A risk assessment method and device based on standard operating procedure semantic parsing

Abstract

The application discloses a risk assessment method and device based on standard operation procedure semantic analysis, and relates to the technical field of laboratory safety management and artificial intelligence cross. The method comprises the following steps: extracting key information based on safety system documents and operation procedures, and constructing a risk rule library index; analyzing monitoring videos in parallel, generating behavior semantic description through visual recognition and time sequence analysis; using rule index to retrieve related rules, inputting semantic description into a large model to infer risks and evaluate credibility; early warning visualization and feedback labeling collection to realize continuous optimization. The application can convert the existing unstructured standard operation procedure documents of the laboratory into computable structured rules without relying on a large amount of labeled data, and realize intelligent risk assessment and early warning in combination with video monitoring data.

Description

Technical Field

[0001] This invention relates to the field of laboratory safety management and artificial intelligence, and in particular to a risk assessment method and apparatus based on semantic parsing of standard operating procedures. Background Technology

[0002] With the rapid development of higher education and scientific research in my country, the scale and number of laboratories in universities and research institutes have continued to expand. Currently, laboratories generally exhibit characteristics of broad disciplinary coverage, diverse equipment and instruments, and numerous types of hazards. Despite continuous updates to equipment and strengthened safety management, laboratory safety accidents still occur frequently due to the high mobility of laboratory personnel, the dispersed nature of experiments, and the heavy workload of management. As core locations for talent cultivation, scientific research, and technological innovation, laboratories face severe challenges in safety management. Traditional laboratory safety management mainly relies on manual patrols and periodic inspections, which have the following prominent problems: limited regulatory coverage, making it difficult to achieve 24 / 7, all-round monitoring; delayed discovery of hidden dangers, often only revealing problems after an accident occurs; low management efficiency, with a large amount of repetitive work consuming human resources; and a lack of data support, making it difficult to conduct scientific risk assessments and decision-making.

[0003] Standard Operating Procedures (SOPs) are a crucial foundation for laboratory safety management, encompassing key aspects such as experimental procedures, safety requirements, and emergency response measures. However, traditional SOPs exist in unstructured text form, making them difficult for computer systems to directly understand and apply. Transforming SOP text into computable and reasonable structured knowledge is a key technical challenge for achieving intelligent safety management. In recent years, Large Language Models (LLMs) have made breakthroughs in natural language understanding and knowledge reasoning, and their zero-shot learning capabilities provide a new technical path for the intelligent upgrading of laboratory safety management. Applying Large Language Models to laboratory safety risk assessment and building a more intelligent and efficient laboratory safety management system remains a significant technical challenge. Summary of the Invention

[0004] To address the technical problems of existing technologies, such as insufficient utilization of Standard Operating Procedure (SOP) knowledge, inadequate model interpretability, difficulty in achieving dynamic risk assessment, and lack of multi-source data fusion capabilities, this invention provides a risk assessment method and apparatus based on semantic parsing of Standard Operating Procedures. The technical solution is as follows:

[0005] On the one hand, a risk assessment method based on standard operating procedure (SOP) semantic parsing is provided. This method is implemented by a risk assessment device based on SOP semantic parsing, and includes: S1: Collect laboratory safety management system documents and standard operating procedure documents, and format them to obtain structured text data; S2: Input structured text data into a general large language model to obtain key information, obtain risk assessment rules based on the key information, store and organize the risk assessment rules to obtain a risk rule base, and obtain a rule index based on the risk assessment rules in the risk rule base. The key information includes operation steps, safety requirements, prohibited items and / or descriptions of hazards. S3: Obtain raw video data by collecting data from video surveillance cameras, preprocess the raw video data to obtain a video image frame sequence, obtain scene structured information through visual recognition, obtain a behavior event sequence through temporal behavior analysis, and input the semantic conversion and encoding module to obtain a behavior semantic description. The behavior event sequence includes experimental operation behavior, violation behavior and / or behavior pattern. S4: Based on the behavioral semantic description, extract relevant risk judgment rules from the risk rule base using the rule index, and input the behavioral semantic description and relevant risk judgment rules into the general large language model to infer the risk prediction result of the current operation and perform credibility assessment. S5: Determine the risk level of the risk prediction result of the current operation, generate and push early warning information when the risk threshold is exceeded, and track the response to obtain the early warning closed-loop processing status. S6: Visualize the warning information and collect feedback to obtain the labeling information of the warning results. Based on the labeling information, obtain standardized emergency response cards and precise education information, and extract modification opinions to obtain updated large language model prompt word templates and / or time series prediction models through manual in-loop fine-tuning.

[0006] Preferably, step S1 involves collecting laboratory safety management system documents and standard operating procedure documents, and then formatting them to obtain structured text data, including: S11: Collect laboratory safety management system documents to obtain the first unstructured text data. The laboratory safety management system documents include laboratory safety management regulations, hazardous chemical management methods and / or laboratory access system. S12: Collect standard operating procedure documents to obtain second unstructured text data. The standard operating procedure documents include experimental standard operating procedures, equipment operation manuals and / or emergency response procedures. S13: The first unstructured text data and the second unstructured text data are formatted to obtain structured text data. The formatting process supports parsing PDF, Word and / or TXT document formats.

[0007] Preferably, in step S2, structured text data is input into a general large language model to obtain key information, risk assessment rules are obtained based on the key information, the risk assessment rules are stored and organized to obtain a risk rule base, and a rule index is obtained based on the risk assessment rules in the risk rule base. The key information includes operating steps, safety requirements, prohibited items, and / or descriptions of hazards, including: S21: Input structured text data into a general large language model for semantic understanding and information recognition to obtain key information; S22: Based on the key information, after semantic extraction and structure transformation, risk determination rules are obtained, and the risk determination rules are executable safe operation specifications and / or risk logic; S23: Construct a rule base, store and organize the risk assessment rules to obtain a risk rule base, the risk rule base having the feature of unified management of the risk assessment rules; S24: Based on the risk assessment rules in the risk rule base, a rule index is obtained through correlation analysis and establishment. The rule index includes the mapping relationship between the risk assessment rules and the experimental scenario, equipment type and / or operation behavior.

[0008] Preferably, step S3 involves acquiring raw video data from a video surveillance camera, preprocessing the raw video data to obtain a sequence of video image frames, obtaining structured scene information through visual recognition, and then obtaining a sequence of behavioral events through temporal behavior analysis. This sequence is then input into a semantic conversion and encoding module to obtain a semantic description of the behavior. The sequence of behavioral events includes experimental operation behaviors, violations, and / or behavioral patterns, including: S31: Collect data from video surveillance cameras through the laboratory video surveillance system to obtain raw video data, which includes real-time video streams and / or historical recording data; S32: Perform video decoding, video frame extraction, and image enhancement on the original video data to obtain a video image frame sequence; S33: Based on the video image frame sequence, perform target detection, posture and motion detection, and protective equipment wearing status recognition to obtain scene structured information. The target detection includes target detection of experimental personnel, equipment, and items. The posture and motion detection includes detection of the human posture and motion of experimental personnel. The protective equipment wearing status recognition includes recognition of the protective equipment wearing status of experimental personnel. The protective equipment includes safety helmets, goggles, laboratory suits, and / or protective gloves. S34: Perform temporal behavior analysis on the structured information of the scenario to obtain a sequence of behavioral events, the sequence of behavioral events including experimental operation behavior, violation behavior and / or behavior pattern; S35: Input the sequence of behavioral events into the semantic conversion and encoding module to obtain a behavioral semantic description.

[0009] Preferably, step S4 involves extracting relevant risk assessment rules from a risk rule base based on the behavioral semantic description using a rule index, and inputting the behavioral semantic description and relevant risk assessment rules into a general large language model to infer the risk prediction result of the current operation and perform a credibility assessment, including: S41: Based on the behavioral semantic description, extract relevant risk judgment rules from the risk rule base using the rule index; S42: Using the prompt word engineering method, the relevant risk judgment rules and the behavioral semantic description are organized into a structured input format and input into a large language model. The structured input format is reasoned using the thought chain technology and zero-shot learning capability to obtain the risk prediction result of the current operation. S43: Evaluate the credibility of the reasoning results.

[0010] Preferably, step S5 involves determining the risk level of the risk prediction result of the current operation, generating and pushing a warning message when the risk threshold is exceeded, and tracking the response to obtain the warning closed-loop processing status, including: S51: Using a preset risk threshold, the risk level in the risk prediction result of the current operation is judged, and a warning trigger command is obtained when the risk level exceeds the preset risk threshold; S52: Generate early warning information according to the early warning triggering instruction, and push the early warning information to relevant personnel through the early warning push channel. The early warning information includes early warning captured images and / or the thought chain reasoning process output by the large language model. The early warning push channel includes mobile APP, SMS, on-site broadcast and / or warning lights. S53: The response status of the early warning information is tracked through the early warning response tracking module to obtain the early warning closed-loop processing status.

[0011] Preferably, step S6 involves visualizing the early warning information and collecting feedback to obtain labeled information of the early warning results, matching standardized emergency response cards and precise educational information based on the labeled information, and extracting modification opinions and manually fine-tuning them in the loop to obtain updated large language model prompt word templates and / or time series prediction models, including: S61: Visualize the warning information to obtain a warning interface that includes the thought chain reasoning process; S62: Collect feedback instructions from safety management personnel on the warning results, and obtain the labeling information of the warning results. The labeling information of the warning results labels the warning results as valid warnings or false alarms, corresponding risk types, modification suggestions, handling results and / or manual feedback. The warning result labeling is used to iteratively optimize the prompt word template and / or risk rule base of the large language model to form an adaptive safety management closed loop. S63: Based on the annotation information of the warning result, after matching with the emergency response plan, a standardized emergency response card is obtained and pushed to relevant personnel. The standardized emergency response card includes standardized emergency response steps for the corresponding risk type. S64: Based on the annotation information of the warning result, obtain the associated educational materials and obtain the precise educational information to be sent to the terminal of the violator. The educational materials include standard operating procedure documents and / or safety micro-lesson videos. S65: Extract modification suggestions from the annotation information of the warning results, and after manual in-loop fine-tuning and optimization, obtain the updated large language model prompt word template and / or time series prediction model. The fine-tuning and optimization process includes: adjusting the prompt word template based on the modification suggestions, and optimizing the time series prediction model by using the false alarm data as negative samples.

[0012] On the other hand, a risk assessment device based on standard operating procedure semantic parsing is provided. This device is applied to a risk assessment method based on standard operating procedure semantic parsing. The device includes: Structured text module: Used to collect laboratory safety management system documents and standard operating procedure documents, and to format them to obtain structured text data; Risk assessment rule module: This module is used to input structured text data into a general large language model to obtain key information, obtain risk assessment rules based on the key information, store and organize the risk assessment rules to obtain a risk rule library, and obtain a rule index based on the risk assessment rules in the risk rule library. The key information includes operation steps, safety requirements, prohibited items and / or descriptions of hazards. Video data module: used to obtain raw video data by collecting data from video surveillance cameras, preprocess the raw video data to obtain video image frame sequences, obtain scene structured information through visual recognition, obtain behavioral event sequences through temporal behavior analysis, and input them into the semantic conversion and encoding module to obtain behavioral semantic descriptions. The behavioral event sequences include experimental operation behaviors, violation behaviors and / or behavioral patterns. Behavioral semantic module: Based on the behavioral semantic description, it extracts relevant risk judgment rules from the risk rule base using rule index, inputs the behavioral semantic description and relevant risk judgment rules into a general large language model, infers the risk prediction result of the current operation, and performs credibility assessment. Risk prediction module: used to determine the risk level of the risk prediction result of the current operation, generate and push early warning information when the risk threshold is exceeded, and track the response to obtain the early warning closed-loop processing status; The annotation information module is used to visualize the early warning information and collect the annotation information of the early warning results. Based on the annotation information, it matches the standardized emergency response card and precise education information, and extracts modification opinions and performs manual on-loop fine-tuning to obtain the updated large language model prompt word template and / or time series prediction model.

[0013] On the other hand, a risk assessment device based on standard operating procedure semantic parsing is provided. The risk assessment device based on standard operating procedure semantic parsing includes: a processor; a memory, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the method described in any one of the above-described risk assessment methods based on standard operating procedure semantic parsing is implemented.

[0014] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: Employing zero-shot learning capabilities of a large language model, risk reasoning can be performed without a large amount of labeled data, reducing system deployment costs and improving adaptability to new scenarios and risks. Existing laboratory SOP documents and safety management systems are transformed into a structured rule base through semantic parsing, enabling effective utilization of existing knowledge and rule-based compliance judgments, thus improving the professionalism and accuracy of risk assessment. Through temporal behavior prediction and SOP rule-driven risk prediction, early warnings are issued before risks actually occur, achieving a proactive safety management model. The integration of document data (SOPs, management systems) and video data (surveillance cameras) enables joint analysis of textual semantics and visual behavior, allowing for a more comprehensive risk assessment. The use of thought chain technology outputs a complete reasoning process, making risk assessment results traceable and auditable, facilitating understanding and adoption by safety management personnel. Through a human-machine collaborative management module, machine intelligence is combined with human judgment, collecting feedback to continuously optimize the system and avoiding the early warning neglect effect caused by false alarms in purely automated systems. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart of a risk assessment method based on semantic parsing of standard operating procedures provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of a system for semantic parsing of standard operating procedures provided in an embodiment of the present invention; Figure 3 This is a block diagram of a risk assessment device based on semantic parsing of standard operating procedures provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of a risk assessment device based on semantic parsing of standard operating procedures provided in an embodiment of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0018] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0019] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0020] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0021] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0022] This invention provides a risk assessment method based on standard operating procedure (SOP) semantic parsing. This method can be implemented by a risk assessment device based on SOP semantic parsing, which can be a terminal or a server. Figure 1 The flowchart shown is for a risk assessment method based on standard operating procedure semantic parsing. The processing flow of this method may include the following steps:

[0023] Preferably, laboratory safety management system documents and standard operating procedure documents are collected and formatted to obtain structured text data, including: Collect laboratory safety management system documents to obtain the first unstructured text data. The laboratory safety management system documents include laboratory safety management regulations, hazardous chemical management methods and / or laboratory access system. Collect standard operating procedure documents to obtain second unstructured text data. The standard operating procedure documents include experimental standard operating procedures, equipment operation manuals and / or emergency response procedures. The first unstructured text data and the second unstructured text data are formatted to obtain structured text data. The formatting process supports parsing PDF, Word and / or TXT document formats.

[0024] In some embodiments, laboratory safety management system documents (including laboratory safety management regulations, hazardous chemical management methods, laboratory access regulations, etc.) and experimental operation SOP documents (including standard operating procedures for various experiments, equipment operation manuals, emergency response procedures, etc., such as standard operating procedures for fatigue testing machines) in PDF, Word, and TXT formats are collected through file import interfaces or system integration methods; the SOP documents are then converted, segmented, and cleaned; and a large language model is used to understand the semantics of the documents and identify operating steps, safety requirements, prohibited items, and descriptions of hazards. The system accesses laboratory video surveillance cameras via RTSP / ONVIF streaming media protocols to acquire real-time video streams and historical recording data; it also performs video data decoding, frame extraction, and image enhancement.

[0025] Preferably, structured text data is input into a general large language model to obtain key information. Risk assessment rules are then derived based on this key information. These risk assessment rules are stored and organized to form a risk rule base. A rule index is obtained based on the risk assessment rules in the risk rule base. The key information includes operational steps, safety requirements, prohibited items, and / or descriptions of hazards, including: Structured text data is input into a general large language model for semantic understanding and information recognition to obtain key information; Based on the key information, risk assessment rules are obtained through semantic extraction and structure transformation. These risk assessment rules are executable safe operation specifications and / or risk logic. A rule base is constructed to store and organize the risk assessment rules, resulting in a risk rule base. The risk rule base has the characteristic of unified management of the risk assessment rules. Based on the risk assessment rules in the risk rule base, a rule index is obtained through correlation analysis and establishment. The rule index includes the mapping relationship between the risk assessment rules and the experimental scenario, equipment type and / or operation behavior.

[0026] In some embodiments, the SOP document is first parsed by the document preprocessing unit to remove interfering information such as headers, footers, and charts, and the document is segmented into sections or operation steps. Subsequently, the semantic understanding unit inputs the segmented text into a pre-deployed large language model (such as a large language model with hundreds of billions of parameters based on the Transformer architecture). By designing specific system prompts, the large language model is guided to recognize "operation steps," "safety requirements" (such as mandatory wearing of safety goggles), "prohibited items" (such as prohibiting putting one's head into a fume hood), and "hazard descriptions" (such as flammable and explosive gases) in the text.

[0027] Preferably, raw video data is obtained by collecting data from video surveillance cameras. The raw video data is preprocessed to obtain a sequence of video image frames, and then visual recognition is used to obtain structured scene information. Temporal behavior analysis is then performed to obtain a sequence of behavioral events, which is input into a semantic conversion and encoding module to obtain a semantic description of the behavior. The sequence of behavioral events includes experimental operation behaviors, violations, and / or behavioral patterns, including: The laboratory video surveillance system collects data from video surveillance cameras to obtain raw video data, which includes real-time video streams and / or historical recording data. The original video data is subjected to video decoding, video frame extraction, and image enhancement to obtain a video image frame sequence; Based on the video image frame sequence, target detection, posture and motion detection, and protective equipment wearing status recognition are performed to obtain scene structured information. The target detection includes target detection of experimental personnel, equipment, and items. The posture and motion detection includes detection of the human posture and motion of experimental personnel. The protective equipment wearing status recognition includes recognition of the protective equipment wearing status of experimental personnel. The protective equipment includes safety helmets, goggles, laboratory suits, and / or protective gloves. Temporal behavior analysis is performed on the structured information of the scenario to obtain a sequence of behavioral events, which includes experimental operation behaviors, violation behaviors, and / or behavioral patterns. The sequence of behavioral events is input into the semantic transformation and encoding module to obtain a behavioral semantic description.

[0028] In some embodiments, the video behavior recognition algorithm is deployed on edge computing nodes to ensure the real-time performance and privacy security of video processing. Specifically, the input video stream is processed by frame extraction at a preset frame rate (e.g., 15fps). YOLO and other target detection algorithms are used to locate targets such as laboratory personnel, experimental equipment (e.g., beakers, alcohol lamps), and hazardous chemical reagents in the scene, and output bounding box coordinates. HRNet or OpenPose algorithms are used to extract the skeletal keypoint sequences of the laboratory personnel. The protective equipment recognition unit combines target detection and image classification algorithms to extract features from areas such as the head, hands, and torso of the personnel, determining the wearing status of safety helmets, goggles, lab coats, and protective gloves (worn / not worn / not worn correctly). Based on a spatiotemporal graph convolutional network (ST-GCN) or a video Transformer model, the skeletal keypoint sequences and target interaction information are input into the model, outputting specific action labels (e.g., "pouring liquid," "lighting alcohol lamp," "reaching for reagents").

[0029] Preferably, based on the behavioral semantic description, relevant risk judgment rules are extracted from the risk rule base using a rule index, and the behavioral semantic description and relevant risk judgment rules are input into a general large language model to infer the risk prediction result of the current operation and perform a credibility assessment, including: Based on the behavioral semantic description, relevant risk assessment rules are extracted from the risk rule base using the rule index; Using the prompt word engineering method, the relevant risk judgment rules and the behavioral semantic description are organized into a structured input format and input into a large language model. The structured input format is then reasoned using the thought chain technique and zero-shot learning capability to obtain the risk prediction result of the current operation. The credibility of the reasoning results is evaluated.

[0030] In some embodiments, such as Figure 2As shown, the structured action labels and equipment wearing status output by the video behavior recognition module are automatically converted into natural language descriptions based on the environmental context. For example, it generates: "The current experimenter is in a chemical laboratory, pouring an unknown transparent liquid, but is not wearing goggles or protective gloves." The rule matching unit retrieves Top-K relevant SOP rules from the rule index unit based on the current scene label (chemical laboratory) and action label (pouring liquid). Using the prompt word engineering method, the retrieved SOP rules and real-time natural language descriptions are organized into a structured input format and input into the large language model. An example of a prompt word template is as follows: "[Role Setting] You are a professional laboratory safety review expert. [SOP Rule] {Retrieved rule JSON}. [Current Observation Situation] {Generated Natural Language Description}. [Task Requirements] Please determine whether the current observation situation violates the SOP rule, and use a thought chain approach to analyze the reasoning process step by step, finally outputting the risk level (Level 1 Low Risk / Level 2 Medium Risk / Level 3 High Risk / Level 4 Extremely High Risk) and confidence score." The large language model, based on its powerful zero-shot learning and logical reasoning capabilities, directly outputs the reasoning results. Thanks to the use of thought chain technology, the system outputs a complete reasoning process, such as "According to rule R001, safety goggles and gloves must be worn when pouring concentrated sulfuric acid liquids. Currently, liquid has been poured without the aforementioned equipment, therefore it is judged as a violation, belonging to level three high risk," which greatly enhances the interpretability and auditability of the system.

[0031] Preferably, the risk prediction result of the current operation is used to determine the risk level, and when the risk threshold is exceeded, an early warning message is generated and pushed, and the response is tracked to obtain the early warning closed-loop processing status, including: Using a preset risk threshold, the risk level in the risk prediction result of the current operation is judged, and a warning trigger command is obtained when the risk level exceeds the preset risk threshold; According to the warning triggering instruction, a warning message is generated and pushed to relevant personnel through the warning push channel. The warning message includes a warning captured image and / or the thought chain reasoning process output by the large language model. The warning push channel includes a mobile APP, SMS, on-site broadcast and / or warning light. The early warning response tracking module tracks the response status of the early warning information to obtain the early warning closed-loop processing status.

[0032] In some embodiments, the risk prediction unit incorporates a time-series prediction model based on the Transformer architecture. This model receives behavioral feature sequences from the past T time windows (e.g., the past 10 seconds) as input and predicts the probability distribution of behavior in the next t time windows (e.g., the next 3 seconds). For example, if the time-series model predicts that a person has a very high probability of "lighting an alcohol lamp" in the next 3 seconds, and the protective equipment identification unit indicates that the person's "long hair is not tied up tightly," the zero-sample risk inference module immediately pre-matches this "predicted behavior" with the "ignition SOP rule," identifying a "Level 4 extremely high risk (fire hazard)." Once the risk level exceeds a preset threshold (e.g., reaching Level 2 or above), the warning generation unit immediately generates a warning message containing the risk type, location, violation details, and handling suggestions. The warning push unit selects different push links based on the risk level: low-risk warnings are pushed to the laboratory manager's mobile app, while high-risk / extremely high-risk warnings simultaneously trigger on-site loudspeaker announcements, flashing red warning lights, and a mandatory SMS message to the safety inspector, ensuring the danger chain is broken before the "ignition" action occurs.

[0033] Preferably, the warning information is visualized and the feedback is collected to obtain the annotation information of the warning results. Based on the annotation information, standardized emergency response cards and precise educational information are matched, and updated large language model prompt word templates and / or time-series prediction models are obtained through manual in-loop fine-tuning of extracted modification opinions. This includes: The warning information is visualized to obtain a warning interface that includes the thought chain reasoning process; The system collects feedback instructions from safety management personnel on the warning results and obtains the labeling information of the warning results. The labeling information of the warning results marks the warning results as valid warnings or false alarms, corresponding risk types, modification suggestions, handling results and / or manual feedback. The warning result labeling is used to iteratively optimize the prompt word template and / or risk rule base of the large language model to form an adaptive safety management closed loop. Based on the annotation information of the warning results, after matching with the emergency response plan, a standardized emergency response card is generated and pushed to relevant personnel. The standardized emergency response card includes standardized emergency response steps for the corresponding risk type. Based on the annotation information of the warning results, relevant educational materials are obtained, and precise educational information is sent to the terminals of the violators. The educational materials include standard operating procedure documents and / or safety micro-lesson videos. Modification suggestions are extracted from the annotation information of the warning results. After manual in-loop fine-tuning and optimization, an updated large language model prompt word template and / or time series prediction model are obtained. The fine-tuning and optimization process includes: adjusting the prompt word template based on the modification suggestions, and optimizing the time series prediction model by using the false alarm data as negative samples.

[0034] In some embodiments, the early warning confirmation unit displays the captured early warning image and the "thinking chain reasoning process" output by the large language model to safety management personnel on the APP. Management personnel can confirm "valid early warning" or "false alarm" with one click. If it is a valid early warning, the handling guidance unit automatically pushes the standard emergency response card (such as "chemical spill emergency response steps") bound to the risk type. After the incident is resolved, the safety education unit pushes relevant SOP documents and safety micro-lesson videos to the personnel who violated the regulations, achieving precise education. The feedback optimization unit collects the "false alarm" data and modification suggestions marked by the management personnel. On the one hand, it fine-tunes the prompt word template of the large language model through manual in-loop processing. On the other hand, it uses it as negative samples to optimize the time series prediction model, so that the risk reasoning ability of the system can continuously evolve in actual use.

[0035] The above is an introduction to the method embodiments. The following describes the solution described in this application through device embodiments.

[0036] Figure 3 This is a block diagram illustrating a risk assessment apparatus based on standard operating procedure (SOP) semantic parsing, according to an exemplary embodiment. The apparatus is used for a risk assessment method based on SOP semantic parsing. (Refer to...) Figure 3 The device includes a structured text module 310, a risk assessment rule module 320, a video data module 330, a behavioral semantic module 340, a risk prediction module 350, and a labeling information module 360.

[0037] Structured text module 310: Used to collect laboratory safety management system documents and standard operating procedure documents, and to format them to obtain structured text data; Risk assessment rule module 320: is used to input structured text data into a general large language model to obtain key information, obtain risk assessment rules based on the key information, store and organize the risk assessment rules to obtain a risk rule library, and obtain a rule index based on the risk assessment rules in the risk rule library. The key information includes operation steps, safety requirements, prohibited items and / or descriptions of hazards. Video data module 330: used to obtain raw video data by collecting data from video surveillance cameras, preprocess the raw video data to obtain a video image frame sequence, obtain scene structured information through visual recognition, obtain a behavior event sequence through temporal behavior analysis, and input the semantic conversion and encoding module to obtain a behavior semantic description. The behavior event sequence includes experimental operation behavior, violation behavior and / or behavior pattern. Behavioral semantic module 340: Based on the behavioral semantic description, it extracts relevant risk judgment rules from the risk rule base using the rule index, and inputs the behavioral semantic description and relevant risk judgment rules into a general large language model to infer the risk prediction result of the current operation and perform a credibility assessment. Risk prediction module 350: used to determine the risk level of the risk prediction result of the current operation, generate and push early warning information when the risk threshold is exceeded, and track the response to obtain the early warning closed-loop processing status; The annotation information module 360 ​​is used to visualize the early warning information and collect the annotation information of the early warning results. Based on the annotation information, it matches the standardized emergency response card and precise education information, and extracts modification opinions and performs manual in-loop fine-tuning to obtain the updated large language model prompt word template and / or time series prediction model.

[0038] A risk assessment device based on standard operating procedure semantic parsing, the risk assessment device based on standard operating procedure semantic parsing includes: a processor; a memory, the memory storing computer-readable instructions, which, when executed by the processor, implement the method described in any one of the above-described risk assessment methods based on standard operating procedure semantic parsing.

[0039] Figure 4 This is a schematic diagram of the structure of a risk assessment device based on semantic parsing of standard operating procedures provided in an embodiment of the present invention, such as... Figure 4 As shown, the risk assessment device based on standard operating procedure semantic parsing may include the above-mentioned Figure 3 The risk assessment device shown is based on standard operating procedure (SOP) semantic parsing. Optionally, the risk assessment device 410 based on SOP semantic parsing may include a first processor 2001.

[0040] Optionally, the risk assessment device 410 based on standard operating procedure semantic parsing may also include a memory 2002 and a transceiver 2003.

[0041] The first processor 2001, memory 2002, and transceiver 2003 can be connected via a communication bus.

[0042] The following is combined Figure 4 A detailed description of each component of the risk assessment device 410 based on standard operating procedure semantic parsing is provided below: The first processor 2001 is the control center of the risk assessment device 410 based on standard operating procedure semantic parsing. It can be a single processor or a collective term for multiple processing elements. For example, the first processor 2001 can be one or more central processing units (CPUs), application-specific integrated circuits (ASICs), or one or more integrated circuits configured to implement embodiments of the present invention, such as one or more digital signal processors (DSPs), or one or more field-programmable gate arrays (FPGAs).

[0043] Optionally, the first processor 2001 can perform various functions of the risk assessment device 410 based on standard operating procedure semantic parsing by running or executing software programs stored in the memory 2002 and calling data stored in the memory 2002.

[0044] In a specific implementation, as one example, the first processor 2001 may include one or more CPUs, for example... Figure 4 CPU0 and CPU1 are shown in the diagram.

[0045] In a specific implementation, as one example, the risk assessment device 410 based on standard operating procedure semantic parsing may also include multiple processors, for example... Figure 4 The first processor 2001 and the second processor 2004 are shown in the diagram. Each of these processors can be a single-core processor or a multi-core processor. Here, a processor can refer to one or more devices, circuits, and / or processing cores used to process data (such as computer program instructions).

[0046] The memory 2002 is used to store the software program that executes the present invention, and is controlled by the first processor 2001 to execute it. The specific implementation method can be referred to the above method embodiment, and will not be repeated here.

[0047] Optionally, the memory 2002 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. The memory 2002 may be integrated with the first processor 2001 or may exist independently, and may be accessed through the interface circuit of the risk assessment device 410 based on standard operating procedure semantic parsing. Figure 4 (Not shown in the image) is coupled to the first processor 2001, and this embodiment of the invention does not specifically limit this.

[0048] The transceiver 2003 is used to communicate with network devices or with terminal devices.

[0049] Alternatively, transceiver 2003 may include a receiver and a transmitter. Figure 4 (Not shown separately). The receiver is used to implement the receiving function, and the transmitter is used to implement the transmitting function.

[0050] Optionally, the transceiver 2003 can be integrated with the first processor 2001 or exist independently, and can be connected to the interface circuit of the risk assessment device 410 based on standard operating procedure semantic parsing. Figure 4 (Not shown in the image) is coupled to the first processor 2001, and this embodiment of the invention does not specifically limit this.

[0051] It should be noted that, Figure 4 The structure of the risk assessment device 410 based on standard operating procedure semantic parsing shown in the diagram does not constitute a limitation on the router. Actual knowledge structure identification devices may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0052] Furthermore, the technical effects of the risk assessment device 410 based on standard operating procedure semantic parsing can be referred to the technical effects of the risk assessment method based on standard operating procedure semantic parsing described in the above method embodiments, and will not be repeated here.

[0053] It should be understood that the first processor 2001 in the embodiments of the present invention may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0054] It should also be understood that the memory in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0055] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0056] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0057] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0058] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0059] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0060] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0061] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0062] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0063] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0064] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0065] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A risk assessment method based on standard operating procedure semantic parsing, characterized in that, The method includes: S1: Collect laboratory safety management system documents and standard operating procedure documents, and format them to obtain structured text data; S2: Input structured text data into a general large language model to obtain key information, obtain risk assessment rules based on the key information, store and organize the risk assessment rules to obtain a risk rule base, and obtain a rule index based on the risk assessment rules in the risk rule base. The key information includes operation steps, safety requirements, prohibited items and / or descriptions of hazards. S3: Obtain raw video data by collecting data from video surveillance cameras, preprocess the raw video data to obtain a video image frame sequence, obtain scene structured information through visual recognition, obtain a behavior event sequence through temporal behavior analysis, and input the semantic conversion and encoding module to obtain a behavior semantic description. The behavior event sequence includes experimental operation behavior, violation behavior and / or behavior pattern. S4: Based on the behavioral semantic description, extract relevant risk judgment rules from the risk rule base using the rule index, and input the behavioral semantic description and relevant risk judgment rules into the general large language model to infer the risk prediction result of the current operation and perform credibility assessment. S5: Determine the risk level of the risk prediction result of the current operation, generate and push early warning information when the risk threshold is exceeded, and track the response to obtain the early warning closed-loop processing status. S6: Visualize the warning information and collect feedback to obtain the labeling information of the warning results. Based on the labeling information, obtain standardized emergency response cards and precise education information, and extract modification opinions to obtain updated large language model prompt word templates and / or time series prediction models through manual in-loop fine-tuning.

2. The risk assessment method based on standard operating procedure semantic parsing of claim 1, wherein, The S1 process collects laboratory safety management system documents and standard operating procedure documents, formats them, and obtains structured text data, including: S11: Collect laboratory safety management system documents to obtain the first unstructured text data. The laboratory safety management system documents include laboratory safety management regulations, hazardous chemical management methods and / or laboratory access system. S12: Collect standard operating procedure documents to obtain second unstructured text data. The standard operating procedure documents include experimental standard operating procedures, equipment operation manuals and / or emergency response procedures. S13: The first unstructured text data and the second unstructured text data are formatted to obtain structured text data. The formatting process supports parsing PDF, Word and / or TXT document formats.

3. The risk assessment method based on standard operating procedure semantic parsing of claim 1, wherein, S2 inputs structured text data into a general large language model to obtain key information, derives risk assessment rules based on the key information, stores and organizes these risk assessment rules to obtain a risk rule base, and obtains a rule index based on the risk assessment rules in the risk rule base. The key information includes operating steps, safety requirements, prohibited items, and / or hazard descriptions, including: S21: Input structured text data into a general large language model for semantic understanding and information recognition to obtain key information; S22: Based on the key information, after semantic extraction and structure transformation, risk determination rules are obtained, and the risk determination rules are executable safe operation specifications and / or risk logic; S23: Construct a rule base, store and organize the risk assessment rules to obtain a risk rule base, the risk rule base having the feature of unified management of the risk assessment rules; S24: Based on the risk assessment rules in the risk rule base, a rule index is obtained through correlation analysis and establishment. The rule index includes the mapping relationship between the risk assessment rules and the experimental scenario, equipment type and / or operation behavior.

4. The risk assessment method based on standard operating procedure semantic parsing of claim 1, wherein, The S3 process involves acquiring raw video data from a video surveillance camera, preprocessing the raw video data to obtain a sequence of video image frames, obtaining structured scene information through visual recognition, and then obtaining a sequence of behavioral events through temporal behavior analysis. This sequence is then input into a semantic conversion and encoding module to obtain a semantic description of the behavior. The sequence of behavioral events includes experimental operation behaviors, violation behaviors, and / or behavioral patterns, including: S31: Collect data from video surveillance cameras through the laboratory video surveillance system to obtain raw video data, which includes real-time video streams and / or historical recording data; S32: Perform video decoding, video frame extraction, and image enhancement on the original video data to obtain a video image frame sequence; S33: Based on the video image frame sequence, perform target detection, posture and motion detection, and protective equipment wearing status recognition to obtain scene structured information. The target detection includes target detection of experimental personnel, equipment, and items. The posture and motion detection includes detection of the human posture and motion of experimental personnel. The protective equipment wearing status recognition includes recognition of the protective equipment wearing status of experimental personnel. The protective equipment includes safety helmets, goggles, laboratory suits, and / or protective gloves. S34: Perform temporal behavior analysis on the structured information of the scenario to obtain a sequence of behavioral events, the sequence of behavioral events including experimental operation behavior, violation behavior and / or behavior pattern; S35: Input the sequence of behavioral events into the semantic conversion and encoding module to obtain a behavioral semantic description.

5. The risk assessment method based on standard operating procedure semantic parsing according to claim 1, characterized in that, S4, based on the behavioral semantic description, extracts relevant risk judgment rules from the risk rule base using a rule index, and inputs the behavioral semantic description and relevant risk judgment rules into a general large language model to infer the risk prediction result of the current operation and perform a credibility assessment, including: S41: Based on the behavioral semantic description, extract relevant risk judgment rules from the risk rule base using the rule index; S42: Using the prompt word engineering method, the relevant risk judgment rules and the behavioral semantic description are organized into a structured input format and input into a large language model. The structured input format is reasoned using the thought chain technology and zero-shot learning capability to obtain the risk prediction result of the current operation. S43: Evaluate the credibility of the reasoning results.

6. The risk assessment method based on standard operating procedure semantic parsing according to claim 1, characterized in that, S5 performs a risk level assessment on the risk prediction result of the current operation, generates and pushes a warning message when the risk threshold is exceeded, and tracks the response to obtain the warning closed-loop processing status, including: S51: Using a preset risk threshold, the risk level in the risk prediction result of the current operation is judged, and a warning trigger command is obtained when the risk level exceeds the preset risk threshold; S52: Generate early warning information according to the early warning triggering instruction, and push the early warning information to relevant personnel through the early warning push channel. The early warning information includes early warning captured images and / or the thought chain reasoning process output by the large language model. The early warning push channel includes mobile APP, SMS, on-site broadcast and / or warning lights. S53: The response status of the early warning information is tracked through the early warning response tracking module to obtain the early warning closed-loop processing status.

7. The risk assessment method based on standard operating procedure semantic parsing of claim 1, wherein, The step S6 involves visualizing the early warning information and collecting feedback to obtain labeled information of the early warning results. Based on the labeled information, standardized emergency response cards and precise educational information are matched, and updated large language model prompt word templates and / or time series prediction models are obtained by extracting modification opinions and performing manual on-loop fine-tuning. S61: Visualize the warning information to obtain a warning interface that includes the thought chain reasoning process; S62: Collect feedback instructions from safety management personnel on the warning results, and obtain the labeling information of the warning results. The labeling information of the warning results labels the warning results as valid warnings or false alarms, corresponding risk types, modification suggestions, handling results and / or manual feedback. The warning result labeling is used to iteratively optimize the prompt word template and / or risk rule base of the large language model to form an adaptive safety management closed loop. S63: Based on the annotation information of the warning result, after matching with the emergency response plan, a standardized emergency response card is obtained and pushed to relevant personnel. The standardized emergency response card includes standardized emergency response steps for the corresponding risk type. S64: Based on the annotation information of the warning result, obtain the associated educational materials and obtain the precise educational information to be sent to the terminal of the violator. The educational materials include standard operating procedure documents and / or safety micro-lesson videos. S65: Extract modification suggestions from the annotation information of the warning results, and after manual in-loop fine-tuning and optimization, obtain the updated large language model prompt word template and / or time series prediction model. The fine-tuning and optimization process includes: adjusting the prompt word template based on the modification suggestions, and optimizing the time series prediction model by using the false alarm data as negative samples.

8. A risk assessment device based on standard operating procedure semantic parsing, the risk assessment device based on standard operating procedure semantic parsing being configured to implement the risk assessment method based on standard operating procedure semantic parsing according to any one of claims 1 to 7, characterized in that, The device includes: Structured text module: Used to collect laboratory safety management system documents and standard operating procedure documents, and to format them to obtain structured text data; Risk assessment rule module: This module is used to input structured text data into a general large language model to obtain key information, obtain risk assessment rules based on the key information, store and organize the risk assessment rules to obtain a risk rule library, and obtain a rule index based on the risk assessment rules in the risk rule library. The key information includes operation steps, safety requirements, prohibited items and / or descriptions of hazards. Video data module: used to obtain raw video data by collecting data from video surveillance cameras, preprocess the raw video data to obtain video image frame sequences, obtain scene structured information through visual recognition, obtain behavioral event sequences through temporal behavior analysis, and input them into the semantic conversion and encoding module to obtain behavioral semantic descriptions. The behavioral event sequences include experimental operation behaviors, violation behaviors and / or behavioral patterns. Behavioral semantic module: Based on the behavioral semantic description, it extracts relevant risk judgment rules from the risk rule base using rule index, inputs the behavioral semantic description and relevant risk judgment rules into a general large language model, infers the risk prediction result of the current operation, and performs credibility assessment. Risk prediction module: used to determine the risk level of the risk prediction result of the current operation, generate and push early warning information when the risk threshold is exceeded, and track the response to obtain the early warning closed-loop processing status; The annotation information module is used to visualize the early warning information and collect the annotation information of the early warning results. Based on the annotation information, it matches the standardized emergency response card and precise education information, and extracts modification opinions and performs manual on-loop fine-tuning to obtain the updated large language model prompt word template and / or time series prediction model.

9. A risk assessment device based on standard operating procedure semantic parsing, characterized by, The risk assessment processor based on standard operating procedure semantic parsing; a memory storing computer-readable instructions, which, when executed by the processor, implement the method as described in any one of claims 1 to 7.

10. A computer readable storage medium, characterized in that, The computer-readable storage medium contains program code that can be invoked by a processor to execute the method as described in any one of claims 1 to 7.

Images