Basic-level power supply enterprise compliance risk early warning system and method based on big data analysis

Abstract

The invention discloses a grassroots power supply enterprise compliance risk intelligent system and method based on big data analysis. The data acquisition unit is used for acquiring business operation data, historical violation records and policy and regulation update data of basic power supply enterprises to form a unified compliance data set. And the natural language processing unit performs text word segmentation and correlation analysis on the policy and regulation and violation record data, extracts key risk factors and labels compliance risk labels. And the risk feature construction unit performs multi-dimensional feature fusion on the business operation data and the compliance risk label data to generate a feature matrix for risk identification. And the intelligent risk assessment unit performs real-time analysis on the feature matrix by using a pre-trained machine learning model, identifies compliance risk categories and levels, and generates early warning information. According to the method, the accuracy of compliance risk identification is improved by using big data analysis and an intelligent algorithm, the compliance management cost of basic-level power supply enterprises is reduced, and the operation safety and compliance of the enterprises are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power intelligent early warning systems, and in particular to a compliance risk early warning system for grassroots power supply enterprises based on big data analysis. Background Art

[0002] Currently, compliance management at grassroots power supply companies primarily relies on manual review and traditional information management systems. These systems typically manage data and identify risks based on rule-based settings and manual data entry. Compliance risk assessments primarily rely on regular inspections and manual comparisons of historical violation records with policy and regulatory changes. At the same time, some companies have introduced information technology, such as database management systems and basic data analysis tools, to improve the efficiency of compliance management.

[0003] However, existing compliance management methods have multiple problems. First, the manual review model is inefficient and difficult to capture the risks brought about by regulatory changes in a timely manner, which can easily lead to delayed compliance risks. Second, traditional information management systems lack intelligent analysis capabilities, are unable to deeply explore historical violation patterns, and have difficulty in making accurate risk predictions. In addition, existing risk identification methods mostly use static rule matching and lack the ability to dynamically learn and adaptively optimize data, resulting in insufficient accuracy and flexibility in compliance risk assessment. Especially in grassroots power supply companies, due to the complexity of business scenarios and diverse data sources, existing technologies have difficulty in achieving comprehensive and automated risk warnings.

[0004] Therefore, it is necessary to develop a compliance risk early warning system for grassroots power supply enterprises based on big data analysis. Summary of the Invention

[0005] This application provides a compliance risk early warning system for grassroots power supply enterprises based on big data analysis to reduce the compliance management costs of grassroots power supply enterprises.

[0006] This application provides a compliance risk early warning system for grassroots power supply enterprises based on big data analysis, including:

[0007] The data collection unit is used to collect business operation data, historical violation records, and policy and regulatory update data from grassroots power supply enterprises, and perform data cleaning and format standardization to form a unified compliance data set;

[0008] A natural language processing unit is used to perform text segmentation, semantic recognition, and association analysis on the policy and regulatory update data and historical violation record data, extract key risk factors, and label them with corresponding compliance risk tags;

[0009] a risk feature construction unit, configured to perform multi-dimensional feature fusion on the business operation data and the risk label data processed by the natural language processing unit to form a feature matrix for risk identification;

[0010] An intelligent risk assessment unit is used to use a pre-trained machine learning model to perform real-time compliance risk identification on the feature matrix, determine the category and level of the risk, and generate corresponding early warning information based on the risk category and level.

[0011] Also includes:

[0012] The early warning and risk avoidance suggestion unit is connected to the intelligent risk assessment unit and is used to automatically generate targeted risk avoidance suggestions based on the identified risk categories and levels, and push the early warning information and avoidance suggestions to the compliance management terminal of the grassroots power supply enterprise to guide the enterprise to actively implement risk prevention measures.

[0013] The data acquisition unit includes an incremental regulatory sensitive data capture module, which is used to update the collected policy and regulatory data and identify sensitive clauses directly related to the specific business activities of grassroots power supply enterprises; automatically generate data capture trigger rules with time stamps and regulatory association identifiers based on the identified sensitive clauses; according to the trigger rules, capture data that meets the characteristics of sensitive clauses in the business operation data stream of the grassroots power supply enterprise, and separately mark them to form a sensitive compliance data subset.

[0014] The data collection unit includes a violation risk data backtracking collection module, which is used to automatically analyze the historical violation record data corresponding to the risk category when the intelligent risk assessment unit detects a specific category of compliance risk and generates early warning information to determine the key violation event characteristics related to the current risk; based on the determined key violation event characteristics, generate data backtracking rules, including the time window of the violation event, the type of equipment involved or the position of the responsible person; based on the data backtracking rules, conduct a secondary backtracking collection of the historical business operation data stored by the grassroots power supply enterprise, and generate a special backtracking data set with violation association marks for continuous optimization of the subsequent risk feature model.

[0015] The natural language processing unit is specifically used for:

[0016] Using a multi-granularity domain word embedding model specifically built for grassroots power supply companies, we dynamically segment and embed the data on policy and regulatory updates and historical violation records to capture the implicit semantic relationships between different regulatory terms and corporate compliance terms.

[0017] Using the word vector mapping results, a semantic relationship network in the field of compliance management of power supply enterprises is constructed through semantic similarity analysis to clarify the semantic associations between key terms in different regulatory provisions and violation records;

[0018] Based on the semantic similarity and co-occurrence frequency between nodes in the semantic relationship network, cross-text risk factor association analysis is performed to form an accurate risk factor combination and annotate the corresponding compliance risk labels.

[0019] The word vector mapping process of the natural language processing unit includes:

[0020] Based on the node weights of key terms in the constructed semantic relationship network, the semantic relevance parameters of the word vector mapping are dynamically adjusted to reflect the semantic change trends of regulatory clause updates while generating semantic similarity.

[0021] Combined with the context of the newly added regulatory provisions, the word vector embedding representation is automatically updated, and through an adaptive iterative process, the word vector is kept consistent with the original semantic network, enhancing the accuracy of the association analysis of cross-text risk factors.

[0022] The risk feature construction unit is specifically used to:

[0023] Based on the business categories covered by regulatory requirements, business operation data is divided into equipment management data, operation and maintenance scheduling data, and financial compliance data. Each data category is further subdivided based on the constraints of regulatory provisions. Equipment management data is categorized by inspection frequency, maintenance record completeness, and equipment load status. Operation and maintenance scheduling data is categorized by scheduling response time, task completion status, and personnel deployment. Financial compliance data is categorized by capital flow records, contract execution status, and expenditure approval process.

[0024] When matching risk tag data with business data, the system extracts the corresponding data fields from the business data based on the numerical restrictions, time requirements, and operational specifications of regulatory provisions, and calculates their degree of compliance with regulatory requirements. For regulatory provisions that include time restrictions, the system calculates the time intervals between data fields and marks data that exceeds the regulatory limits as anomalies. For regulatory provisions that include operational process requirements, the system calculates the task completion status of the business data and marks unfinished or overdue data as anomalies.

[0025] When forming the feature matrix, based on the time series analysis method, the time interval difference, historical violation frequency and cumulative violation days are extracted, and the trend characteristics of historical data are added to the feature matrix, so that the system can identify the cumulative effect of long-term violation risks and provide a basis for evaluating regulatory compliance in different time windows.

[0026] When calculating the time interval of the data field, the risk feature construction unit dynamically adjusts the selection range of the time window based on the timeliness requirements of the regulatory provisions and the historical operating model of the enterprise; among them, for the case where the regulatory requirements clearly stipulate a fixed time period, the time threshold stipulated by the regulatory requirements is directly used for calculation; for the case where the regulations do not clearly provide a fixed time period but there are industry practices or internal management standards of the enterprise, the most common time interval is calculated based on historical business data, and the time interval is used as a benchmark for compliance judgment; in the case of regulatory updates or business process adjustments, the time window change trends of the new and old regulatory requirements are compared, and the time interval calculation rules are dynamically adjusted to ensure that the feature matrix can adapt to regulatory changes and improve the accuracy of compliance risk assessment.

[0027] The machine learning model of the intelligent risk assessment unit includes a compliance feature extraction network, a risk classification network and a dynamic adaptive adjustment module;

[0028] The input of the compliance feature extraction network is the time dimension features, business indicator features, historical behavior features, and regulatory adaptability features in the feature matrix. It is used to perform time series expansion of the time dimension features based on the hierarchical feature decomposition method, extract task interval deviations and regulatory execution time limit compliance, calculate the time decay weights of key business data, and perform normalization processing on the business indicator features so that the equipment operation status, inspection frequency, and task completion rate features are calculated under a unified numerical scale.

[0029] The input of the risk classification network is the normalized compliance feature vector generated by the compliance feature extraction network. Based on the regulation matching rules and classification enhancement mechanism, the network identifies the compliance risk category of the input features, and combines the number of violations and rectification efficiency information in the historical behavior characteristics to calculate the risk level and generate a risk category label.

[0030] The input of the dynamic adaptive adjustment module includes regulatory update information and historical prediction deviation data of the risk classification network. The module parses the regulatory adjustment content based on the regulatory change detection mechanism, and combines the regulatory adaptability characteristics to calculate the degree of deviation between the criticality of regulatory requirements and corporate behavior, and dynamically adjusts the classification weights and risk level division thresholds of the risk classification network.

[0031] A compliance risk early warning method for grassroots power supply enterprises based on big data analysis, including:

[0032] Collect business operation data, historical violation records, and policy and regulatory update data from grassroots power supply companies, cleanse and standardize the data to form a unified compliance data set;

[0033] Perform text segmentation, semantic recognition, and association analysis on the policy and regulatory update data and historical violation record data to extract key risk factors and label them with corresponding compliance risk tags;

[0034] Based on the business operation data and the annotated compliance risk labels, a feature matrix for risk identification is constructed, wherein the feature matrix includes fusion features of multiple dimensions;

[0035] The feature matrix is ​​input into the trained machine learning model to perform real-time compliance risk identification, determine the risk category and risk level corresponding to each business data instance, and generate compliance risk warning information based on the risk category and risk level.

[0036] This application has the following beneficial technical effects:

[0037] Through big data analysis and machine learning technology, the system can automatically identify compliance risks of grassroots power supply companies and assess risk categories and levels in real time. Compared with traditional manual review models, it improves identification accuracy and reduces potential violations caused by lags. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a schematic diagram of a compliance risk warning system for grassroots power supply enterprises based on big data analysis provided in the first embodiment of this application.

[0039] Figure 2 This is a schematic diagram of a compliance risk warning method for grassroots power supply enterprises based on big data analysis provided in the second embodiment of this application. DETAILED DESCRIPTION

[0040] The following description sets forth many specific details to facilitate a thorough understanding of the present application. However, the present application can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the scope of the present application. Therefore, the present application is not limited to the specific implementations disclosed below.

[0041] The first embodiment of this application provides a compliance risk warning system for grassroots power supply enterprises based on big data analysis. Figure 1 , which is a schematic diagram of the first embodiment of the present application. Figure 1 The first embodiment of this application provides a detailed description of a compliance risk warning system for grassroots power supply enterprises based on big data analysis.

[0042] The compliance risk early warning system for grassroots power supply enterprises based on big data analysis includes a data acquisition module 101, a natural language processing unit 102, a risk feature construction unit 103 and an intelligent risk assessment unit 104.

[0043] The data collection unit 101 is used to collect business operation data, historical violation record data, and policy and regulation update data of grassroots power supply enterprises, and perform data cleaning and format standardization to form a unified compliance data set.

[0044] The data acquisition unit 101 is the basic module of the present invention, which is responsible for acquiring original data related to the compliance management of grassroots power supply enterprises from multiple sources and pre-processing the data to ensure the accuracy and stability of subsequent analysis. The unit includes multiple data interface modules to connect to the business management system, power dispatching system, financial system, human resources management system, etc. within the power supply enterprise to realize real-time collection of daily business operation data. In addition, the unit is also equipped with an external data acquisition module for accessing industry regulatory agencies, policy release platforms, government public data resources, etc., to ensure that enterprises can obtain the latest regulatory change information in the first place. In addition, the unit also supports docking with the historical violation record database to obtain the company's past compliance and violation information, providing a reference for subsequent risk identification.

[0045] In order to improve the availability of data, the data acquisition unit 101 integrates a set of efficient data cleaning and preprocessing mechanisms. During the data acquisition process, the unit first performs a data integrity check and automatically removes data entries with serious missing values ​​or incorrect formats. Secondly, the unit performs structured conversion on the data based on the set standard rules. For example, different systems may use different timestamp formats, unit representations, or coding standards. The unit can unify the time format, convert data units, and map codes through standardized rules, so that the data can be analyzed under a unified framework. In addition, the unit uses a redundant deduplication algorithm to compare data from different sources, eliminate duplicate data, and improve data quality.

[0046] The data collection unit 101 also includes a dynamic data weight allocation mechanism to enhance the adaptability of data in compliance risk identification. Since different types of data have different impact weights on compliance risks, for example, policy and regulatory update information usually has a higher influence, while historical violation records provide a reference for risk trends. The unit can dynamically adjust the weights of various types of data based on historical statistics and machine learning algorithms. For example, when new regulations are detected, the unit can temporarily increase the priority of regulatory data to ensure its influence in subsequent analysis and improve the real-time nature of early warning. In addition, the unit also has an abnormal data labeling function that can identify possible false alarms or mis-collected data, and correct or isolate the data before it enters the next step of processing.

[0047] Data acquisition unit 101 also features an adaptive data sampling mechanism. To address the vast volume of business data generated by power supply companies, it employs an intelligent data extraction algorithm that dynamically adjusts sampling frequency and data granularity based on system load and analysis requirements. For example, for routine business data, the unit can use batch sampling to reduce computational pressure. When high-risk business activity is detected, the unit automatically switches to high-frequency sampling to ensure the integrity of critical event data.

[0048] In summary, the data acquisition unit 101 not only provides comprehensive data collection functions, but also improves the accuracy, real-time nature and adaptability of the data through technical means such as data cleaning, standardization, dynamic weight adjustment, adaptive sampling and storage optimization, laying a solid foundation for subsequent intelligent compliance risk analysis.

[0049] Furthermore, the data acquisition unit includes an incremental regulatory sensitive data capture module, which is used to update the collected policy and regulatory data and identify sensitive clauses directly related to the specific business activities of the grassroots power supply enterprises; automatically generate data capture trigger rules with time stamps and regulatory association identifiers based on the identified sensitive clauses; according to the trigger rules, capture data that meets the characteristics of sensitive clauses in the business operation data flow of the grassroots power supply enterprise, and separately mark them to form a sensitive compliance data subset.

[0050] The incremental regulatory-sensitive data capture module plays a key role in the entire data collection unit. Its core function is to ensure that grassroots power supply companies can accurately identify sensitive clauses directly related to their own business activities after regulatory updates, and proactively adjust data collection strategies to ensure that the collection process always complies with the latest regulatory requirements, thereby improving the timeliness and pertinence of compliance management.

[0051] During the specific implementation process, the module first needs to parse the policy and regulatory update data to extract sensitive clauses that may affect the operations of grassroots power supply companies. This process usually involves natural language processing technology, including text segmentation, named entity recognition, syntactic analysis, and semantic understanding methods to ensure that the system can accurately understand the key content of the regulatory clauses. For example, when new power safety regulatory requirements are issued, the module can automatically extract clauses related to substation inspections, fault reporting, and electricity customer information protection, and structure the content of the clauses to further analyze their scope of application. In order to improve the accuracy of recognition, the module can combine the historical compliance data of the power supply company to determine which types of regulatory changes have had a direct impact on the company's operations in the past, thereby prioritizing the identification of similar clauses and automatically performing correlation analysis to ensure that the system can accurately focus on regulatory changes that have a real impact on the company.

[0052] After extracting sensitive clauses, the module automatically generates data capture trigger rules based on the identified results. These rules not only include time stamps for regulatory clauses but also establish links between regulations and data based on the regulatory scope, the company's business type, and compliance requirements. For example, if a regulation requires an enterprise to submit a detailed incident report within 24 hours of a power grid anomaly, the module will automatically generate a time-constrained trigger rule to ensure that the system prioritizes the collection of relevant operation and maintenance data, dispatch records, and personnel disposition information after the anomaly occurs, and to check whether a compliance report has been generated after 24 hours. Furthermore, the module can generate categorized trigger rules based on the different application scenarios of the regulation. For example, for equipment management regulations, the trigger rules may require the system to focus on equipment operation logs, while for financial compliance requirements, the focus may be on collecting cash flow data and contract execution. Trigger rule generation can be based on rule template matching or can be combined with machine learning models to dynamically optimize through historical data analysis, allowing the rules to adapt to evolving regulatory requirements.

[0053] After the regulatory sensitive clauses and data capture trigger rules are established, the module will actively filter and mark data that meets the characteristics of sensitive clauses in the business operation data stream of the grassroots power supply enterprise based on the trigger rules, forming an independent sensitive compliance data subset. During the data capture process, the module can monitor the enterprise's business data stream in real time, including substation maintenance records, grid load data, customer electricity usage behavior, etc., and compare it with regulatory requirements to ensure that data that meets specific regulatory standards can be accurately identified and stored. For example, when regulations require enterprises to record the entire process of specific high-risk operations, the module can continuously monitor the relevant operation data and automatically filter the log content that meets the regulatory requirements to avoid compliance risks caused by missing data. In addition, the module also supports incremental data capture, that is, after the regulations are updated, only the data related to the new or changed clauses will be additionally collected, without the need to reprocess all historical data, thereby reducing the system's computing burden and improving the efficiency of data collection.

[0054] The sensitive compliance data subset generated by this module not only has independent storage properties, but also contains complete time stamps and regulatory association identifiers, so that in the subsequent compliance risk assessment process, the system can quickly query the data corresponding to specific regulations to ensure the accuracy of regulatory adaptability analysis. In addition, when the module detects regulatory adjustments, it can trigger a data backtracking mechanism to check whether the company has collected data that meets the requirements of the new regulations before and after the regulations come into effect, and supplement the collection of missing data when necessary to ensure compliance during the regulatory transition period. This mechanism enables companies to be more proactive in responding to sudden regulatory changes, rather than passively waiting for manual adjustments to data collection strategies, thereby reducing compliance risks caused by delayed regulatory adaptation.

[0055] The design of the incremental regulatory-sensitive data capture module enables grassroots power supply companies to automatically adapt their collection strategies when regulations are adjusted, ensuring the accuracy, timeliness and regulatory compliance of data collection, reducing the need for manual intervention, and optimizing the system's data processing efficiency, thereby improving the intelligence level of compliance management.

[0056] Furthermore, the data collection unit includes a violation risk data backtracking collection module, which is used to automatically analyze the historical violation record data corresponding to the risk category when the intelligent risk assessment unit detects a specific category of compliance risk and generates early warning information to determine the key violation event characteristics related to the current risk; based on the determined key violation event characteristics, generate data backtracking rules, including the time window of the violation event, the type of equipment involved or the position of the responsible personnel; based on the data backtracking rules, conduct secondary backtracking collection of the historical business operation data stored by the grassroots power supply enterprise, and generate a special backtracking data set with violation association marks for continuous optimization of the subsequent risk characteristic model.

[0057] The Violation Risk Data Retrospective Collection Module plays a key role in data tracing and deep mining within the entire system. Its primary mission is to ensure that, after the intelligent risk assessment unit detects a specific compliance risk, it can effectively trace back historical data, identify potential violation patterns, and enhance the accuracy and specificity of risk signature modeling through secondary data collection. This module not only focuses on current business data but also, through in-depth analysis of historical violation records and business data, builds a compliance risk assessment system more tailored to the operational characteristics of grassroots power supply enterprises.

[0058] In actual applications, the module first automatically triggers the analysis of historical violation record data after the intelligent risk assessment unit generates an early warning message. To ensure the targeted analysis, the module does not simply query all historical violation data, but uses pattern recognition and feature matching methods to screen out violation cases that are highly relevant to the current risk event. For example, if the intelligent risk assessment unit identifies that a certain type of substation inspection process does not meet the latest regulatory requirements, the module will retrieve past violation records and screen out historical violation events involving excessive inspection time, incomplete key point inspections, data upload delays, etc. In order to improve the accuracy of the analysis, the module will also use text analysis and data mining technology to perform semantic analysis on the description content of historical violation records to ensure that violations with different expressions but similar essence can be accurately classified.

[0059] After identifying the characteristics of historical violations, the module further generates a set of data backdating rules based on statistical and time series analysis. The core function of these rules is to determine which historical business data needs to be recollected to more comprehensively assess current compliance risks. These rules typically include multiple dimensions, such as time windows, the types of equipment involved, and the positions of responsible personnel. For example, if historical data analysis indicates that the failure rate of a certain type of equipment increases significantly during specific seasons or climate conditions, and the current risk warning involves operation and maintenance issues with this type of equipment, the module may automatically set a backdating rule requiring the system to retrieve operating data for the equipment under similar seasons or specific load conditions. Furthermore, based on the responsible personnel's historical operation records, the module can filter out relevant data on operation and maintenance personnel who have been involved in similar violations, ensuring that the risk assessment fully considers the impact of human factors. This mechanism not only helps to accurately identify the causes of current risks but also identifies potential systemic management loopholes, thereby improving the foresight of compliance management.

[0060] Based on the generated data backtracking rules, the module will conduct a secondary backtracking collection of the historical business operation data stored by the grassroots power supply enterprises and generate a special backtracking data set. During the backtracking collection process, the module will not simply copy the existing data, but adopt a differentiated data extraction method to ensure that the newly collected data can supplement the deficiencies of the current risk assessment. For example, if some inspection records are missing in the historical data, the system will prioritize retrieving data from adjacent time periods based on the time window rules to infer the missing information, and call the work logs or voice records of the relevant responsible persons when necessary to obtain supplementary information. In addition, the module supports cross-data source matching, that is, when it is found that a violation involves multiple business links, it will automatically extract relevant information from different data sources, such as integrating data from the equipment monitoring system, scheduling record system and manual inspection reports to form a more comprehensive violation event backtracking data set. This collection method makes the data more timely and complete, thereby improving the accuracy of subsequent risk feature modeling.

[0061] The generated special retrospective data set is not only used for in-depth analysis of current risk events, but also fed back to the risk feature construction unit to optimize the system's risk prediction capabilities. This module can automatically adjust the parameters of the risk assessment model based on the retrospectively collected data, so that the model can more accurately reflect the actual operation of grassroots power supply enterprises. For example, if the incidence of a certain type of violation risk under specific conditions is much higher than other conditions, the module will automatically set this condition as a high-weight feature, making the system more sensitive when predicting similar risks in the future. In addition, this module can also be used for regulatory adaptability analysis, that is, after the regulations are adjusted, historical business data is traced back to assess the potential compliance risks of enterprises under the new regulations, thereby providing data support for enterprises to formulate compliance strategies in advance.

[0062] The design of the violation risk data retrospective collection module enables the system to quickly locate relevant historical data upon identifying compliance risks and optimize risk signature models based on retrospective data collection. This mechanism not only improves the accuracy of compliance management but also enables the system to continuously optimize in response to regulatory adjustments and business changes, providing grassroots power supply companies with more intelligent and precise risk warning and management solutions.

[0063] The natural language processing unit 102 is used to perform text segmentation, semantic recognition and association analysis on the policy and regulation update data and historical violation record data, extract key risk factors, and mark corresponding compliance risk tags.

[0064] Natural language processing unit 102 undertakes the critical task of parsing and understanding relevant regulatory information and historical violation records for grassroots power supply enterprises, ensuring the system can accurately identify changes in policies and regulations, patterns of non-compliance, and provide accurate data support for risk assessment. This unit primarily consists of a text data parsing module, a semantic recognition module, a risk factor extraction module, and a compliance risk label generation module. These modules work together to achieve in-depth analysis of regulations and historical data.

[0065] First, the text data parsing module processes regulatory text and historical violation records related to the power supply industry. Regulatory data typically comes from government regulatory agencies, industry standards-setting organizations, and internal corporate compliance management systems, while historical violation records come from publicly available data within the company itself or from peers. Because this data comes in a variety of formats, including PDF files, HTML web pages, plain text files, and structured database records, this module first performs format conversion, converting different types of text data into a standardized, parseable text format. It then uses optical character recognition (OCR) technology to identify text content within scanned documents to ensure data integrity and readability.

[0066] After text parsing is complete, the semantic recognition module uses natural language processing (NLP) technology to gain a deep understanding of the regulatory text and violation records. This module first performs word segmentation on the text to facilitate subsequent grammatical and semantic analysis. The power supply industry is known for its extensive terminology. To improve recognition accuracy, the module is pre-trained with a specialized dictionary and corpus based on the industry, enabling it to accurately identify specialized terminology, specific business activities, and relevant compliance requirements. Furthermore, the module applies syntactic analysis and dependency parsing techniques to extract the core structure of the regulatory text, ensuring the system understands key constraints, prohibitions, and exemptions within the text. For example, in a regulation stating that "grassroots power supply enterprises shall report accidents to the competent authorities within 24 hours of occurrence, otherwise they will be fined," the module can automatically identify "reporting within 24 hours of occurrence" as a compliance requirement and "failure to do so will result in a fine" as a consequence of the violation.

[0067] After the semantic recognition module completes the parsing of the regulatory text, the risk factor extraction module further analyzes the key risk points in the regulations. This module combines the operational characteristics of power supply companies to establish a set of compliance risk factor models, which include time sensitivity factors, business process impact factors, financial risk factors, and personnel management factors. For example, in regulatory requirements, if a business needs to be completed within a specific time but is not completed, it may be considered a compliance risk; if a regulation involves financial penalties, the impact may be high and therefore requires special attention. This module uses a combination of rule matching and machine learning to ensure that content related to compliance risks in regulatory texts can be accurately extracted while reducing misidentification. For historical violation record data, the module uses pattern recognition technology to extract past violation behavior patterns and compare them with regulatory requirements to identify potential loopholes in corporate compliance management.

[0068] After completing the risk factor extraction, the compliance risk label generation module classifies and labels the compliance risk points obtained through analysis. This module uses a preset compliance label system to map different regulatory requirements and violations to standardized risk categories. For example, if the regulations involve financial penalties, the regulation may be labeled as "financial risk"; if the regulations involve requirements for equipment maintenance cycles, it may be labeled as "operational risk." This module supports dynamic label adjustment and can automatically update the label classification system according to changes in regulations to ensure that the system is always consistent with the latest regulatory requirements. In addition, the module also supports a risk level assessment mechanism, which can dynamically calculate the severity of compliance risks based on the binding strength of regulatory provisions, the severity of penalties, and the company's past violations, to facilitate subsequent intelligent assessment and early warning.

[0069] Overall, the natural language processing unit 102 achieves efficient analysis of power supply industry regulations and historical violation records through functions such as text parsing, semantic recognition, risk factor extraction, and risk label generation, enabling the system to accurately understand regulatory requirements, identify potential compliance risks, and provide high-quality data support for subsequent intelligent assessment units.

[0070] Furthermore, the natural language processing unit is specifically configured to:

[0071] Using a multi-granularity domain word embedding model built specifically for grassroots power supply enterprises, dynamic word segmentation and word vector mapping are performed on policy and regulatory update data and historical violation record data to capture the implicit semantic relationship between different regulatory terms and corporate compliance terms; using the word vector mapping results, a semantic relationship network in the field of power supply enterprise compliance management is constructed through semantic similarity analysis to clarify the semantic association between key terms in different regulatory clauses and violation records; based on the semantic similarity and co-occurrence frequency between nodes in the semantic relationship network, cross-text risk factor association analysis is performed to form an accurate risk factor combination and annotate the corresponding compliance risk labels.

[0072] The natural language processing unit in this system undertakes the critical task of parsing and processing unstructured data, such as regulatory text and historical violation records. To ensure highly accurate compliance risk analysis for grassroots power supply companies, this unit relies on a specially constructed multi-granularity domain word embedding model. This model is constructed through four main phases: data collection, corpus preprocessing, model training, and iterative optimization. This ensures that it accurately captures the specialized terminology and semantic relationships involved in regulatory provisions, historical violation descriptions, and operational management for grassroots power supply companies.

[0073] During the data collection phase, the system first collected a large amount of textual data related to the power supply industry. This primarily includes national and local power regulatory policies, management regulations of power grid companies and grassroots power supply enterprises, operation and maintenance process manuals, internal audit reports, historical violation penalty announcements, and industry standards. These texts were sourced from government regulatory agencies, power enterprise management systems, public industry documents, and internal enterprise data storage systems. To ensure data integrity, the system deduplicated the texts, standardized their formats, and filtered out irrelevant data.

[0074] During the corpus preprocessing stage, the system uses rule-based and statistical methods to segment and clean the text. First, regular expressions and dictionary matching methods are used to remove meaningless symbols, special characters, and malformed text data. Then, word form normalization is performed to standardize the mapping of synonyms, industry abbreviations, and professional terms from different regions. For example, "distribution network", "distribution network", and "distribution network system" all refer to the same concept and are uniformly replaced with standard terms during the processing process. In addition, the system uses BERT-based named entity recognition (NER) technology to identify key entities in regulations, such as equipment categories, operation and maintenance personnel positions, and regulatory agency names, and provide basic data for subsequent semantic modeling.

[0075] During the model training phase, the system employs a hierarchical word embedding approach to capture the multi-granular semantic features of power supply companies. First, at the first level, the system uses traditional embedding methods such as Word2Vec to train a large amount of regulatory text to obtain basic word vectors. A co-occurrence-based statistical approach is then used to generate preliminary word vector representations. Secondly, at the second level, the system uses a pre-trained language model based on the Transformer architecture (such as BERT) for deep training to enhance its ability to capture dependencies in long texts and dynamically adjust the word vector representations based on context. For example, within regulatory text, the term "substation inspection" may refer to safety compliance, equipment management, or personnel responsibility under different provisions. The system automatically adjusts the word vector for "inspection" based on context, giving it different semantic representations under different regulatory provisions. Finally, at the third level, the system employs a contrastive learning-based approach to construct a semantic alignment model between regulatory text and historical violation records. This ensures that the model not only learns the standard expression of regulations but also understands the informal expression of violation records within the company. For example, in historical violation records, "delayed inspection time" may imply "failure to inspect on time," a regulatory requirement. The system can link these two through semantic matching.

[0076] During the model optimization and iteration phase, the system employs an active learning strategy. Whenever regulations are updated or new expressions appear in corporate compliance records, the system updates the model using incremental learning techniques to continuously optimize its semantic understanding capabilities. Furthermore, the system supports manual feedback from corporate users. If the system makes mismatches during compliance analysis, human reviewers can manually correct them. The system then uses this annotated data to further train the model, ensuring its long-term high accuracy.

[0077] After completing the training of the word embedding model, the system uses semantic similarity analysis methods based on the constructed word vectors to construct a semantic relationship network in the field of compliance management of power supply enterprises. In this network, each node represents a regulatory clause, a description of a violation, an equipment operating specification, or a management requirement, and the edge weights between different nodes represent the semantic similarity or co-occurrence frequency between them. The system uses a preliminary screening method based on cosine similarity to determine semantically related nodes, and further uses deep semantic matching (such as Siamese-BERT) to accurately calculate the semantic relationship. For example, if a certain regulatory clause requires that "substation operation and maintenance personnel must inspect once a week", the system can automatically identify that "the lack of inspection records in the operation and maintenance records" may constitute a violation and establish a strong association in the semantic network.

[0078] After constructing the semantic relationship network, the system uses a graph neural network to analyze the regulatory provisions, violation incidents, and management regulations within the network to identify potential correlations among high-risk factors. First, the system extracts node features and inputs the word vectors for each regulatory provision, along with statistical information on historical violation data and the degree of interdependence between regulations into the graph neural network. Second, the system uses a graph attention mechanism to calculate the correlation weights between regulatory provisions and historical violations, automatically filtering out the most influential risk factors. For example, if regulations require equipment maintenance to be no more than 60 days, and historical records show frequent delays in maintenance intervals for a certain type of equipment, the system will automatically identify a strong correlation between this equipment type and maintenance violations and further predict potential future compliance risks for this equipment type. Finally, the system uses an aggregation-based risk label generation method that combines the text content of regulatory provisions, historical violation patterns, and regulatory adjustment trends to generate accurate compliance risk labels for enterprises, enabling them to proactively identify potential compliance risks and take targeted measures.

[0079] This system uses a multi-level natural language processing approach. First, it trains a multi-granularity domain word embedding model using data specifically targeted at grassroots power supply companies to achieve accurate semantic parsing of regulations. Second, it constructs semantic associations between regulations and historical violation records through a semantic relationship network. It then uses graph neural networks for in-depth analysis to extract high-risk factor combinations. Finally, it automatically generates risk labels to provide companies with accurate compliance risk markers. This solution not only improves the accuracy of regulatory understanding but also enables in-depth analysis of the associations between regulations and corporate business operations, enabling the system to adapt to dynamic changes in regulations and provide grassroots power supply companies with more accurate and real-time compliance risk warning capabilities.

[0080] For example, inspection management is a critical component of the daily operations of grassroots power supply companies. Imagine a local power regulatory agency updates a regulation requiring power supply companies to ensure that key substation equipment undergoes a complete inspection at least once every 30 days and records the inspection results. Failure to do so will be considered a violation. Upon the release of this regulation, companies need to immediately adjust their operational strategies to ensure compliance. However, grassroots power supply companies' inspection records may be scattered across various systems, such as the operations and maintenance management system, dispatching system, and equipment logging system. Furthermore, different operations and maintenance personnel may use inconsistent recording methods, with some using "inspection completed," others simply recording "equipment operating normally," and still others recording "routine inspection" or "safety inspection." Using keyword matching alone makes it difficult to accurately determine which records truly comply with the new regulations and which may present compliance risks.

[0081] In this case, the system's natural language processing unit will first use a multi-granularity domain word embedding model to parse the terms of the new regulation and perform hierarchical semantic modeling on key concepts involved, such as "substation," "inspection," "record," and "30 days." The system will use domain word embedding technology to map terms such as "inspection," "patrol," and "inspection" to the same semantic space, and through contextual analysis, ensure that the potential association between "equipment operating normally" and "inspection completed" is identified. At the same time, the system will search for similar violation cases in the company's historical violation records, analyze past penalty data resulting from non-compliant inspections, and further optimize its understanding of regulatory requirements, so that the model not only focuses on surface text similarities, but also conducts risk prediction based on actual compliance experience.

[0082] When the regulations are parsed, the system will construct a semantic relationship network based on semantic similarity calculation and deep dependency analysis. In this network, the provisions of the new regulations are regarded as core nodes, while the company's historical inspection data, operation and maintenance records, compliance reports and other data are regarded as associated nodes. Suppose the system finds that the inspection records of some substations in the past year are incomplete or submitted late, and some of the records are described as "routine inspections" but do not clearly state whether the inspections are completed. The system will calculate the degree of match between these records and the new regulations through the semantic network and identify possible compliance risks. For example, if the inspection records of a substation frequently contain "equipment operating normally" but do not include specific inspection times, the system will determine that these records may not be sufficient to prove the inspection compliance required by the regulations, and will mark them as potential risk points.

[0083] Furthermore, the system will use graph neural networks to analyze the distribution of historical violation records, calculate which types of inspection omissions are more likely to lead to actual penalties, and optimize the risk factor weights based on this data. For example, if historical violation data shows that certain types of equipment (such as high-voltage switchgear) have a higher frequency of penalties for irregular inspections, the system will automatically increase the compliance risk weight of this type of equipment, giving it priority attention in subsequent risk assessments. In addition, the system will also include the historical inspection situation of different responsible personnel in the analysis. If an operation and maintenance personnel has frequently submitted incomplete inspection records in the past, the system may predict that the person's future compliance risk is higher and generate personalized compliance recommendations on the enterprise management terminal, such as increasing training or strengthening the review of inspection reports.

[0084] Ultimately, based on all analysis results, the system generates a compliance risk tag for the regulation and sends detailed risk warnings to the enterprise. For example, on the compliance management terminal, the system might prompt, "Substation A's inspection records may not comply with the new regulatory requirements. Within the last 30 days, there are only records of 'equipment operating normally,' but no complete inspection reports." Furthermore, the system automatically generates corrective action suggestions, such as, "It is recommended to add inspection timestamps and ensure that all inspection records clearly include a list of inspected equipment and descriptions of any anomalies." As the system continues to operate, it automatically monitors changes in regulatory requirements and adaptively adjusts analysis strategies, enabling enterprises to dynamically adjust operational compliance and ensure efficient compliance management even as regulations change.

[0085] Furthermore, the word vector mapping process of the natural language processing unit includes:

[0086] Based on the node weights of key terms in the constructed semantic relationship network, the semantic relevance parameters of the word vector mapping are dynamically adjusted to reflect the semantic change trend of regulatory clause updates while generating semantic similarity; combined with the context of the newly added regulatory clauses, the word vector embedding representation is automatically updated, and through an adaptive iterative process, the word vector is kept consistent with the original semantic network, thereby enhancing the accuracy of the association analysis of cross-text risk factors.

[0087] The system's natural language processing unit plays a key role in parsing regulatory text and analyzing historical violation records. A key component of this process is the dynamic adjustment of word vector mapping to capture subtle shifts in the semantics of regulatory provisions and improve the accuracy of cross-text risk factor analysis. In practice, this process primarily involves calculating node weights in the semantic relationship network, dynamically adjusting word vector mapping parameters, and updating word vectors based on regulatory context.

[0088] When constructing a semantic relationship network, the system uses key terms appearing in regulatory texts and violation records as nodes, and the semantic relevance and historical co-occurrence frequency between different terms are represented by edges. In this network structure, the weight of each node is determined not only by its frequency in the regulatory text but also by its importance in historical violation records. For example, if the term "equipment inspection" frequently appears in violation records involving penalties, its node weight will be higher than the term "operation log," which only occasionally appears in general regulatory clauses. In addition, the system uses a graph attention mechanism to dynamically calculate the relevance of each node, ensuring that the semantic connections in the network reflect real business scenarios and regulatory priorities.

[0089] Based on the above-mentioned semantic relationship network, the system dynamically adjusts the semantic relevance parameters when performing word vector mapping. Traditional word embedding methods usually generate a fixed vector for each term, but this approach makes it difficult to reflect the characteristics of regulatory semantics that change over time and context. To address this problem, the system introduces a dynamic weight adjustment mechanism when generating word vectors, that is, according to the weights of the nodes in the semantic relationship network and the correlation of the edges, the parameters of the word vector generation model are adjusted in real time. For example, when a new regulatory clause contains a similar but stricter expression than an existing clause, the system will automatically detect the enhanced semantic association, thereby shortening the distance between related terms in the word vector space. As a result of this dynamic adjustment, the word vector can not only maintain an accurate representation of the existing regulatory semantics, but also reflect the trend of regulatory semantic changes, providing a more realistic semantic basis for subsequent cross-text analysis.

[0090] When regulations are updated or new clauses are added, the system triggers an automatic update process for word vectors. New regulatory clauses often contain new terms, different expressions, or completely new business rules. In order to adapt to these changes, the system first performs contextual semantic analysis on the new clauses through a pre-trained language model (such as BERT) to extract new semantic features. Subsequently, combined with the existing semantic relationship network, the system performs an adaptive iteration to update the weights of relevant nodes and the correlation of edges in the network. Through this iterative process, the semantic information of the new clauses is smoothly integrated into the original semantic structure without the need to completely rebuild the entire semantic network. This not only improves efficiency, but also ensures the coordination of the word vectors of the new clauses with the existing semantic structure.

[0091] During the adaptive update process, the system also verifies the effectiveness of the newly added word vectors to ensure that they perform better than their initial state in risk factor association analysis. Specifically, the system verifies the changes in semantic similarity between a set of representative regulatory provisions and historical violation records, adjusts the regularization parameters of the word vectors, and ultimately generates word vector representations that provide optimal support for cross-text association analysis. This optimization process ensures that after new regulatory provisions are added to the semantic network, they will not cause analytical errors due to semantic drift, nor will they reduce the accuracy of association analysis due to insufficient adaptation to the new context.

[0092] In summary, the system's natural language processing unit successfully tracks and updates the semantics of regulatory clauses by dynamically adjusting word embedding parameters, combined with weight calculation and edge-association evaluation within the semantic relationship network. Whenever regulations change or new clauses are added, the system adaptively updates the word embedding representation, ensuring the accuracy and consistency of cross-text risk factor analysis. This provides more reliable and accurate technical support for intelligent early warning of compliance risks for grassroots power supply companies.

[0093] The risk feature construction unit 103 is used to perform multi-dimensional feature fusion on the business operation data and the risk label data processed by the natural language processing unit to form a feature matrix for risk identification.

[0094] The risk feature construction unit 103 is responsible for the core data integration and feature extraction functions in the entire system. Its main task is to perform multi-dimensional feature fusion on the business operation data from the data collection unit, the historical violation record data, and the regulatory update information parsed by the natural language processing unit, and form a feature matrix for compliance risk identification, thereby providing data support for subsequent intelligent risk assessment.

[0095] The unit first receives business operation data from the data acquisition unit, including the operating status of power supply equipment, grid load conditions, customer complaint records, operation and maintenance personnel operation logs, etc. These data are usually stored in a structured or semi-structured form and organized according to time series or business events. To ensure data consistency, the unit has built-in data alignment and time window sliding mechanisms, which can align the timestamps of different data sources to avoid analysis errors caused by data asynchrony. In addition, for historical violation records, the unit will combine key information such as the time of occurrence of the violation, the scope of impact, related equipment and personnel, and map them to a standardized violation pattern database so that similar risk scenarios can be identified in subsequent analysis.

[0096] For regulatory update data, this unit receives regulatory text elements parsed by the natural language processing unit, including key obligations, prohibitions, penalties, and applicable scenarios, and compares them with the company's business data. For example, if new regulations require substation operators to submit safety reports within 48 hours after equipment maintenance, and this unit finds from business operation data that most maintenance reports are submitted after 48 hours, it will automatically mark the company's compliance risk point in the feature matrix, ensuring that subsequent intelligent risk assessments can identify this issue.

[0097] On the basis of data integration, the unit further performs multi-dimensional feature extraction and fusion to generate a feature matrix that can accurately reflect the compliance status of the enterprise. To this end, the unit adopts a rule-based feature construction method combined with a machine learning-driven automatic feature extraction method to ensure that it can not only utilize the compliance rules set by expert experience, but also mine potential risk patterns from massive data. The rule construction part mainly predefines a series of risk features based on industry standards and regulatory requirements, such as "equipment maintenance frequency is lower than the specified value", "customer complaint rate exceeds the industry average", "critical equipment load exceeds the standard for a long time", etc. These regularized features can be directly used to construct the basic columns of the feature matrix to ensure that the system has strong interpretability.

[0098] The machine learning-driven automatic feature extraction component uses statistical learning and deep learning models based on a company's historical violation data and related business data to extract implicit features that may affect compliance. For example, this unit can employ principal component analysis or autoencoder technology to automatically reduce the redundancy of high-dimensional data and extract key variables that can effectively distinguish between compliance and non-compliance status. Furthermore, for time series data, this unit can combine time window analysis, sliding mean calculation, and trend analysis methods to extract characteristics of data that change over time, thereby identifying compliance risks with lags or cumulative effects.

[0099] The constructed feature matrix includes data from multiple dimensions. Each row represents a specific business instance (such as an operation and maintenance task, the status of a piece of power equipment, or a customer complaint record), and each column represents a compliance-related feature, such as time dimension features (such as submission deadlines and fault intervals), business indicator features (such as equipment operating hours and inspection frequency), historical behavior features (such as the number of past corporate violations and rectification efficiency), and regulatory compatibility features (such as the degree of deviation between the criticality of regulatory requirements and corporate behavior). After the feature matrix is ​​constructed, the unit performs normalization to ensure that different features have the same numerical scale, thereby improving the stability of the model calculation.

[0100] To further enhance the system's adaptability, the unit also integrates an adaptive feature selection mechanism that automatically adjusts the weights of the feature matrix based on an enterprise's historical violation patterns. For example, if an enterprise has a high violation rate regarding equipment maintenance compliance, the unit will dynamically increase the weights of features related to equipment maintenance, allowing the enterprise's compliance risk assessment to focus more closely on its key issues.

[0101] Furthermore, the risk feature construction unit is specifically used to:

[0102] Based on the business categories involved in regulatory requirements, business operation data is divided into equipment management data, operation and maintenance scheduling data, and financial compliance data. Each type of data is further subdivided based on the constraints of regulatory provisions. Equipment management data is classified according to inspection frequency, maintenance record integrity, and equipment load status. Operation and maintenance scheduling data is classified according to scheduling response time, task completion status, and personnel deployment. Financial compliance data is classified according to capital flow records, contract execution status, and expenditure approval process. When risk tag data is matched with business data, the corresponding data is extracted from the business data based on the numerical restrictions, time requirements, and operating specifications of regulatory provisions. fields and calculates their compliance with regulatory requirements. For regulatory clauses containing time limits, the time interval of the data fields is calculated, and data that exceeds the regulatory limit is marked as anomalies. For regulatory clauses containing operational process requirements, the task completion status of the business data is calculated, and unfinished or timed-out data is marked as anomalies. When forming the feature matrix, based on the time series analysis method, the time interval difference, historical violation frequency and cumulative violation days are extracted, and the trend characteristics of historical data are added to the feature matrix, so that the system can identify the cumulative effect of long-term violation risks and provide a basis for evaluating regulatory compliance within different time windows.

[0103] The Risk Signature Construction Unit within this system performs the core function of deeply analyzing and integrating power supply enterprise operational data with regulatory requirements to ensure accurate and adaptable risk assessments. This unit's implementation involves three primary phases: data classification, regulatory matching, and signature matrix construction. Each phase incorporates the specific business scenarios of grassroots power supply enterprises to ensure accurate identification of compliance risks.

[0104] During the data classification phase, the unit first divides business operation data into equipment management data, operation and maintenance scheduling data, and financial compliance data based on the scope of application of regulatory provisions and the company's operating business, and further refines each type of data based on the specific constraints of regulatory provisions. Equipment management data mainly includes equipment inspection, maintenance, and operating status, and is specifically divided into three categories: inspection frequency, maintenance record completeness, and equipment load status. Inspection frequency involves regular inspection records of equipment, and is further subdivided according to factors such as the completion of inspection tasks, the completeness of inspection result reports, and the qualifications of inspection execution personnel. Maintenance record completeness assesses the standardization of the equipment maintenance process, including a comparison of equipment status before and after maintenance, the level of detail in maintenance record filling, and the timeliness of maintenance task completion. Equipment load status involves the long-term operating data of the equipment, including load fluctuations, the duration and number of overload operations, and whether there are records of abnormal shutdowns due to overload. Operations and maintenance scheduling data is categorized based on the execution of scheduling tasks, primarily including scheduling response time, task completion status, and personnel deployment. Scheduling response time measures the time interval from task assignment to execution, task completion status assesses the compliance of scheduling task execution, and personnel deployment status addresses the qualifications of operations and maintenance personnel, the rationality of task assignments, and the degree of compliance with actual execution. Financial compliance data is categorized across cash flow records, contract execution status, and expenditure approval processes. Cash flow records are used to monitor compliance with corporate fund payment pathways. Contract execution status includes contract payments, contract fulfillment progress, and default risks. The expenditure approval process assesses corporate compliance in budget approvals, ensuring that all expenditures adhere to established financial rules.

[0105] During the regulation matching phase, the unit extracts relevant fields from business data based on the constraints in the regulatory clauses and calculates the data's compliance. For regulations with numerical constraints, such as requiring equipment inspection intervals to exceed a specified number of days, the unit extracts the inspection time field and calculates whether the actual interval exceeds the regulatory threshold. For regulations with time requirements, such as requiring scheduled tasks to be completed within a specified time limit, the unit calculates the task execution interval and marks any tasks that have exceeded the specified time limit. For regulations with operational process constraints, such as requiring a complete report to be generated after equipment repairs, the unit analyzes the completeness of maintenance records and marks any tasks that are missing key records. During this process, the unit also incorporates a dynamic matching mechanism for regulatory requirements. When regulatory adjustments result in changes in standards, the system automatically identifies the differences between the old and new regulations and adjusts the data filtering criteria. For example, if a regulatory update changes the inspection cycle from 30 days to 25 days, the unit recalculates the intervals for all inspection records and adjusts the compliance judgment based on the new standard.

[0106] During the feature matrix construction phase, this unit uses time series analysis to extract long-term trend features, enabling the system to identify the cumulative effects of compliance risks. First, the unit calculates the time interval difference—the deviation between the regulatory time limit and the actual execution time—as a key feature for measuring compliance. Second, the unit calculates the frequency of historical violations to assess whether an enterprise has a history of persistent noncompliance. For example, if a device has repeatedly exceeded its inspection deadline over the past year, its risk score will increase accordingly. Finally, the unit calculates the cumulative number of days of noncompliance—the number of days within the regulatory time window—to measure compliance trends. For example, if a maintenance task is consistently not completed within the regulatory timeframe, the task's risk level will increase. In the feature matrix, this unit not only records individual violations but also incorporates historical trend features, enabling the system to predict an enterprise's future compliance risks and providing a basis for assessing long-term noncompliance risks.

[0107] Through the above-mentioned multi-level data classification, regulation matching and time series analysis, the unit can effectively integrate business data and regulatory requirements and form a refined feature matrix, enabling the system to accurately identify and predict the compliance risks of power supply enterprises, thereby improving the risk warning capabilities of grassroots power supply enterprises.

[0108] Furthermore, when calculating the time interval of the data field, the risk feature construction unit dynamically adjusts the selection range of the time window based on the timeliness requirements of the regulatory provisions and the historical operating model of the enterprise; among them, for the case where the regulatory requirements clearly stipulate a fixed time period, the time threshold stipulated by the regulatory provisions is directly used for calculation, and for the case where the regulations do not clearly give a fixed time period but there are industry practices or internal management standards of the enterprise, the most common time interval is calculated based on historical business data, and the time interval is used as a benchmark for compliance judgment; in the case of regulatory updates or business process adjustments, the time window change trends of the new and old regulatory requirements are compared, and the time interval calculation rules are dynamically adjusted to ensure that the feature matrix can adapt to regulatory changes and improve the accuracy of compliance risk assessment.

[0109] When calculating the time interval of a data field, the risk feature construction unit needs to ensure that the time window setting not only meets the requirements of the regulatory provisions, but also adapts to the actual operating model of the enterprise to improve the accuracy of the compliance risk assessment. During the time interval calculation process, the unit first determines the initial calculation method based on the timeliness requirements of the regulatory provisions. If the regulations clearly stipulate a fixed time period, such as a 30-day inspection cycle or a 7-day financial approval cycle, the system directly uses the time threshold specified in the regulations for calculation, and matches the timestamps of the corresponding fields in the business data to calculate the deviation between the actual execution interval and the regulatory requirements. If the time interval of a business exceeds the regulatory threshold, the system will mark the data as an anomaly and add a violation time deviation field to the feature matrix so that the subsequent risk assessment unit can identify high-risk business links.

[0110] When regulations do not explicitly stipulate a fixed time period, but there are industry practices or internal corporate management standards, the unit will conduct statistical analysis based on historical business data to calculate the most common execution time intervals for this type of business. For example, if a regulation only requires "regular inspections of equipment" but does not specify the inspection cycle, the system will analyze the company's internal inspection records and calculate the average inspection cycle of this business over the past period of time, and select the time interval with the highest frequency as a reference standard. If the company's actual business execution interval far exceeds the historical average, the system will mark the business as a potential compliance risk and add abnormal fluctuation features to the feature matrix to improve the sensitivity of compliance risk detection. Through this data-driven approach, the system can still provide a reasonable time interval benchmark even when the regulations do not clearly specify time requirements, thereby improving its adaptability to corporate operational data.

[0111] In the event of regulatory updates or business process adjustments, this unit needs to compare the changing trends in the time windows required by the new and old regulations and dynamically adjust the time interval calculation rules to adapt to the new compliance standards. When a regulatory adjustment results in a change in time requirements, such as shortening the inspection cycle from 30 days to 25 days, the unit automatically detects the regulatory update and recalculates the time intervals for all relevant business data, making the new regulatory time requirements effective immediately. At the same time, the unit also analyzes data trends before and after the regulatory adjustment to determine which businesses have become new compliance risk points due to the regulatory change. For example, if, after a regulatory update, certain business operations of an enterprise are still executed according to the old standards, the system will specifically annotate the execution data of these businesses and add regulatory transition period features to the feature matrix to enable a more accurate compliance risk assessment of the enterprise in the early stages of the regulation's implementation. In this way, the unit not only ensures that the system can quickly adapt to regulatory adjustments but also provides transition period monitoring during regulatory changes, improving the enterprise's compliance adaptability.

[0112] The unit's dynamic time interval calculation method enables the system to find the optimal balance between regulatory requirements, industry practices and the company's actual operating model, thereby improving the accuracy of compliance judgments for time-sensitive regulations and ensuring that the feature matrix can comprehensively and accurately reflect the company's compliance risk status.

[0113] The intelligent risk assessment unit 104 is used to use a pre-trained machine learning model to perform real-time compliance risk identification on the feature matrix, determine the category and level of the risk, and generate corresponding warning information based on the risk category and level.

[0114] The intelligent risk assessment unit 104 performs core analysis and decision support tasks throughout the system. Its primary function is to analyze the feature matrix generated by the risk feature construction unit in real time, based on a pre-trained machine learning model. It then identifies potential compliance risks, determines the risk category and level, and generates corresponding early warning information based on the risk category and level. This unit not only needs to efficiently process large amounts of historical and real-time business data but also possesses adaptive learning capabilities to continuously optimize the accuracy of risk identification and improve the effectiveness of early warnings.

[0115] In the first step of risk assessment, the unit receives the feature matrix from the risk feature construction unit and inputs it into a pre-trained machine learning model for analysis. The core of this machine learning model can be based on a variety of algorithms, such as support vector machines (SVMs), random forests, gradient boosting decision trees, or deep learning neural networks. The choice of which algorithm depends on the specific needs and data characteristics of the power supply company. For example, if the company's historical violation data is relatively complete, supervised learning methods can be used for classification prediction. If the violation pattern is more complex and has a certain degree of uncertainty, semi-supervised learning or unsupervised learning methods, such as cluster analysis and autoencoders, can be introduced to discover potential risk patterns. Regardless of the model used, the unit needs to perform feature importance assessment to ensure that key variables can fully play a role in risk prediction. For example, if the regulatory adaptability feature has a greater impact on the violation, the weight of this feature in the model should be increased accordingly to improve the accuracy of the model's prediction.

[0116] Another key function of this unit is the determination of risk categories and levels. During the assessment process, the system will classify risks into multiple levels based on different feature inputs, such as low risk, medium risk, high risk and serious risk, and further subdivide them into categories such as financial risk, operational risk, equipment management risk, and human resources compliance risk. For example, if a company's equipment maintenance interval exceeds the threshold required by regulations and has been punished for similar problems in the past, the system will determine it as a high-risk event related to equipment management. If it is only an occasional small overdue amount and the company has a good compliance record in the past, it may be classified as a low-risk event. This classification can not only help corporate managers quickly understand the nature of the risk, but also facilitate subsequent disposal and decision-making.

[0117] After the risk identification is completed, the unit will generate early warning information and push it to the company's compliance management terminal. Early warning information usually includes risk categories, risk levels, possible violation clauses, scope of impact, recommended corrective measures, etc. For example, if a compliance problem is detected in a certain operation and maintenance process, the system will send an early warning to the relevant responsible person, reminding him of the relevant regulatory requirements and providing rectification suggestions for similar historical cases. In addition, the unit supports a graded early warning mechanism, that is, different response strategies are adopted for different levels of risks. For example, low-risk events may only be notified to relevant departments through internal reminders, while high-risk or serious risk events may trigger automatic rectification suggestions, or even issue compliance risk alerts to senior corporate executives or regulatory agencies.

[0118] In addition, to further improve the accuracy of risk prediction, the unit has adaptive learning capabilities, that is, based on the differences between historical warning results and actual violation handling results, it continuously optimizes model parameters. Specifically, the system will compare the actual situation after the warning is generated, such as whether a warning ultimately leads to a violation penalty, or whether the risk is eliminated after rectification. If the system finds that the prediction deviation of certain risks is large, it will automatically adjust the feature weights or retrain the model to enhance the adaptability of the model. For example, if the system finds that updates to certain policies and regulations often lead to increased corporate compliance risks, then when similar regulations are updated in the future, the unit will automatically assign them a higher risk sensitivity to ensure the accuracy and timeliness of the warning.

[0119] In terms of system architecture, this unit utilizes a parallel computing framework to improve the computational efficiency of risk assessment. Given the massive amount of data generated by grassroots power supply companies, this unit can leverage distributed computing platforms such as Hadoop to enable parallel processing of large-scale data. For real-time data streams, this unit can be combined with streaming computing frameworks such as Kafka to achieve low-latency risk detection, ensuring that early warning information reaches relevant personnel before or immediately after a risk occurs.

[0120] The design of the intelligent risk assessment unit 104 ensures that the system can efficiently identify the compliance risks of grassroots power supply enterprises, and provide accurate risk assessment and early warning by combining historical violation data, business operation data and regulatory change information, helping enterprises optimize compliance management strategies, reduce violation costs, and improve operational safety.

[0121] Furthermore, the machine learning model of the intelligent risk assessment unit includes a compliance feature extraction network, a risk classification network, and a dynamic adaptive adjustment module;

[0122] The input of the compliance feature extraction network is the time dimension features, business indicator features, historical behavior features, and regulatory adaptability features in the feature matrix. It is used to perform time series expansion of the time dimension features based on the hierarchical feature decomposition method, extract task interval deviations and regulatory execution time limit compliance, calculate the time decay weights of key business data, and perform normalization processing on the business indicator features so that the equipment operation status, inspection frequency, and task completion rate features are calculated under a unified numerical scale.

[0123] The input of the risk classification network is the normalized compliance feature vector generated by the compliance feature extraction network. Based on the regulation matching rules and classification enhancement mechanism, the network identifies the compliance risk category of the input features, and combines the number of violations and rectification efficiency information in the historical behavior characteristics to calculate the risk level and generate a risk category label.

[0124] The input of the dynamic adaptive adjustment module includes regulatory update information and historical prediction deviation data of the risk classification network. The module parses the regulatory adjustment content based on the regulatory change detection mechanism, and combines the regulatory adaptability characteristics to calculate the degree of deviation between the criticality of regulatory requirements and corporate behavior, and dynamically adjusts the classification weights and risk level division thresholds of the risk classification network.

[0125] The system's intelligent risk assessment unit's machine learning model analyzes power companies' business data, regulatory requirements, and historical violation records to identify potential compliance risks and generate early warnings. This unit comprises a compliance feature extraction network, a risk classification network, and a dynamic adaptive adjustment module. Each component has clear inputs, processing flows, and outputs, working together to ensure accurate and adaptable risk identification.

[0126] The compliance feature extraction network takes as input the time dimension features, business indicator features, historical behavior features, and regulatory compliance features from the feature matrix. The network deeply processes different data categories using a hierarchical feature decomposition method and outputs a structured compliance feature vector. To process the time dimension features, the network first performs time series expansion, dynamically analyzing the execution times of key tasks in the business data. It calculates the gap between the actual task completion time and the regulatory requirement, and generates a task interval deviation feature. The network then calculates compliance with regulatory deadlines by matching the time thresholds specified in the regulation. For example, if the regulation requires that inspection intervals must not exceed 30 days, the network analyzes all inspection data, calculates whether the actual inspection intervals meet the requirement, and generates a deadline compliance score. Furthermore, the network incorporates a time-decay weighting method, assigning lower weights to older business events and higher weights to more recent events, thereby enhancing the system's sensitivity to real-time compliance risks. For business indicator features, the network first performs normalization to convert indicators such as device operating status, inspection frequency, and task completion rate to the same numerical scale, thus avoiding calculation errors caused by differences in feature numerical ranges. The network then uses a hierarchical clustering approach to group devices or business instances with similar operating modes and calculate their compliance risk scores, ensuring that the feature representation more accurately reflects the company's actual operations. The network output is a normalized compliance feature vector, which includes multiple risk features such as task interval deviation, compliance with regulatory execution deadlines, time decay weight adjustment value, device operating status score, inspection frequency score, and task completion rate score.

[0127] The risk classification network takes as input the compliance feature vector generated by the compliance feature extraction network. Based on regulation matching rules and a classification enhancement mechanism, the network identifies compliance risks from the input features and outputs specific risk categories and risk levels. First, the network analyzes the key constraints of the regulatory provisions based on the regulation matching rules and maps them to the dimensions of the compliance feature vector. For example, if the regulation requires that "substation equipment overload must not exceed 80%," the network identifies the overload rate of a specific device in the equipment operating status score and calculates a violation probability score based on the regulatory constraint. Subsequently, the network, combined with the classification enhancement mechanism, analyzes the company's past performance under the same compliance risk by calculating historical behavioral features such as the number of violations and rectification efficiency. For example, if a piece of equipment at an enterprise has been recorded for multiple overload violations within the past year, the network will automatically increase the risk weight of that device during the classification process, making it more likely to be classified as high-risk. Furthermore, the network incorporates a time-window hierarchical classification approach, which categorizes compliance risks across different time windows. For example, short-term violations may be classified as low-risk events, while long-term violations may be classified as high-risk events. The network's output includes the company's compliance risk category labels, such as "low risk," "medium risk," and "high risk," as well as specific risk scores, and generates a structured risk report containing risk categories, violation probabilities, and rectification recommendations.

[0128] The dynamic adaptive adjustment module, which takes as input regulatory update information and historical prediction deviation data from the risk classification network, analyzes regulatory changes and dynamically optimizes the classification weights and risk level thresholds within the risk classification network. First, based on a regulatory change detection mechanism, the module automatically analyzes regulatory changes, extracts new or changed regulatory provisions, and identifies the business categories affected. For example, if a regulatory update causes the inspection cycle to be adjusted from 30 days to 25 days, the module automatically identifies this change and transmits the new inspection requirements to the risk classification network, ensuring timely updates to the classification standards. Second, the module calculates regulatory fit features to measure the degree of alignment between an enterprise's business behavior and regulatory requirements. It also dynamically adjusts the parameters of the risk classification network based on the enterprise's historical violation trends. For example, if an enterprise had a tendency to proactively rectify issues before a regulatory update, the system can lower its initial violation risk score to avoid false positives caused by regulatory adjustments. Furthermore, the module evaluates the classification accuracy of the risk classification network based on historical prediction deviation data, comparing past predictions with actual violations. Based on this feedback, the module optimizes the classification thresholds, ensuring that risk predictions are more tailored to the enterprise's actual situation. The output of this module is the updated classification weights, risk level adjustment parameters and regulatory matching optimization strategy, and it automatically adjusts the risk classification network to enable it to adapt to regulatory changes and improve the stability and accuracy of long-term predictions.

[0129] Through the above-mentioned multi-level data processing process, the intelligent risk assessment unit can accurately identify the compliance risks of enterprises based on regulatory provisions, historical behavioral data and real-time business data, and adaptively optimize classification strategies when regulations are adjusted, so that the system can maintain efficient and accurate compliance risk identification capabilities in the long term.

[0130] The following is the reference implementation code of the machine learning model:

[0131]

[0132]

[0133]

[0134]

[0135] Furthermore, when calculating the risk level, the risk classification network of the intelligent risk assessment unit further dynamically adjusts the weight of the compliance feature vector based on the time decay weighted prediction model to optimize the accuracy of long-term violation risk identification.

[0136] Among them, for the violation risk of a specific business instance i at time t, the comprehensive risk score R is defined i (t) is a nonlinear function of multiple factors such as time, historical violation frequency, regulatory adjustment weight, and rectification efficiency. It is calculated using the following formula 1:

[0137]

[0138] α, β, and γ are the business indicator characteristic influencing factors, historical behavior characteristic influencing factors, and regulatory adjustment influencing factors, respectively. Their recommended values ​​can be set according to the characteristic importance of the data. For example, when regulations change frequently, α=0.5, β=0.3, and γ=0.2 can be adjusted to α=0.3, β=0.3, and γ=0.4 to increase the impact weight of regulatory adjustments.

[0139] F j (X i ) represents the feature matrix X i The normalized value of the jth business indicator feature or time dimension feature in , including submission time limit, fault interval, equipment operating hours, inspection frequency, etc.; the normalization calculation method adopts the minimum-maximum normalization method:

[0140]

[0141] Among them, X min and X max are the minimum and maximum values ​​of the feature in the historical data respectively.

[0142] G k (H i ) represents the feature matrix H i The normalized value of the kth historical behavior feature in , including the number of past violations and rectification efficiency of the enterprise; these data are calculated based on the historical compliance data of the enterprise. The rectification efficiency can be expressed as the ratio of the number of successful rectifications of historical violations to the total number of violations, that is:

[0143]

[0144] w j (t) is the time decay factor of the j-th feature, which is defined as the following formula 2:

[0145]

[0146] Where μ is the time attenuation coefficient, and the recommended value is 0.01. t represents time.

[0147] t0 is the timestamp of the last violation of the feature, ensuring that the impact of earlier violations gradually decays.

[0148] λ k (t) is the dynamic adjustment weight of historical behavior characteristics, which is defined as the following formula 3:

[0149]

[0150] Among them, v is the adjustment parameter, and the recommended value is 0.1.

[0151] σ k is the value of the kth historical behavior feature, such as the cumulative value of the number of past violations.

[0152] σ0 is the baseline violation impact threshold, used to measure the extent of an enterprise's deviation from regulatory standards, ensuring that risk assessments are adaptable to the historical violation trends of different enterprises. It is set based on expert knowledge or calculated based on empirical data. For example, if σ0 = 5, then when the number of violations by an enterprise is less than 5, the calculated result is close to 0.5. When the number of violations exceeds 5, the result rapidly approaches 1, increasing the impact of historical violations on the current risk assessment.

[0153] ΔQ i (t) is the incremental risk caused by regulatory adjustments to business instance i, calculated as follows:

[0154]

[0155] Where P represents the total number of regulatory provisions adjusted; ξ m (t) represents the applicability score of the mth regulatory clause to the business instance at time t; ξm (t-1) represents the applicability score of the mth regulatory clause to the business instance at time t-1. The applicability score calculation process integrates the requirements of the regulatory clause, the degree of compliance with the business data, the effective date of the regulation, and the impact of internal corporate rules. First, the system parses the regulatory clause and extracts key requirements, such as equipment inspection cycles, approval deadlines, or load rate caps. These requirements are then compared with actual business data to determine the regulatory applicability to the business instance.

[0156] The applicability score is calculated through a weighted comprehensive approach, taking into account three key factors: compliance (weighted 0.5), timeliness (weighted 0.3), and internal rule adjustments (weighted 0.2). The system first calculates a score for each factor separately: compliance is scored based on the degree of data deviation (1 for full compliance, 0.2 for severe deviation); timeliness is assessed based on the time since the regulation came into effect; and internal rules are adjusted based on the stringency of corporate standards. Finally, the system multiplies these three scores by their respective weights and adds them together to form a final applicability score, which comprehensively reflects the applicability of the regulation to a specific business case.

[0157] η m is the weight coefficient of this clause. When the regulatory adjustment leads to a large change in the applicability score, the value increases, indicating that the impact of the regulatory adjustment on the risk assessment is enhanced. The recommended value is 0.5.

[0158] Finally, the risk level L i Based on the comprehensive risk score R i (t) is graded, with index i representing a specific business instance. That is, during the risk assessment process, the system independently calculates the compliance risk level for each business unit (e.g., an inspection task, the operating status of a piece of equipment, a financial approval process, etc.). In other words, i corresponds to a row in the feature matrix and represents the compliance risk level of the business instance corresponding to that row's data.

[0159] For example, if i = 1 represents the inspection task of substation A on May 1, 2025, then L1 represents the compliance risk level of the inspection task;

[0160] If i = 2 represents the monthly maintenance record of a certain power equipment, then L2 represents the compliance risk level of the equipment maintenance task;

[0161] If i=3 represents a customer complaint record, then L3 reflects the compliance risk level of the complaint handling process.

[0162] The specific rules are shown in Formula 5 below:

[0163]

[0164] θ1, θ2, and θ3 are risk level thresholds, which are updated based on the company's historical violation trends, regulatory changes, and the data distribution in the feature matrix to improve the adaptability and accuracy of risk assessment. For example, the initial values ​​of these thresholds can be set to θ1 = 0.2, θ2 = 0.5, and θ3 = 0.8.

[0165] Here, low risk (L i =1): The business instance basically complies with regulatory requirements and has no tendency to violate regulations in the short term; medium-low risk (L i =2): Some indicators are close to the violation limit, but have not yet exceeded the compliance range, and adjustments are needed; medium-high risk (L i =3): Some compliance features have deviated from regulatory requirements. If this trend continues, it may lead to non-compliance; high risk (L i =4): The business instance has seriously deviated from regulatory requirements and requires immediate rectification.

[0166] Furthermore, the compliance risk early warning system for grassroots power supply enterprises based on big data analysis also includes:

[0167] The early warning and risk avoidance suggestion unit is connected to the intelligent risk assessment unit and is used to automatically generate targeted risk avoidance suggestions based on the identified risk categories and levels, and push the early warning information and avoidance suggestions to the compliance management terminal of the grassroots power supply enterprise to guide the enterprise to actively implement risk prevention measures.

[0168] The early warning and risk avoidance suggestion unit plays a key role in decision-making support and proactive prevention in the entire system. Its main function is to automatically generate corresponding risk avoidance suggestions based on the risk categories and levels identified by the intelligent risk assessment unit, and to promptly push early warning information and avoidance measures to the compliance management terminals of grassroots power supply enterprises, enabling enterprises to take effective measures before or at an early stage of risk occurrence, reduce the possibility of violations, and improve the initiative and accuracy of compliance management.

[0169] After the early warning information is generated, the unit will first analyze the risk categories, levels and related influencing factors provided by the intelligent risk assessment unit, and build targeted avoidance suggestions based on the company's historical violation data, business processes and regulatory requirements. To ensure the accuracy of the suggestions, the unit uses a combination of rule-based and machine learning methods for processing. In terms of rule matching, the unit has a built-in compliance knowledge base that stores regulatory requirements, standardized risk avoidance plans and best practices for the power industry. For example, for a certain type of equipment maintenance risk, the knowledge base may contain multiple feasible solutions such as "strengthening the frequency of daily inspections", "adopting an intelligent monitoring system", and "adjusting maintenance personnel arrangements". The unit will automatically select the best avoidance measures based on the specific circumstances of the enterprise.

[0170] In order to improve the personalization of risk avoidance recommendations, the unit also uses machine learning methods, combined with the company's historical compliance records and rectification effects for optimization. Through data mining, the unit can analyze the measures taken by the company in similar risk situations in the past and evaluate their effectiveness. For example, if a company violated regulations in the past due to delayed equipment maintenance, the unit will analyze the corrective measures taken at the time, such as whether the frequency of inspections was increased and whether the process was improved, and infer the most suitable avoidance plan for the current situation based on the company's compliance history. In addition, the unit also supports adaptive adjustment. If a certain avoidance suggestion is not effective in the subsequent execution process, the unit can automatically optimize the recommendation logic to ensure that recommendations in similar situations in the future are more accurate.

[0171] The risk avoidance recommendations generated by this unit not only include specific corrective measures but also provide detailed implementation guidance to ensure prompt implementation by the enterprise. For example, when pushing the suggestion to "increase the frequency of inspections," the unit will further provide recommended inspection intervals, the types of equipment involved, and the required personnel deployment suggestions. It will also analyze the actual effectiveness of inspections in reducing violations based on historical data. In addition, to enhance corporate compliance awareness, the unit can also provide regulatory interpretations, detailing the possible consequences of violations and relevant regulatory provisions, so that corporate managers can intuitively understand the importance of risks and thus improve the execution of compliance management.

[0172] In terms of pushing early warning information and avoidance suggestions, the unit adopts a multi-channel push mechanism to ensure that the information can be conveyed to the relevant responsible persons and management in a timely manner. For different levels of risks, the unit will adopt different notification strategies. For example, for low-risk warnings, they are only archived and notified to relevant personnel on the compliance management terminal, while for high-risk or serious risk events, the unit may push information at multiple levels through emails, text messages, internal corporate notification systems, etc., and even trigger emergency response mechanisms to ensure that the company can take quick measures. In addition, the unit supports hierarchical push based on permissions, that is, different risk types will be pushed to managers at different levels. For example, risks related to equipment maintenance will be directly notified to the operation and maintenance team, while risks involving financial compliance will be pushed to the financial director to ensure accurate information delivery and improve response efficiency.

[0173] Furthermore, the compliance risk early warning system for grassroots power supply enterprises based on big data analysis is characterized by further comprising:

[0174] A causal chain traceability analysis unit, connected to the intelligent risk assessment unit, is used to construct a multi-level causal reasoning map for tracing the cause of violations based on the time dimension characteristics, historical behavior characteristics, and regulatory adaptability characteristics in the feature matrix after the compliance risk category and level are determined, so as to identify the key causal paths and evolution patterns that cause the current compliance risk;

[0175] The causal chain traceability analysis unit receives compliance risk labels corresponding to high-risk levels, reversely traces the evolution of key features of relevant business instances in its feature matrix within the historical time window, extracts the time dimension feature change trend of the risk concentration segment, and combines the joint variation characteristics of the number of violations and rectification efficiency in the historical behavior characteristics, as well as the mutation points of the regulatory adaptability characteristics, to identify the multi-factor coupling relationship that leads to increased risk.

[0176] The identified causal paths are structured and a causal chain weight network is constructed based on the chronological order of indicator changes, the cumulative degree of violation history, and sensitive nodes of regulatory adaptability changes. The causal weights are then reallocated to the time dimension features and historical behavior features in the feature matrix.

[0177] Output risk tracing report based on traceable causal reasoning, which marks the dominant feature combination that leads to risk intensification, key time windows and their corresponding regulatory adaptability critical points, and is used to provide source rectification basis and precise prevention and control suggestions for compliance management terminals.

[0178] The present invention relates to a compliance risk early warning system for grassroots power supply enterprises based on big data analysis, which further includes a causal chain traceability analysis unit, which is used to construct a set of traceable and deducible multi-level causal reasoning maps based on the formed feature matrix content for the identified compliance risk categories and risk levels after completing the intelligent risk assessment, so as to clarify the causal path and inducement evolution process of compliance risks, thereby providing traceability reference and rectification decision support for the compliance governance of grassroots power supply enterprises.

[0179] The causal chain traceability analysis unit is communicatively connected to the intelligent risk assessment unit. When the intelligent risk assessment unit outputs the risk level result (such as high risk or serious risk) and matches it with a compliance risk label, the unit starts the traceability analysis process. First, the system reads the various characteristic values ​​of the business instance corresponding to the high-risk result in the characteristic matrix, including but not limited to time dimension characteristics (such as business execution interval, response delay), historical behavior characteristics (such as past violation frequency, rectification completion efficiency) and regulatory adaptability characteristics (such as the degree of deviation of key fields, satisfaction of required time limits, etc.). Each row in the characteristic matrix corresponds to a specific business instance, and each column is the standardized numerical representation of the instance under a certain type of compliance assessment dimension.

[0180] The analysis process uses the current risk tag as the tracing entry point, and reviews the historical evolution trajectory of the business instance at multiple time nodes in the past. Specifically, the system matches the instance with its historical records and extracts its continuous changes in the time dimension characteristics, such as whether there is a trend of accumulated delays in the task response time and whether the work cycle fluctuates significantly. Next, the system identifies the joint variation characteristics in the historical behavior characteristics, that is, it jointly models the number of violations and the rectification efficiency to determine whether there are trend characteristics of "frequent violations but slow rectification" or "deterioration of rectification efficiency". At the same time, it analyzes whether there is a sudden change in the regulatory adaptability characteristics, such as the situation where a regulatory indicator shows a cliff-like drop in adaptability around a certain time point. These trends, sudden changes and joint changes will be considered as possible precursors to risk intensification.

[0181] After comprehensively analyzing the above, the system extracts the multi-factor coupling relationship that leads to an increase in the risk level. For example, when a business instance experiences a sharp decline in regulatory compliance, there are also characteristics of repeated delays in task execution time and a simultaneous decline in rectification rate. The system will identify this combination as a potential dominant causal path. To achieve structured modeling, the system constructs a causal chain weight network, plotting each influencing factor into a multi-level graph structure according to the time sequence, triggering order, and historical impact. The nodes represent key business indicators or compliance factors, and the edges represent the causal direction with causal strength weights. The weight is determined by the following factors: the time difference between a certain feature change and the intensification of the risk, the statistical strength of the historical correlation between the feature and the risk level, and the frequency of joint action with other features.

[0182] After the causal chain weight network is established, the system will redistribute the causal weights of the time dimension features and historical behavior features in the original feature matrix based on the network. In other words, for fields with strong causal correlations, the system assigns them higher causal influence weights for subsequent training and optimization of compliance models or for weight adjustment in predictions. Finally, the causal chain traceability analysis unit outputs a complete risk tracing report, which clearly points out: 1. The dominant feature combination that leads to the current increased risk; 2. The critical time window in which the combination is located, such as two weeks before the start of the task scheduling anomaly, within one month after the regulatory update, etc.; 3. The key nodes with the most significant changes in the corresponding regulatory adaptability characteristics. All information is annotated in a structured data format and presented visually, making it easy for compliance managers to quickly locate the root cause of the problem.

[0183] The system also supports batch causal path analysis of multiple high-risk instances. When common causal chains exist, it generates causal clustering reports, identifying potential systemic management risks and assisting grassroots power supply companies in developing long-term governance strategies. Furthermore, the analysis unit includes a response mechanism for regulatory updates. When regulatory adjustments occur, the causal map is re-compared and revised to ensure the timeliness and rationality of the causal paths.

[0184] The following is an explanation using a specific embodiment. Assume that in a grassroots power supply enterprise, the system detects that a certain equipment inspection task is assessed as "high risk", and the risk level output by the intelligent risk assessment unit is 4 (serious risk), and the corresponding compliance risk label is "inspection overdue and record missing". At this time, the causal chain traceability analysis unit is started, taking the business instance as input, and first reads the following fields in its feature matrix: task_interval (task interval time) is 45 days, inspection_frequency (inspection frequency) is 3 times / month, task_completion_rate (task completion rate) is 0.52, past_violations (number of past violations) is 4, rectification_success_rate (rectification success rate) is 0.21, and compliance_adaptability (regulatory adaptability) is 0.55. The system traced the task schedule for this business over the past six months and found that task_interval had gradually increased from 21 days to 45 days over three consecutive cycles. Meanwhile, compliance_adaptability had decreased after the last two tasks, falling from 0.78 to 0.55. The rectification success rate was also below the industry average. The system identified the linkage between these three indicators as a coupled risk pattern and, based on historical statistical models, determined that similar combinations had a co-occurrence probability of over 80% in past risk level increases.

[0185] Therefore, the system constructs a causal path: reduced inspection frequency → longer intervals between tasks → decreased task completion rate → reduced adaptability → increased risk level, assigning the "longer intervals" node the highest causal weight in the causal chain. The system ultimately generates a causal report, recommending an increase in the frequency of future inspection scheduling and establishing a compliance warning threshold for "inter-task intervals exceeding 30 days" to prevent similar serious risks from recurring.

[0186] By introducing the aforementioned causal chain traceability analysis unit, this invention achieves a shift from outcome-based risk identification to source-based behavioral modeling. This enables compliance risk management to move beyond post-event warnings and further enable pre-event intervention and process control, significantly improving the governance capabilities and regulatory response efficiency of grassroots power supply companies. The entire analysis process is feasible, relying on data and calculations that can be completed through existing power supply companies' business data systems and analysis platforms, ensuring a high degree of engineering feasibility and legal clarity.

[0187] In the above embodiment, a compliance risk warning system for grassroots power supply enterprises based on big data analysis is provided. Correspondingly, this application also provides a compliance risk warning method for grassroots power supply enterprises based on big data analysis. Figure 2 , which is a flowchart of an embodiment of a compliance risk warning method for grassroots power supply enterprises based on big data analysis in this application. Since this embodiment, i.e., the second embodiment, is basically similar to the first embodiment, the description is relatively simple. For relevant details, please refer to the partial description of the first embodiment. The method embodiment described below is merely illustrative.

[0188] The second embodiment of the present application provides a compliance risk early warning method for grassroots power supply enterprises based on big data analysis, including:

[0189] Step S201: collecting business operation data, historical violation record data, and policy and regulation update data of grassroots power supply enterprises, and cleaning and formatting the data to form a unified compliance data set;

[0190] Step S202: Performing text segmentation, semantic recognition, and association analysis on the policy and regulation update data and historical violation record data to extract key risk factors and label them with corresponding compliance risk tags;

[0191] Step S203: constructing a feature matrix for risk identification based on the business operation data and the marked compliance risk labels, wherein the feature matrix includes fusion features of multiple dimensions;

[0192] Step S204: Input the feature matrix into the trained machine learning model to perform real-time compliance risk identification, determine the risk category and risk level corresponding to each business data instance, and generate compliance risk warning information based on the risk category and risk level.

[0193] Although the present application is disclosed as above with the preferred embodiments, it is not intended to limit the present application. Any person skilled in the art may make possible changes and modifications without departing from the spirit and scope of the present application. Therefore, the scope of protection of the present application shall be based on the scope defined by the claims of the present application.

Claims

1. A compliance risk early warning system for grassroots power supply enterprises based on big data analysis, characterized by: include: The data collection unit is used to collect business operation data, historical violation records, and policy and regulatory update data from grassroots power supply enterprises, and perform data cleaning and format standardization to form a unified compliance data set; A natural language processing unit is used to perform text segmentation, semantic recognition, and association analysis on the policy and regulatory update data and historical violation record data, extract key risk factors, and label them with corresponding compliance risk tags; a risk feature construction unit, configured to perform multi-dimensional feature fusion on the business operation data and the risk label data processed by the natural language processing unit to form a feature matrix for risk identification; An intelligent risk assessment unit is used to use a pre-trained machine learning model to perform real-time compliance risk identification on the feature matrix, determine the category and level of the risk, and generate corresponding early warning information based on the risk category and level.

2. The compliance risk early warning system for grassroots power supply enterprises based on big data analysis according to claim 1 is characterized in that: Also includes: The early warning and risk avoidance suggestion unit is connected to the intelligent risk assessment unit and is used to automatically generate targeted risk avoidance suggestions based on the identified risk categories and levels, and push the early warning information and avoidance suggestions to the compliance management terminal of the grassroots power supply enterprise to guide the enterprise to actively implement risk prevention measures.

3. The compliance risk early warning system for grassroots power supply enterprises based on big data analysis according to claim 1 is characterized in that: The data collection unit includes an incremental regulatory sensitive data capture module for updating the collected policy and regulatory data and identifying sensitive clauses directly related to specific business activities of grassroots power supply enterprises; Data capture trigger rules with time stamps and regulatory association identifiers are automatically generated based on the identified sensitive clauses; based on the trigger rules, data that meets the characteristics of sensitive clauses is captured in the business operation data flow of the grassroots power supply enterprise, and is separately marked to form a sensitive compliance data subset.

4. The compliance risk early warning system for grassroots power supply enterprises based on big data analysis according to claim 1 is characterized in that: The data collection unit includes a violation risk data backtracking collection module, which is used to automatically analyze the historical violation record data corresponding to the risk category when the intelligent risk assessment unit detects a specific category of compliance risk and generates early warning information to determine the key violation event characteristics related to the current risk; based on the determined key violation event characteristics, generate data backtracking rules, including the time window of the violation event, the type of equipment involved or the position of the responsible person; based on the data backtracking rules, conduct a secondary backtracking collection of the historical business operation data stored by the grassroots power supply enterprise, and generate a special backtracking data set with violation association marks for continuous optimization of the subsequent risk feature model.

5. The compliance risk early warning system for grassroots power supply enterprises based on big data analysis according to claim 1 is characterized in that: The natural language processing unit is specifically used for: Using a multi-granularity domain word embedding model specifically built for grassroots power supply companies, we dynamically segment and embed the data on policy and regulatory updates and historical violation records to capture the implicit semantic relationships between different regulatory terms and corporate compliance terms. Using the word vector mapping results, a semantic relationship network in the field of compliance management of power supply enterprises is constructed through semantic similarity analysis to clarify the semantic associations between key terms in different regulatory provisions and violation records; Based on the semantic similarity and co-occurrence frequency between nodes in the semantic relationship network, cross-text risk factor association analysis is performed to form an accurate risk factor combination and annotate the corresponding compliance risk labels.

6. The compliance risk early warning system for grassroots power supply enterprises based on big data analysis according to claim 5 is characterized in that: The word vector mapping process of the natural language processing unit includes: Based on the node weights of key terms in the constructed semantic relationship network, the semantic relevance parameters of the word vector mapping are dynamically adjusted to reflect the semantic change trends of regulatory clause updates while generating semantic similarity. Combined with the context of the newly added regulatory provisions, the word vector embedding representation is automatically updated, and through an adaptive iterative process, the word vector is kept consistent with the original semantic network, enhancing the accuracy of the association analysis of cross-text risk factors.

7. The compliance risk early warning system for grassroots power supply enterprises based on big data analysis according to claim 1 is characterized in that: The risk feature construction unit is specifically used to: Based on the business categories covered by regulatory requirements, business operation data is divided into equipment management data, operation and maintenance scheduling data, and financial compliance data. Each data category is further subdivided based on the constraints of regulatory provisions. Equipment management data is categorized by inspection frequency, maintenance record completeness, and equipment load status. Operation and maintenance scheduling data is categorized by scheduling response time, task completion status, and personnel deployment. Financial compliance data is categorized by capital flow records, contract execution status, and expenditure approval process. When matching risk tag data with business data, the system extracts the corresponding data fields from the business data based on the numerical restrictions, time requirements, and operational specifications of regulatory provisions, and calculates their degree of compliance with regulatory requirements. For regulatory provisions that include time restrictions, the system calculates the time intervals between data fields and marks data that exceeds the regulatory limits as anomalies. For regulatory provisions that include operational process requirements, the system calculates the task completion status of the business data and marks unfinished or overdue data as anomalies. When forming the feature matrix, based on the time series analysis method, the time interval difference, historical violation frequency and cumulative violation days are extracted, and the trend characteristics of historical data are added to the feature matrix, so that the system can identify the cumulative effect of long-term violation risks and provide a basis for evaluating regulatory compliance in different time windows.

8. The compliance risk early warning system for grassroots power supply enterprises based on big data analysis according to claim 7 is characterized in that: When calculating the time interval of the data field, the risk feature construction unit dynamically adjusts the selection range of the time window based on the timeliness requirements of the regulatory provisions and the historical operating model of the enterprise; among them, for the case where the regulatory requirements clearly stipulate a fixed time period, the time threshold stipulated by the regulatory requirements is directly used for calculation; for the case where the regulations do not clearly provide a fixed time period but there are industry practices or internal management standards of the enterprise, the most common time interval is calculated based on historical business data, and the time interval is used as a benchmark for compliance judgment; in the case of regulatory updates or business process adjustments, the time window change trends of the new and old regulatory requirements are compared, and the time interval calculation rules are dynamically adjusted to ensure that the feature matrix can adapt to regulatory changes and improve the accuracy of compliance risk assessment.

9. The compliance risk early warning system for grassroots power supply enterprises based on big data analysis according to claim 1 is characterized in that: The machine learning model of the intelligent risk assessment unit includes a compliance feature extraction network, a risk classification network and a dynamic adaptive adjustment module; The input of the compliance feature extraction network is the time dimension features, business indicator features, historical behavior features, and regulatory adaptability features in the feature matrix. It is used to perform time series expansion of the time dimension features based on the hierarchical feature decomposition method, extract task interval deviations and regulatory execution time limit compliance, calculate the time decay weights of key business data, and perform normalization processing on the business indicator features so that the equipment operation status, inspection frequency, and task completion rate features are calculated under a unified numerical scale. The input of the risk classification network is the normalized compliance feature vector generated by the compliance feature extraction network. Based on the regulation matching rules and classification enhancement mechanism, the network identifies the compliance risk category of the input features, and combines the number of violations and rectification efficiency information in the historical behavior characteristics to calculate the risk level and generate a risk category label. The input of the dynamic adaptive adjustment module includes regulatory update information and historical prediction deviation data of the risk classification network. The module parses the regulatory adjustment content based on the regulatory change detection mechanism, and combines the regulatory adaptability characteristics to calculate the degree of deviation between the criticality of regulatory requirements and corporate behavior, and dynamically adjusts the classification weights and risk level division thresholds of the risk classification network.

10. A compliance risk early warning method for grassroots power supply enterprises based on big data analysis, characterized in that: include: Collect business operation data, historical violation records, and policy and regulatory update data from grassroots power supply companies, cleanse and standardize the data to form a unified compliance data set; Perform text segmentation, semantic recognition, and association analysis on the policy and regulatory update data and historical violation record data to extract key risk factors and label them with corresponding compliance risk tags; Based on the business operation data and the annotated compliance risk labels, a feature matrix for risk identification is constructed, wherein the feature matrix includes fusion features of multiple dimensions; The feature matrix is ​​input into the trained machine learning model to perform real-time compliance risk identification, determine the risk category and risk level corresponding to each business data instance, and generate compliance risk warning information based on the risk category and risk level.

Images