: Different personnel are assigned, stratified, and categorized based on different risks; 4) Repair and handling of existing problems: Different repair and handling strategies are developed for different types of problems; 5) For persistent problems, these are included in risk analysis for continuous observation or monitoring; 6) For management issues, the handling results are promptly sent to relevant management departments and feedback is provided to the risk management department. For recurring and frequently occurring problems, relevant suggestions are made to strengthen management or monitoring.

[0053] An example of risk warning application in penicillin production workshops for penicillin raw material (hazardous material) storage tanks: 1. Risk Data System Construction: The risk knowledge graph includes rules such as "Penicillin storage temperature must be controlled between 2-8℃" and "Storage tank pressure must not exceed 0.3MPa"; three types of risks are defined: Temperature abnormality risk i=1; pressure abnormality risk i=2; personnel operation risk i=3.

[0054] The weights are set as follows: =0.4, =0.4, =0.2; Severity = =0.9, which may lead to raw material failure or leakage. =0.5, so improper operation has a relatively small impact.

[0055] 2. Real-time data acquisition: The IoT system acquires real-time data from the storage tank: temperature 9℃, pressure 0.32MPa, operator not wearing protective gloves as required; the large-scale model predicts the probability of occurrence of the three types of risks as follows: =0.8, temperature exceeds the standard; =0.7, pressure exceeds the standard; =0.6, operation violated.

[0056] 3. Risk Index Calculation: Based on the formula:

[0057] The calculation yields: RI = (0.4 × 0.9 × 0.8) + (0.4 × 0.9 × 0.7) + (0.2 × 0.5 × 0.6) = 0.288 + 0.252 + 0.06 = 0.6. Recalculated: RI = (4 × 0.9 × 0.8) + (4 × 0.9 × 0.7) + (2 × 0.5 × 0.6) = 2.88 + 2.52 + 0.6 = 6.0, which is considered medium risk (3 < 6.0 ≤ 7).

[0058] 4. Early Warning and Response: The system generates multimodal early warnings and text reports such as "Penicillin storage tank temperature and pressure exceed limits, operator protection is inadequate"; short videos: real-time monitoring screens mark abnormal points and push them to the material management department and workshop safety officer; after on-site verification by the safety officer, the cooling and depressurization procedure is initiated, and the operator is required to wear protective equipment properly. The handling results are fed back to the system to form a closed loop.

[0059] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for identifying and processing risks in a pharmaceutical workshop using a large-scale risk early warning model, characterized in that, This includes the following: constructing a risk data system and integrating the risk data with safety production standards for training; Acquire real-time production data and use it to detect production risk indices in real time. Based on historical production data and the company's management system, a large-scale risk management model is constructed to implement multimodal generation and processing strategies for data distribution and early warning.

2. The method for identifying and processing risks in a pharmaceutical workshop using a large-scale risk warning model according to claim 1, characterized in that, The construction of a risk data system and the integration of risk data with safety production standards for training include the following: The sources of risk include material risk, equipment risk, environmental risk, and operational risk. In the construction of the risk data system, the data includes safety production standards, historical production data, and corresponding data related to the risk content; Integrated training involves merging risk data with safety production standards to form a structured risk knowledge graph; Knowledge is extracted from the risk knowledge graph using a BERT-based entity recognition model, knowledge is fused using entity linking technology, and knowledge reasoning is performed using graph neural networks, ultimately enabling knowledge application.

3. The method for identifying and processing risks in a pharmaceutical workshop using a large-scale risk warning model according to claim 1, characterized in that, The acquisition of real-time production data and the real-time detection of production risk index through real-time production data include the following: the acquisition of real-time production data includes collection by the Internet of Things system and acquisition of special data, including chemical raw materials, production equipment and production environment; Based on real-time production data, real-time risk prediction is performed through a large model, and a real-time risk index is calculated by combining risk weights and severity, including chemical handling risks, equipment malfunction risks, and environmental exceedance risks. Multimodal early warning data is distributed according to risk type, including text reports, image reports, audio reports, short video reports, and animation reports.

4. The method for identifying and processing risks in a pharmaceutical workshop using a large-scale risk warning model according to claim 3, characterized in that, The acquisition of real-time production data includes collection through the Internet of Things system and acquisition of specialized data. For chemical raw material management, relevant information of each batch of raw materials is entered through electronic chemical codes to establish a material management system and form a dynamic material batch and production batch relationship database. By uploading and verifying the production and inventory status of each batch of materials in real time, the production speed of batch materials is controlled.

5. The method for identifying and processing risks in a pharmaceutical workshop using a large-scale risk warning model according to claim 3, characterized in that, The acquisition of real-time production data includes collection by the Internet of Things system and acquisition of specialized data. For the monitoring and maintenance of production equipment, various production parameters are monitored in real time. During the production process, production abnormalities are detected through intelligent risk warning models and manual inspections, and targeted maintenance solutions are proposed to provide auxiliary decision-making for production equipment management.

6. The method for identifying and processing risks in a pharmaceutical workshop using a large-scale risk warning model according to claim 3, characterized in that, The acquisition of real-time production data includes collection by the Internet of Things system and acquisition of specialized data. Among these, the monitoring and continuous improvement of the production environment involves real-time monitoring of the safety status of the production environment, generation of production environment indicator reports, continuous improvement, establishment of an environmental management system, recording production environment data in different regions, locations, and time periods, and realizing real-time monitoring, management, and analysis of production environment data.

7. The method for identifying and processing risks in a pharmaceutical workshop using a large-scale risk warning model according to claim 1, characterized in that, The construction of a large-scale risk management model, and the multimodal generation and processing strategies for data distribution and early warning, include the following: risk identification through the large-scale risk management model; risk assessment through the large-scale risk management model; generation and release of early warnings through risk managers; and risk elimination and rectification through risk management strategies.

8. The method for identifying and processing risks in a pharmaceutical workshop using a large-scale risk warning model according to claim 7, characterized in that, The risk identification achieved through the risk processing big model includes data processing and identification. Data processing includes real-time data filtering, transformation, and cleaning to form standardized data. Identification includes using big data identification algorithms to identify and analyze data and filter out abnormal data; and using Transformer-based deep neural network models combined with historical data and knowledge graphs for supervised learning. Feature engineering is used to extract risk features and train a risk inference model library.

9. The method for identifying and processing risks in a pharmaceutical workshop using a large-scale risk warning model according to claim 7, characterized in that, The risk assessment using a large-scale risk processing model includes calculating the real-time risk index (RI) and classifying risks according to the RI. The risk assessment process specifically includes: calculating the real-time risk index (RI) using the large-scale model and outputting it in real time; making a preliminary assessment by comparing it with a preset threshold; further reviewing and confirming it manually; and generating early warning information based on the final assessment results.

10. The method for identifying and processing large-scale risk early warning models in pharmaceutical workshops according to claim 7, characterized in that, The generation and release of early warnings through risk managers includes multimodal early warnings, including text, images, audio, short videos, and animations; it also includes graded processing of early warnings, implementing different levels of processing schemes according to the severity of the early warning. The emergency response is communicated through system messages, SMS, security broadcasts, and manual notifications, forming a closed loop.

Images