Power production system operation compliance monitoring and early warning method

Abstract

The invention provides an electric power production system operation compliance monitoring and early warning method. The method comprises the following steps: constructing a system library according to a system document of an electric power production system; generating a standard log document according to the operation record document and the notification record document of the power production system; the standard log document is input into the large model for key content extraction, and a structured log text is obtained; performing system searching and matching in a system library according to the structured log text to obtain a target system text; the structured log text, the target system text and contextual information are input into a large model for compliance judgment, a compliance analysis text is obtained, and the contextual information comprises historical record information of related equipment, materials and personnel related to the structured log text; and if the compliance analysis text characterizes that the operation is not compliant, triggering an early warning generation process. According to the method and the device, the limitation of the traditional method in the aspects of efficiency, accuracy, real-time performance and adaptability is effectively solved.

Description

Technical Field

[0001] The present application relates to the field of power production technology, and in particular to a method for monitoring and early warning the operational compliance of a power production system. Background Art

[0002] Power generation systems are subject to stringent requirements such as safe production and environmental protection. Any non-compliant operation can lead to serious consequences. Therefore, operational compliance monitoring and early warning are crucial for ensuring the safe and stable operation of power systems. Traditional compliance monitoring methods primarily rely on manual review and rule-based automated systems. In addition to these two traditional approaches, there are also system early warning systems based on machine learning.

[0003] Manual review relies on the experience of professionals. While flexible, it is highly subjective, inefficient, and costly. It also struggles to fully cover the complex power production process and leaves blind spots. Secondly, while rule-based automation systems offer efficiency and stability, their fixed rules make them difficult to adapt to the ever-changing dynamics of power production. They also lack adaptability, and the high cost of rule maintenance makes them inadequate for business development and regulatory updates. Furthermore, while machine learning-based system alerts can automatically discover patterns in data and dynamically optimize, they place extremely high demands on data quality. Data bias, missing data, or noise can lead to model misjudgments. Furthermore, their "black box" nature makes the models difficult to interpret, making it difficult for staff to understand and trust the alert results, impacting their practical application. In summary, existing technologies struggle to strike a balance between efficiency, adaptability, interpretability, and cost control. A more intelligent and flexible solution is urgently needed. Summary of the Invention

[0004] This application provides a method for monitoring and early warning of operational compliance of an electric power production system, which can solve the technical problem existing in the prior art of finding a balance between efficiency, adaptability, explainability and cost control.

[0005] In a first aspect, an embodiment of the present application provides a method for monitoring and providing early warning of operational compliance of a power production system, the method comprising: Build a system library based on the system documents of the power production system; Generate standard log files based on the operation log files and notification log files of the power production system; Input the standard log document into the big model to extract key content and obtain the structured log text output by the big model; According to the structured log text, the system is searched and matched in the system database to obtain the target system text; Input the structured log text, target policy text, and contextual information into the big model for compliance judgment, and obtain the compliance analysis text output by the big model. The contextual information includes the historical records of the relevant equipment, materials, and personnel involved in the structured log text. If the compliance analysis text representation operation is non-compliant, the warning generation process is triggered.

[0006] Furthermore, in one embodiment, the step of constructing a policy library based on the policy documents of the power production system includes: extracting a first preset number of keywords from each institutional document of the power production system; Slicing and dividing each policy document into multiple text blocks according to a preset slice length and a preset overlap length; Perform word embedding encoding on each text block to obtain the corresponding first word embedding vector, and create an index for each first word embedding vector; The step of searching and matching the system in the system database according to the structured log text to obtain the target system text includes: Based on the structured log text and the title and keywords of each policy document, similarity measurement is performed on the structured log text and each policy document, and a second preset number of policy documents with the highest similarity are selected as candidate policy documents; Perform word embedding encoding on the structured log text to obtain the second word embedding vector; Based on the index of the first word embedding vector, a third preset number of first word embedding vectors having the highest similarity to the second word embedding vector are retrieved from the first word embedding vectors corresponding to the candidate system documents as candidate word embedding vectors; The candidate word embedding vector whose similarity meets the threshold requirement is used as the target word embedding vector; The text blocks corresponding to the target word embedding vectors are integrated to obtain the target institutional text.

[0007] Furthermore, in one embodiment, the step of measuring the similarity between the structured log text and each policy document based on the structured log text and the title and keywords of each policy document includes: For each institutional document: the edit distance similarity is calculated based on the structured log text and the title of the institutional document. The Jaccard similarity coefficient is calculated based on the word segmentation results of the structured log text and the keywords of the institutional document. The edit distance similarity and the Jaccard similarity coefficient are weighted and summed to obtain the similarity between the structured log text and the institutional document.

[0008] Furthermore, in one embodiment, the similarity between the first word embedding vector and the second word embedding vector adopts cosine similarity.

[0009] Furthermore, in one embodiment, the step of generating a standard log document based on the operation record document and notification record document of the power production system includes: De-noising the operation record documents and notification record documents of the power production system to obtain de-noised record documents; Perform data integrity check and data filling based on the document type of the denoised record document to obtain a standard record document. If necessary data is missing, a data missing alarm is issued and the manually entered supplementary value is filled in. If non-essential data is missing, the default value is filled in. Generate a standard log document based on the log format template and the corresponding type of standard record document.

[0010] Furthermore, in one embodiment, the step of inputting the standard log document into the large model to extract key content and obtaining the structured log text output by the large model includes: Input the standard log document into the large model to determine the log type and obtain the log type output by the large model; The log type and standard log document are input into the big model to extract key content, and the structured log text output by the big model is obtained. Among them, the log type affects the method of extracting key content.

[0011] Furthermore, in one embodiment, the context information also includes sensitivity adjustment instructions for compliance judgment.

[0012] Furthermore, in one embodiment, the steps of triggering the warning generation process include: Input the target system text and compliance analysis text into the big model to perform risk assessment and generate rectification strategies, and obtain the structured warning text output by the big model; Generate warning messages based on structured warning text and push them to relevant personnel; Generate maintenance work orders based on structured warning texts and push them to relevant personnel.

[0013] Furthermore, in one embodiment, after the step of generating a warning message based on the structured warning text and pushing the warning message to relevant personnel, the following steps are further included: Receive evaluation and feedback from relevant personnel regarding early warning messages; Regularly conduct statistical analysis on evaluation feedback information to identify common causes of false positives or omissions and generate improvement reports, which are used to guide the optimization of large models and system search and matching algorithms.

[0014] Furthermore, in one embodiment, after the step of generating a maintenance work order based on the structured warning text and pushing the maintenance work order to relevant personnel, the following steps are further included: Receive processing result information of maintenance work orders from relevant personnel; The processing result information and the corresponding structured warning text are input into the big model for rectification strategy learning.

[0015] In this application, a system library is constructed based on the system documents of the power production system; a standard log document is generated based on the operation record documents and notification record documents of the power production system; the standard log document is input into the big model to extract key content and obtain structured log text; the system is searched and matched in the system library based on the structured log text to obtain the target system text; the structured log text, the target system text and the context information are input into the big model for compliance judgment to obtain the compliance analysis text; if the compliance analysis text indicates that the operation is non-compliant, the early warning generation process is triggered. Through this application, the compliance, safety and efficiency of power production operations are significantly improved, the shortcomings of traditional methods in real-time, accuracy and initiative are solved, and strong support is provided for the intelligent management of power production. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 Schematic diagram of a flow chart of a method for monitoring and early warning of operational compliance of a power production system in one embodiment of the present application; Figure 2 Schematic diagram of the functional modules of the power production system operation compliance monitoring and early warning system in one embodiment of the present application. DETAILED DESCRIPTION

[0017] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0018] First, some technical terms in this application are explained to facilitate those skilled in the art to understand this application.

[0019] Manual Auditing: Professionals regularly collect various documents and records from the power production process, such as operation tickets, work tickets, equipment maintenance records, and operation logs, and verify their completeness, accuracy, and compliance with relevant specifications and standards. They also conduct on-site inspections at power production sites and manually verify various data within the power production system, such as power consumption data, voltage and current data, and equipment operating parameters. Faced with complex and ever-changing situations, auditors can make flexible, qualitative judgments based on their experience. However, subjectivity can easily lead to discrepancies in judgment, resulting in high labor costs, low efficiency, and difficulty in fully and comprehensively covering the vast power production process.

[0020] A rule-based automation system builds a rules system based on power industry standards and operating procedures. Sensors are used to collect data on equipment operation and operational processes, which is then compared against established rules in real time. If the data violates the rules, an alarm is triggered, notifying personnel. Its high efficiency and accuracy, along with stable and repeatable judgment results, ensure timely monitoring. However, due to the limitations of fixed rules, it struggles to adapt to ever-changing situations and lacks flexibility. Furthermore, maintaining these rules to accommodate business development and regulatory updates is labor-intensive and time-consuming.

[0021] System warnings based on machine learning methods: Machine learning models can learn from massive amounts of historical power production data and automatically discover potential patterns and regularities in the data, rather than relying on fixed rules set by humans. Moreover, with the continuous input of new data, machine learning models can continuously optimize themselves and adapt to the dynamic changes in the power production system, unlike traditional rule-based systems that have high update and maintenance costs. However, machine learning methods are not perfect. They have extremely high requirements for data quality. If the data is biased, missing, or noisy, it may lead to poor model learning results and output incorrect compliance judgment results. At the same time, the model's interpretability is relatively poor. When an early warning occurs, it is difficult for staff to understand the specific logic on which the model makes this judgment.

[0022] In order to make the objectives, technical solutions and advantages of this application clearer, the implementation methods of this application will be further described in detail below with reference to the accompanying drawings.

[0023] In a first aspect, an embodiment of the present application provides a method for monitoring and early warning of operational compliance of a power production system.

[0024] Figure 1 A flow chart of a method for monitoring and early warning the operation compliance of a power production system in one embodiment of the present application is shown.

[0025] Reference Figure 1 In one embodiment, a method for monitoring and early warning of operational compliance of a power production system includes the following steps: S1. Build a system library based on the system documents of the power production system.

[0026] Specifically, it is first necessary to collect the power production system documents within the power station and organize them electronically. To facilitate management and maintenance, they can be classified and stored according to the business areas involved in the system documents (such as photovoltaic power generation, wind power generation, hydropower generation, etc.).

[0027] It should be noted that the policy database needs to fully store the semantic content of the policy documents. The storage format of the policy documents in the policy database is related to the policy search and matching method, which is not specifically limited in this embodiment.

[0028] S2. Generate a standard log document based on the operation record document and notification record document of the power production system.

[0029] Specifically, the power production system includes multiple data sources such as the power production management system, equipment monitoring system, and operation ticket management system. Each data source will record power production operation data in real time, such as equipment operation records, operating parameters, work ticket information, operation instructions, etc., to form an operation record document.

[0030] Furthermore, during the power generation process, relevant notices are issued as needed to guide power generation operations. Unlike regulatory documents, notices are valid for a shorter period and have more specific terms and conditions. Notices are summarized into a notice record document.

[0031] It is understandable that the data in the original operation record documents and notification record documents are relatively scattered, and there may be problems such as noise and missing data. They need to be sorted according to preset rules to generate standard log documents for use in subsequent steps.

[0032] Optionally, the data of a standard log document may come from multiple operation record documents and notification record documents, and the data flow direction is set according to actual needs.

[0033] S3. Input the standard log document into the big model to extract key content and obtain the structured log text output by the big model.

[0034] Specifically, this embodiment selects a suitable pre-trained large model, such as the GPT series or BERT derivative model based on the Transformer architecture, and loads the corresponding model weights according to the language characteristics of standard log documents (such as English, Chinese or a mixture of multiple languages) and business fields.

[0035] For example, the key contents of the operation record data include the name of the operating equipment, the operation type (such as closing, opening, maintenance, etc.), the operation time, the operator, etc., and the key contents of the notification record data include terms and constraints.

[0036] For example, the prompt word instruction is set as: "Based on the log document provided above, please extract the key content in the log document, including but not limited to the operating device name, operation type, operation time, operator, etc. in the operation record data, the terms and constraints in the notification record data, and output the extracted key content in a structured form."

[0037] S4. Search and match the system in the system database based on the structured log text to obtain the target system text.

[0038] Specifically, the information in the structured log text is more concise, which facilitates system search and matching. The target system text obtained by system search and matching contains rules and regulations used to determine whether the relevant operations involved in the structured log text are compliant.

[0039] S5. Input the structured log text, target system text, and context information into the big model to perform compliance judgment, and obtain the compliance analysis text output by the big model. The context information includes historical records of relevant equipment, materials, and personnel involved in the structured log text.

[0040] Specifically, the large model integrates the relevant rules and regulations in the system documents, the relevant terms and constraints in the notification record documents, and the historical record information of relevant equipment, materials, and personnel to make compliance judgments on the actual operations in the operation record documents, to determine whether the relevant operations are compliant and whether there are potential risks or violations.

[0041] For example, the structured log text is: {"Operator Zhang San performed a closing operation on device A at 2023-10-01 10:00", "The voltage of device A before the operation was 0V, and the voltage after the operation was 220V"}.

[0042] The context information is: {Device A had an abnormal alarm at 2023-09-18 10:00}.

[0043] For example, the prompt word instruction is set as: "Based on the provided [structured log text] and [context information], determine whether the operations in the log text comply with the relevant regulations in the [target system text], give a conclusion of compliance or non-compliance, and provide relevant reason analysis."

[0044] S6. If the compliance analysis text characterizes the operation as non-compliant, the warning generation process is triggered.

[0045] Optionally, the warning push program is automatically triggered through a script or interface.

[0046] Optionally, the compliance analysis text is formatted and captured, and output in one of two states, True (compliant) or False (non-compliant), in a computer programming language. When the False state is output, an early warning generation process is triggered.

[0047] This embodiment introduces the semantic understanding capability of the large model, domain knowledge enhancement, and multimodal data fusion technology. The specific implementation method is as follows: Real-time monitoring and rapid response: Traditional methods may rely on post-analysis or periodic inspections, failing to capture operational irregularities in real time. However, large-scale model-driven systems can process massive amounts of data in real time, quickly identifying abnormal operations and ensuring that problems are discovered in their infancy, thereby reducing potential risks.

[0048] Intelligent compliance checking: Using deep learning technology, the large model can accurately understand and apply complex compliance standards, automatically identify non-compliant behaviors in operations, reduce errors and omissions in human judgment, and improve the accuracy and comprehensiveness of compliance checks.

[0049] Accurate early warning and proactive prevention: The system can not only detect existing illegal operations, but also predict potential operational risks by analyzing historical data and real-time information, issue early warnings, and help operators take preventive measures to avoid accidents.

[0050] Efficient processing and automation: Traditional methods may require a lot of manual intervention and tedious data processing, while large model-driven systems enable automated monitoring and early warning, reducing labor costs, improving processing efficiency, and enabling power production systems to operate more efficiently and safely.

[0051] Strong adaptability and scalability: The large model can adapt to different scenarios and complex operations, continuously improving the accuracy of monitoring and early warning through continuous learning and optimization. Furthermore, the system can be expanded to meet the needs of power production, supporting more data sources and more complex analysis requirements.

[0052] Through the above methods, this embodiment significantly improves the compliance, safety and efficiency of power production operations, solves the shortcomings of traditional methods in real-time, accuracy and initiative, and provides strong support for the intelligent management of power production.

[0053] Furthermore, in one embodiment, the step of constructing a policy library based on the policy documents of the power production system includes: extracting a first preset number of keywords from each institutional document of the power production system; Slicing and dividing each policy document into multiple text blocks according to a preset slice length and a preset overlap length; Perform word embedding encoding on each text block to obtain the corresponding first word embedding vector, and create an index for each first word embedding vector; The step of searching and matching the system in the system database according to the structured log text to obtain the target system text includes: Based on the structured log text and the title and keywords of each policy document, similarity measurement is performed on the structured log text and each policy document, and a second preset number of policy documents with the highest similarity are selected as candidate policy documents; Perform word embedding encoding on the structured log text to obtain the second word embedding vector; Based on the index of the first word embedding vector, a third preset number of first word embedding vectors having the highest similarity to the second word embedding vector are retrieved from the first word embedding vectors corresponding to the candidate system documents as candidate word embedding vectors; The candidate word embedding vector whose similarity meets the threshold requirement is used as the target word embedding vector; The text blocks corresponding to the target word embedding vectors are integrated to obtain the target institutional text.

[0054] In this embodiment, the system search and matching adopts the strategy of fuzzy matching + fine matching. The candidate system documents are located from all system documents through fuzzy matching, and the target system text is located from the candidate system documents through fine matching. While improving the retrieval efficiency and reducing the retrieval resource consumption, the retrieval accuracy is ensured.

[0055] Specifically, fuzzy matching uses text calculation, which has low computational complexity and high efficiency. The text involved in the calculation includes keywords in the policy documents, so when building the policy database, it is necessary to perform keyword extraction on the policy documents.

[0056] Fine-grained matching uses word embedding vector calculations, which offer high accuracy but are computationally intensive. When building the policy database, a fixed-length strategy is used for slicing and dicing, which improves processing efficiency. Overlap is set to avoid semantic gaps, and indexes are established for word embedding vectors to speed up retrieval.

[0057] Fuzzy matching selects candidate system documents based on the relative size of similarity, which helps reduce missed detections. Fine matching selects target word embedding vectors based on the relative size of similarity and the absolute size of similarity (threshold), which helps reduce false detections.

[0058] For example, the first preset number is 20, the preset slice length is 300, the preset overlap length is 100, the BERT (Bidirectional Encoder Representations from Transformers) word embedding model is used for word embedding encoding, and the product quantization (PQ) technology is used to establish the index. The second preset number is 5 and the third preset number is 3.

[0059] Furthermore, in one embodiment, the step of measuring the similarity between the structured log text and each policy document based on the structured log text and the title and keywords of each policy document includes: For each institutional document: the edit distance similarity is calculated based on the structured log text and the title of the institutional document. The Jaccard similarity coefficient is calculated based on the word segmentation results of the structured log text and the keywords of the institutional document. The edit distance similarity and the Jaccard similarity coefficient are weighted and summed to obtain the similarity between the structured log text and the institutional document.

[0060] Specifically, the edit distance (Levenshtein distance) is an algorithm that measures the degree of difference between two strings. Its core definition is the minimum number of edit operations (insertion, deletion, or substitution) required to transform string A into string B. Its recursive formula is as follows: , Here, i and j represent the lengths of strings A and B respectively, and c(A[i]≠B[j]) indicates whether the characters are equal, which is 0 if they are equal and 1 otherwise.

[0061] In order to facilitate the comparison of similarities between strings of different lengths, the edit distance is usually normalized to the interval [0, 1] to obtain the edit distance similarity. The formula is as follows: .

[0062] The Jaccard similarity coefficient measures the similarity between two sets by calculating the ratio of the intersection to the union. Assuming S1 and S2 are two sets of text after word segmentation, the Jaccard similarity coefficient formula is: , Among them, |S1∩S2| represents the size of the intersection of sets S1 and S2, and |S1∪S2| represents the size of the union of sets S1 and S2.

[0063] The formula for weighted calculation is: , Among them, w1 and w2 represent the weights of edit distance similarity and Jaccard similarity coefficient respectively, and satisfy w1+w2=1.

[0064] Furthermore, in one embodiment, the similarity between the first word embedding vector and the second word embedding vector adopts cosine similarity.

[0065] Specifically, the calculation formula of cosine similarity is: .

[0066] Furthermore, in one embodiment, the step of generating a standard log document based on the operation record document and notification record document of the power production system includes: De-noising the operation record documents and notification record documents of the power production system to obtain de-noised record documents; Perform data integrity check and data filling based on the document type of the denoised record document to obtain a standard record document. If necessary data is missing, a data missing alarm is issued and the manually entered supplementary value is filled in. If non-essential data is missing, the default value is filled in. Generate a standard log document based on the log format template and the corresponding type of standard record document.

[0067] Specifically, different types of standard record documents are controlled by preset rules to input corresponding log format templates to generate standard log documents.

[0068] For example, the noise data includes garbled characters, special characters, repeated records, etc.

[0069] For example, the necessary data includes timestamp, source identification, device identification, operation type, operator, event description, etc.

[0070] Furthermore, in one embodiment, the step of inputting the standard log document into the large model to extract key content and obtaining the structured log text output by the large model includes: Input the standard log document into the large model to determine the log type and obtain the log type output by the large model; The log type and standard log document are input into the big model to extract key content, and the structured log text output by the big model is obtained. Among them, the log type affects the method of extracting key content.

[0071] In this embodiment, the log type is determined first, and then the key content is extracted based on the log type, which helps to improve the accuracy of the key content extraction.

[0072] For example, the prompt word instruction is set as: "Based on the log document provided above, determine which of the following types the log document belongs to: equipment operation log, inspection operation log, maintenance operation log, scheduling operation log, safety operation log, shift handover log, material borrowing and returning log, training and assessment", "Based on the log document and log type provided above, please extract the key content in the log document, including but not limited to the operating equipment name, operation type, operation time, operator, etc. in the operation record data, and the terms and constraints in the notification record data, and output the extracted key content in a structured form."

[0073] Furthermore, in one embodiment, the context information also includes sensitivity adjustment instructions for compliance judgment.

[0074] In this embodiment, the sensitivity of compliance judgment is flexibly adjusted with the help of the context information field to meet the needs of different scenarios.

[0075] For example, the sensitivity adjustment instructions for compliance judgment are set as: "During high-load operation, the sensitivity can be appropriately reduced to reduce false alarms" and "Now is a special maintenance period, the sensitivity should be increased to ensure safety."

[0076] Furthermore, in one embodiment, the steps of triggering the warning generation process include: Input the target system text and compliance analysis text into the big model to perform risk assessment and generate rectification strategies, and obtain the structured warning text output by the big model; Generate warning messages based on structured warning text and push them to relevant personnel; Generate maintenance work orders based on structured warning texts and push them to relevant personnel.

[0077] In this embodiment, the structured warning text obtained with the help of the big model contains the results of the risk assessment and the recommended rectification strategy. The structured warning text is generated into a warning message and a maintenance work order, which are pushed to the relevant personnel respectively. Non-compliance situations can be notified in a timely manner and countermeasures can be taken to ensure that the problem can be handled in a timely manner.

[0078] For example, the prompt word instruction is set as: "Please evaluate the severity of the violation according to the three risk levels of low, medium, and high based on the provided system text and compliance analysis text, and analyze the possible consequences (such as equipment damage, safety accidents, etc.), generate a rectification strategy, and output the assessment results and rectification strategy in a structured form."

[0079] For example, the structured warning text includes the following fields: Log information: operation time, operator, device name, operation type, etc. Basis of violation: the specific rules and regulations and their constraints that are matched; Risk level: determined based on risk assessment results; Corrective measures: Provide targeted solutions.

[0080] For example, using instant messaging tools (such as DingTalk or WeChat Work) or SMS services, you can send warning messages to operations personnel, security managers, or business managers. Warning message content can be customized using templates, allowing you to generate personalized warning messages based on the needs of different roles.

[0081] Optionally, the power production system's aerial dashboard displays warning status using color coding (e.g., green for normal, yellow for warning, and red for emergency). Interactive query functionality is also supported, allowing users to view detailed information by clicking on a specific warning point.

[0082] Furthermore, in one embodiment, after the step of generating a warning message based on the structured warning text and pushing the warning message to relevant personnel, the following steps are further included: Receive evaluation and feedback from relevant personnel regarding early warning messages; Regularly conduct statistical analysis on evaluation feedback information to identify common causes of false positives or omissions and generate improvement reports, which are used to guide the optimization of large models and system search and matching algorithms.

[0083] In this embodiment, closed-loop management is achieved with the help of evaluation feedback information, thereby continuously improving the accuracy of compliance monitoring and early warning during use.

[0084] For example, the evaluation feedback information includes accuracy evaluation, false positive / false negative analysis, improvement suggestions, etc.

[0085] Optionally, based on the improvement report, technicians can adjust the prompt word instructions and parameters of the large model or retrain the model to improve its judgment accuracy and optimize the fuzzy matching algorithm, such as adjusting the weight parameters to reduce the false alarm rate.

[0086] Furthermore, in one embodiment, after the step of generating a maintenance work order based on the structured warning text and pushing the maintenance work order to relevant personnel, the following steps are further included: Receive processing result information of maintenance work orders from relevant personnel; The processing result information and the corresponding structured warning text are input into the big model for rectification strategy learning.

[0087] In this embodiment, closed-loop management is achieved with the help of processing result information. When generating a rectification strategy, successful rectification cases in historical data can be used as reference, thereby continuously improving the reliability of the rectification strategy during use.

[0088] Figure 2 A schematic diagram of the functional modules of a power production system operation compliance monitoring and early warning system in one embodiment of the present application is shown.

[0089] Reference Figure 2 The power production system operation compliance monitoring and early warning system includes data collection and access module, data cleaning and standardization module, semantic understanding and feature extraction module, rule matching and association analysis module, compliance intelligent judgment module, early warning generation and distribution module and feedback learning and system optimization module.

[0090] The data collection and access module is used to collect power production operation data and regulatory text data from multiple data sources in real time. It collects equipment operation records, operating parameters, work ticket information, and operating instructions from power production management systems, equipment monitoring systems, and operation ticket management systems. It also collects electronic regulatory text data from the enterprise or organization's regulatory database. It provides a real-time data access interface and supports multiple data formats.

[0091] The Data Cleansing and Standardization module cleans and standardizes log data, providing high-quality data input for subsequent analysis. It removes noise from logs, including garbled characters, special characters, and duplicate records. It also fills in or removes records missing key fields (such as timestamps and operator names). It also standardizes logs from different sources according to predefined log format templates, ensuring that each log entry contains key fields such as timestamps, source identifiers, and event descriptions.

[0092] The semantic understanding and feature extraction module uses a large model to perform log semantic analysis and extract key information. It loads a pre-trained large model and applies corresponding model weights based on the log's language characteristics and business domain. The large model extracts key operational information from the log, such as the device name, operation type, operation time, and operator. It also extracts key clauses and constraints from the rules and regulations, and combines this with contextual information from the log (such as device status and operator permissions) to understand the semantics of the operation record.

[0093] The rule matching and association analysis module performs fuzzy matching in the rules and regulations index to identify rules and regulations related to the key information in the logs. It then electronically organizes the text of the rules and regulations and builds an index based on the subject matter, keywords, and business areas involved. It uses a fuzzy string matching algorithm (such as the edit distance algorithm and the Jaccard similarity coefficient algorithm) combined with semantic similarity judgment to perform matching. Based on the matching score and semantic relevance, it selects a set of rules and regulations related to the key information in the logs.

[0094] The compliance intelligent judgment module is responsible for integrating information and submitting it to the big model for compliance judgment. It integrates the set of rules and regulations selected by fuzzy matching with the original log text and key information extracted in the semantic analysis stage to form a complete prompt word input. The big model is used to conduct in-depth analysis and judgment of the integrated input information to determine whether the operation is compliant and whether there are potential risks or violations. According to the severity of the violation, the risk level is assessed (such as low risk, medium risk, and high risk).

[0095] The early warning generation and distribution module is used to issue early warning notifications and collect feedback information. It generates detailed early warning information based on the compliance judgment results, including log information, violation judgment basis, risk level, etc., and sends the early warning information to relevant operation and maintenance personnel, security management personnel or business managers. It supports multiple notification methods and records the early warning information in the database for subsequent query and analysis.

[0096] The feedback learning and system optimization module optimizes the system based on feedback information. It collects feedback from processing personnel on the warning information, such as whether the warning is accurate and whether there are any false positives or missed warnings. Based on this feedback, it fine-tunes the large model or performs domain adaptation to improve the model's accuracy and adaptability. It also adjusts the fuzzy matching algorithm and rules based on this feedback to reduce false positives and missed warnings. It also monitors the system's operational performance to promptly identify and resolve potential issues.

[0097] It should be noted that the serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0098] The terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned drawings are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally includes steps or units that are not listed, or optionally includes other steps or units inherent to these processes, methods, products or devices. The terms "first", "second" and "third" are used to distinguish different objects, etc., and do not represent a sequence, nor do they limit the "first", "second" and "third" to different types.

[0099] In the description of the embodiments of this application, the words "exemplary," "for example," or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary," "for example," or "for example" in the embodiments of this application should not be construed as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplary," "for example," or "for example" is intended to present the relevant concepts in a concrete manner.

[0100] In the description of the embodiments of the present application, unless otherwise specified, “ / ” means or, for example, A / B can mean A or B; “and / or” in the text is merely a description of the association relationship of associated objects, indicating that three relationships may exist, for example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, in the description of the embodiments of the present application, “multiple” refers to two or more than two.

[0101] In some processes described in the embodiments of the present application, multiple operations or steps are included that appear in a specific order. However, it should be understood that these operations or steps may not be performed in the order in which they appear in the embodiments of the present application or may be performed in parallel. The sequence numbers of the operations are only used to distinguish between different operations, and the sequence numbers themselves do not represent any order of execution. In addition, these processes may include more or fewer operations, and these operations or steps may be performed in sequence or in parallel, and these operations or steps may be combined.

[0102] Through the description of the above embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, or the part that contributes to the existing technology, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above and includes a number of instructions for enabling a terminal device to execute the methods described in each embodiment of this application.

[0103] The above are only preferred embodiments of the present application and do not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method for monitoring and early warning of operational compliance of a power production system, characterized in that: The power production system operation compliance monitoring and early warning method includes: Build a system library based on the system documents of the power production system; Generate standard log files based on the operation log files and notification log files of the power production system; Input the standard log document into the big model to extract key content and obtain the structured log text output by the big model; According to the structured log text, the system is searched and matched in the system database to obtain the target system text; Input the structured log text, target policy text, and contextual information into the big model for compliance judgment, and obtain the compliance analysis text output by the big model. The contextual information includes the historical records of the relevant equipment, materials, and personnel involved in the structured log text. If the compliance analysis text representation operation is non-compliant, the warning generation process is triggered.

2. The method for monitoring and early warning of operational compliance of a power production system according to claim 1, wherein: The step of constructing a system library based on the system documents of the power production system includes: extracting a first preset number of keywords from each institutional document of the power production system; Slicing and dividing each policy document into multiple text blocks according to a preset slice length and a preset overlap length; Perform word embedding encoding on each text block to obtain the corresponding first word embedding vector, and create an index for each first word embedding vector; The step of searching and matching the system in the system database according to the structured log text to obtain the target system text includes: Based on the structured log text and the title and keywords of each policy document, similarity measurement is performed on the structured log text and each policy document, and a second preset number of policy documents with the highest similarity are selected as candidate policy documents; Perform word embedding encoding on the structured log text to obtain the second word embedding vector; Based on the index of the first word embedding vector, a third preset number of first word embedding vectors having the highest similarity to the second word embedding vector are retrieved from the first word embedding vectors corresponding to the candidate system documents as candidate word embedding vectors; The candidate word embedding vector whose similarity meets the threshold requirement is used as the target word embedding vector; The text blocks corresponding to the target word embedding vectors are integrated to obtain the target institutional text.

3. The method for monitoring and early warning of operational compliance of a power production system according to claim 2, wherein: The step of measuring the similarity between the structured log text and each policy document based on the structured log text and the title and keywords of each policy document includes: For each institutional document: the edit distance similarity is calculated based on the structured log text and the title of the institutional document. The Jaccard similarity coefficient is calculated based on the word segmentation results of the structured log text and the keywords of the institutional document. The edit distance similarity and the Jaccard similarity coefficient are weighted and summed to obtain the similarity between the structured log text and the institutional document.

4. The method for monitoring and early warning of operational compliance of a power production system according to claim 2, wherein: The similarity between the first word embedding vector and the second word embedding vector adopts cosine similarity.

5. The method for monitoring and early warning of operational compliance of a power production system according to claim 1, wherein: The step of generating a standard log document based on the operation record document and notification record document of the power production system includes: De-noising the operation record documents and notification record documents of the power production system to obtain de-noised record documents; Perform data integrity check and data filling based on the document type of the denoised record document to obtain a standard record document. If necessary data is missing, a data missing alarm is issued and the manually entered supplementary value is filled in. If non-essential data is missing, the default value is filled in. Generate a standard log document based on the log format template and the corresponding type of standard record document.

6. The method for monitoring and early warning of operational compliance of a power production system according to claim 1, wherein: The step of inputting the standard log document into the large model to extract key content and obtaining the structured log text output by the large model includes: Input the standard log document into the large model to determine the log type and obtain the log type output by the large model; The log type and standard log document are input into the big model to extract key content, and the structured log text output by the big model is obtained. Among them, the log type affects the method of extracting key content.

7. The method for monitoring and early warning of operational compliance of a power production system according to claim 1, wherein: The contextual information also includes sensitivity adjustment instructions for compliance judgments.

8. The method for monitoring and early warning of operational compliance of a power production system according to claim 1, wherein: The steps of triggering the early warning generation process include: Input the target system text and compliance analysis text into the big model to perform risk assessment and generate rectification strategies, and obtain the structured warning text output by the big model; Generate warning messages based on structured warning text and push them to relevant personnel; Generate maintenance work orders based on structured warning texts and push them to relevant personnel.

9. The method for monitoring and early warning of operational compliance of a power production system according to claim 8, wherein: After the step of generating a warning message according to the structured warning text and pushing the warning message to relevant personnel, the method further includes: Receive evaluation and feedback from relevant personnel regarding early warning messages; Regularly conduct statistical analysis on evaluation feedback information to identify common causes of false positives or omissions and generate improvement reports, which are used to guide the optimization of large models and system search and matching algorithms.

10. The method for monitoring and early warning of operational compliance of a power production system according to claim 8, wherein: After the step of generating a maintenance work order according to the structured warning text and pushing the maintenance work order to relevant personnel, the following steps are also included: Receive processing result information of maintenance work orders from relevant personnel; The processing result information and the corresponding structured warning text are input into the big model for rectification strategy learning.

Images