Power system sensitive word negative list detection method and system based on knowledge graph

Abstract

The invention relates to the technical field of information identification, in particular to a power system sensitive word negative list detection method and system based on a knowledge graph. The method comprises the steps of preprocessing to-be-recognized sensitive data of the power system, and performing sensitive word recognition prediction on the sensitive data of the power system by using a pre-trained deep learning model to obtain sensitive word data; inputting the preprocessed sensitive data word sequence into a knowledge graph to obtain a reasoning relationship between graph entities; according to the reasoning relation and the sensitive word data, whether the sensitive data of the power system belong to a negative list or not is judged, if yes, the sensitive data of the power system is labeled with a negative list label, and if not, the sensitive data of the power system is labeled with a non-negative list label. According to the method, reasoning and judgment are performed in combination with the knowledge graph, the sensitive words in the power system data can be effectively recognized, manpower needed by negative list sorting work is greatly reduced, the detection accuracy and efficiency are improved, and the power system data safety is guaranteed.

Description

Technical Field

[0001] The present invention relates to the field of information recognition technology, and in particular to a method and system for detecting a negative list of sensitive words in a power system based on a knowledge graph. Background Art

[0002] As the power system continues to become more information-rich, massive amounts of data are generated, transmitted, and stored within it. This data includes a significant amount of sensitive word negative list data. This data, which includes personal privacy data and various sensitive data, is complex and diverse. Currently, the negative list process relies heavily on manual identification of what constitutes negative list data. This makes the process of sorting out negative list data tedious and repetitive, requiring significant manpower. Identifying sensitive word negative list data is crucial for ensuring power system data security.

[0003] Currently, there are some research results in sensitive word detection. The industry typically uses machine learning-based models to identify negative lists of sensitive words. For example, Chinese invention application CN114091436A provides a sensitive word detection method based on decision trees and variant recognition. However, in practice, while machine learning models can predict new data and flag potentially sensitive content, these models typically operate based on feature matching and statistical laws. They may not fully understand the meaning of words in a specific context, potentially leading to false positives or false negatives.

[0004] For example, CN114860897A provides a sensitive word detection method, device, electronic device and readable storage medium, which discloses that in response to the update of new corpus containing sensitive words in the corpus, the knowledge graph is updated according to the new corpus, and the training samples constructed by extracting and precipitating the graph knowledge and combining the target corpus in the corpus are more suitable for strong adversarial amplifier scenarios such as product reviews. The training samples are used to optimize the discriminant model, and the detection accuracy is continuously improved through continuous corpus updates and model training. However, it is oriented to Internet scenarios (such as user comments, nicknames, UGC content), and detects common sensitive words such as sensitive people, illegal information, and bad content, and is suitable for scenarios such as e-commerce and social platforms. However, the accuracy of existing detection methods needs to be improved when processing complex data structures and professional terms of power systems, and there is a lack of effective integration and reasoning of knowledge related to specific negative lists of power systems. Summary of the Invention

[0005] To address the problems in the existing technology, the present invention provides a method and system for detecting a negative list of sensitive words in the power system based on a knowledge graph. This method can effectively combine a machine learning sensitive word recognition model with a knowledge graph, thereby improving the accuracy and efficiency of detecting a negative list of sensitive words in power system data. The specific technical solution is as follows:

[0006] A method for detecting a negative list of sensitive words in a power system based on a knowledge graph includes the following steps:

[0007] Step S1, pre-processing the power system sensitive data to be identified to obtain a sensitive data word sequence;

[0008] Step S2: Use the pre-trained deep learning model to identify and predict sensitive words in the sensitive data word sequence to obtain sensitive word data;

[0009] Step S3: Input the sensitive data word sequence into the knowledge graph for processing to obtain the inference relationship between graph entities;

[0010] Step S4: Based on the inference relationship and the sensitive word data, determine whether the power system sensitive data belongs to the negative list. If so, mark the power system sensitive data with a negative list label; otherwise, mark the power system sensitive data with a non-negative list label.

[0011] Preferably, the step S1 pre-processes the sensitive data of the power system to be identified, specifically including:

[0012] Identifying noise information of the sensitive data of the power system and cleaning the noise information of the sensitive data of the power system;

[0013] After cleaning the noise information, the sensitive data of the power system is segmented to obtain the sensitive data word sequence.

[0014] Preferably, cleaning noise information of sensitive data of the power system includes:

[0015] Clean HTML tags, extra spaces, and special characters in sensitive data in the power system.

[0016] Preferably, the deep learning model is a BiLSTM-CRF model.

[0017] Preferably, the training steps of the deep learning model are:

[0018] Establish a sensitive word library;

[0019] Collect a negative list of historical power system data, annotate the historical power system data according to the sensitive word library, and construct a training set;

[0020] Build a BiLSTM-CRF model;

[0021] The training set is input into the BiLSTM-CRF model for training, and the model with the highest recognition accuracy is output as the trained deep learning model.

[0022] Preferably, step S3 further includes: constructing a knowledge graph; wherein the construction of the knowledge graph specifically includes the following steps:

[0023] Extract graph entities based on sensitive data word sequences and define the characteristic attributes of graph entities;

[0024] Construct a power system data knowledge graph framework, where the power system data knowledge graph framework includes: power equipment name, data type and operation behavior.

[0025] Preferably, step S5 is also included, in which the judgment result is fed back to the constructed knowledge graph to update the iterative knowledge graph.

[0026] A knowledge graph-based power system sensitive word negative list detection system, applying the method described above, includes: a data preprocessing module for preprocessing power system sensitive data to be identified to obtain a sensitive data word sequence; a sensitive word detection module for using a pre-trained deep learning model to identify and predict sensitive words in the power system sensitive data to obtain sensitive word data;

[0027] The knowledge graph reasoning module is used to input the sensitive data word sequence into the knowledge graph to obtain the reasoning relationship between the graph entities; the result judgment module is used to judge whether the power system sensitive data belongs to the negative list based on the reasoning relationship and the obtained sensitive word data. If so, the power system sensitive data is marked with a negative list label; otherwise, the power system sensitive data is marked with a non-negative list label.

[0028] A computer-readable storage medium includes a stored program, wherein when the program is run, the device where the computer-readable storage medium is located is controlled to execute the method for detecting a negative list of sensitive words in a power system based on a knowledge graph.

[0029] A processor is used to run a program, wherein when the program is running, the method for detecting a negative list of sensitive words in a power system based on a knowledge graph is executed.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] The present invention combines knowledge graphs for reasoning and judgment, which can effectively identify sensitive words in power system data, greatly reduce the manpower required for sorting out negative lists, improve detection accuracy and efficiency, and ensure the security of power system data. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly describes the drawings required for the specific embodiments or the description of the prior art. Similar elements or parts are generally identified by similar reference numerals throughout the drawings. Elements or parts in the drawings are not necessarily drawn to scale.

[0033] Figure 1 Flow chart of the method of the present invention. DETAILED DESCRIPTION

[0034] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0035] It will be understood that when used in this specification and the appended claims, the terms “comprises” and “comprising” indicate the presence of described features, integers, steps, operations, elements and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0036] It should also be understood that the terms used in the present specification are only for the purpose of describing particular embodiments and are not intended to limit the present invention. As used in the present specification and the appended claims, the singular forms "a", "an", and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0037] It should be further understood that the term "and / or" used in the present description and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0038] A method for detecting a negative list of sensitive words in a power system based on a knowledge graph includes the following steps:

[0039] Step S1 : pre-processing the sensitive data of the power system to be identified, removing irrelevant information, making the data convenient for subsequent processing, and obtaining a sensitive data word sequence.

[0040] Comprehensive data collection is conducted from multiple data sources within the power system, including but not limited to databases (which store equipment parameters and user information), monitoring systems (which collect real-time equipment operating data), and communication logs (which record communication information and operating instructions within the system). This ensures that the data covers all aspects of power system operation to ensure comprehensive and accurate testing. For example, historical maintenance records and configuration parameters of power equipment can be obtained from databases, while real-time operating data such as voltage and current of transmission lines can be obtained from monitoring systems. Furthermore, information such as operator control instructions for equipment can be extracted from communication logs.

[0041] The preprocessing of the power system sensitive data to be identified specifically includes identifying noise information of the power system sensitive data, cleaning and word segmentation of the noise information of the power system sensitive data, and converting the text into a word sequence.

[0042] Cleaning noise from sensitive power system data involves removing HTML tags (such as when extracting data from web-based power reports), redundant spaces (to prevent them from affecting text analysis accuracy), and special characters (such as meaningless characters that are not specialized for the power system). When processing power system data, cleaning and segmenting the raw data, which contains a large amount of redundant information, improves data quality and usability, laying the foundation for subsequent knowledge graph construction and sensitive word identification.

[0043] Word segmentation involves using specialized word segmentation tools to convert text data into words or word sequences. For example, a text record containing the operating status of an equipment, "#1 Transformer oil temperature 65°C, operating normally," would be cleaned and segmented into "#1 Transformer oil temperature 65°C, operating normally." This provides clear, structured data for subsequent knowledge graph construction and sensitive word identification.

[0044] Step S2: Use a pre-trained deep learning model to identify and predict sensitive words in the power system sensitive data. The deep learning model is a BiLSTM-CRF model. The training steps of the deep learning model are:

[0045] When training the model, we collected a large amount of text data related to the power system, including equipment operation reports, maintenance records, and user operation logs. Based on the existing negative list data, we collected sensitive words from power system-related regulations (such as descriptive terms for illegal operations in power safety regulations), policies (such as key indicator terms in energy management policies), safety standards (such as specific parameter terms in electrical equipment safety standards), and industry practices (such as common descriptions of specific fault phenomena in the power industry) to build a sensitive word library.

[0046] Collect a series of sensitive words and related data combinations that are contrary to the safety, compliance, and stable operation of the power system, define them as a negative list, annotate them, and construct a training set;

[0047] Build a BiLSTM-CRF model;

[0048] The training set is fed into the BiLSTM-CRF model for training, enabling it to accurately identify sensitive words related to power system safety and compliance. The model with the highest recognition accuracy is output as the trained deep learning model.

[0049] The negative list refers to a collection of sensitive words and related data that are contrary to the safety, compliance, and stable operation of the power system. These sensitive words typically involve the risk of power equipment failure, illegal or irregular user operations, violations of power system safety policies, and other key information that may cause power safety incidents or violate industry standards, laws, and regulations.

[0050] By learning sensitive words and contextual semantic information from large amounts of annotated data, the model can more accurately identify potential sensitive words. The model then combines the identification results with the knowledge graph used to determine negative lists. For example, if the model identifies the sensitive word "illegal operation," it uses the knowledge graph to search for information related to "illegal operation," such as equipment, personnel, and operational procedures, to further determine the specific impact and risk level of this sensitive word in the power system.

[0051] Step S3: Input the sensitive data word sequence into the knowledge graph for processing to obtain the inference relationship between the graph entities. This also includes constructing the knowledge graph; the construction of the knowledge graph specifically includes the following steps:

[0052] Based on the word sequence of sensitive data, graph entities are extracted and the characteristic attributes of graph entities are defined; for example, power equipment (including generators, substations, transmission lines, etc.), users (including industrial users, commercial users, residential users, etc.), data (such as real-time operating data such as voltage, current, power, equipment configuration parameters, etc.) are taken as entities, and their attributes include name, type, status (such as equipment operation, shutdown, fault status), etc.

[0053] Specifically, in constructing the knowledge graph, key entities in the power system are identified, such as substations (including attributes such as their geographic location and region), transmission lines (including attributes such as line length and conductor type), and power users (covering attributes such as user type and power consumption level), along with their attributes. Through in-depth analysis of the power system's operating logic and business rules, relationships between entities are established. For example, based on the power system's topological structure, the connection between substations and transmission lines (e.g., a substation supplies power to a certain area via a specific transmission line) and the ownership relationship between users and power equipment (e.g., an industrial user owns multiple motors of a specific power) are determined. Furthermore, consideration can be given to the collaborative working relationships between devices (e.g., load transfer relationships between multiple substations) and the association between equipment and environmental factors (e.g., the relationship between outdoor equipment and meteorological conditions), enabling the knowledge graph to more comprehensively reflect the actual operation of the power system.

[0054] Construct a knowledge graph framework for power system data. This framework includes information about power equipment names (such as the specific models and names of transformers and circuit breakers), data types (such as real-time monitoring data and historical statistics), and operational behaviors (such as equipment startup, shutdown, and adjustment operations) to accurately describe power system information. By defining these entities and attributes, various elements in the power system can be presented in a structured manner, facilitating relationship analysis and reasoning.

[0055] Based on the power system's operating principles and safety regulations, reasoning relationships between graph entities are defined, such as the association between equipment and operations (e.g., specific equipment can only be operated by specific personnel) and the relationship between data and users (e.g., user access rights to certain data). This provides a foundation for subsequent knowledge reasoning. These reasoning relationships, based on the power system's operating principles, safety regulations, and business rules, help the system infer potential risks and issues based on known information. Through in-depth analysis of the power system's operating logic and business rules, relationships between entities are established, such as the connection between substations and transmission lines, and the ownership relationship between users and electrical equipment, forming a complete knowledge graph structure.

[0056] Inference relationship definition: Clarify the inference relationships between graph entities, such as the association between equipment failure (e.g., transformer oil temperature is too high) and data anomalies (e.g., oil temperature monitoring data exceeds a threshold), and the relationship between user operations (e.g., unauthorized modification of equipment parameters) and data access permissions (e.g., the user does not have modification permissions). For example, when the current of a transmission line exceeds its rated value, based on the physical characteristics of the power system and the operating patterns of the equipment, it can be inferred that there may be a risk of equipment failure (e.g., line overheating may cause insulation damage). When users frequently access sensitive data and have unusual data access patterns (e.g., downloading large amounts of configuration data for critical equipment in a short period of time), security regulations and data protection principles may indicate a data leakage risk. Furthermore, inference relationships can be defined between equipment aging and maintenance needs (e.g., equipment operating beyond a certain age requires increased maintenance frequency), and between power load changes and grid stability (e.g., a sudden and significant increase in load may affect grid voltage stability), to more comprehensively cover various potential risks and correlations in the power system. These inference relationships are based on power system operating principles, safety regulations, and practical experience, providing a logical basis for subsequent outcome judgments.

[0057] Step S4: Based on the inference relationship and the sensitive word data, determine whether the power system sensitive data belongs to the negative list. If so, mark the power system sensitive data with a negative list label; otherwise, mark the power system sensitive data with a non-negative list label.

[0058] Specifically, based on defined inference rules, the entity relationships and sensitive word data in the knowledge graph are used to obtain negative list determination results. For example, if a sensitive word is associated with the failure risk of critical equipment, and this equipment plays an important role in the power system, then according to the inference rules, the data can be determined to be on the negative list.

[0059] For example, if the data contains sensitive words related to the risk of equipment failure (such as "short circuit", "overload", etc.), and the equipment belongs to critical power facilities (such as the main transformer of a hub substation), it will be judged as negative list data; if the user's access to the data involves sensitive words (such as "illegal access") and violates the access permission rules (such as an ordinary user trying to access data with administrator privileges), it will also be judged as negative list data. At the same time, consider the combined impact of multiple factors, such as the frequency of occurrence of sensitive words, the degree of correlation with other key information, etc. For example, if a sensitive word frequently appears in data related to a certain device over a period of time, even if each occurrence may not constitute a serious problem individually, the combined appearance may suggest that the device has potential risks, and thus be judged as negative list data.

[0060] The present invention integrates the recognition results with the information in the knowledge graph to further improve the accuracy of sensitive word recognition.

[0061] In step S5, the judgment results are fed back into the constructed knowledge graph to update and iterate the knowledge graph. Throughout the implementation process, the structure and inference rules of the knowledge graph are continuously optimized, and feedback adjustments are made based on actual detection results to improve the accuracy and effectiveness of the system's detection of the negative list of sensitive terms in the power system. At the same time, the sensitive term library is regularly updated to adapt to changes in power system data security requirements (for example, new sensitive terms are promptly incorporated as new safety regulations are introduced or power technology develops).

[0062] This example tests and compares the present invention with the sensitive word detection method, device, electronic device, and readable storage medium of a Chinese invention application (CN 114860897 A). The specific results are as follows:

[0063] Table 1 Dataset size

[0064]

[0065] Table 2 Performance index comparison

[0066]

[0067] It can be seen that the present invention combines the knowledge graph for reasoning and judgment, which can effectively identify sensitive words in power system data, greatly reduce the manpower required for sorting out the negative list, improve detection accuracy and efficiency, and ensure the security of power system data.

[0068] Example 2:

[0069] Based on the same inventive concept as Example 1, this embodiment provides a knowledge graph-based detection system for a negative list of sensitive words in a power system, and the method described herein includes:

[0070] A data preprocessing module is used to preprocess the sensitive data of the power system to be identified and obtain a sensitive data word sequence;

[0071] The sensitive word detection module is used to use the pre-trained deep learning model to identify and predict sensitive words in the power system sensitive data to obtain sensitive word data;

[0072] The knowledge graph reasoning module is used to input sensitive data word sequences into the knowledge graph for processing and obtain the reasoning relationship between graph entities;

[0073] The result judgment module is used to determine whether the sensitive data of the power system belongs to the negative list based on the reasoning relationship and sensitive word data. If so, the sensitive data of the power system is marked with a negative list label; otherwise, the sensitive data of the power system is marked with a non-negative list label.

[0074] Example 3:

[0075] Based on the same inventive concept as Example 1, this embodiment provides a computer-readable storage medium, which includes a stored program, wherein when the program is running, the device where the computer-readable storage medium is located is controlled to execute the method for detecting a negative list of sensitive words in a power system based on a knowledge graph.

[0076] Example 4:

[0077] Based on the same inventive concept as Example 1, this embodiment provides a processor, which is used to run a program, wherein the program executes the method for detecting a negative list of sensitive words in a power system based on a knowledge graph when running.

[0078] Those skilled in the art will appreciate that the units of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition of each example has been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0079] In the embodiments provided by the present invention, it should be understood that the division of units is merely a logical function division, and there may be other division methods in actual implementation, for example, multiple units can be combined into one unit, one unit can be split into multiple units, or some features can be ignored, etc.

[0080] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0081] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-0nly Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk, etc., various media that can store program code.

[0082] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some or all of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the claims and description of the present invention.

Claims

1. A method for detecting a negative list of sensitive words in a power system based on a knowledge graph, characterized in that: The following steps are involved: Step S1, pre-processing the power system sensitive data to be identified to obtain a sensitive data word sequence; Step S2: Use the pre-trained deep learning model to identify and predict sensitive words in the sensitive data word sequence to obtain sensitive word data; Step S3: Input the sensitive data word sequence into the knowledge graph for processing to obtain the inference relationship between graph entities; Step S4: Based on the inference relationship and the sensitive word data, determine whether the power system sensitive data belongs to the negative list. If so, mark the power system sensitive data with a negative list label; otherwise, mark the power system sensitive data with a non-negative list label.

2. A method for detecting a negative list of sensitive words in a power system based on a knowledge graph according to claim 1, characterized in that: In step S1, the sensitive data of the power system to be identified is preprocessed, specifically including: Identifying noise information of the sensitive data of the power system and cleaning the noise information of the sensitive data of the power system; After cleaning the noise information, the sensitive data of the power system is segmented to obtain the sensitive data word sequence.

3. The method for detecting a negative list of sensitive words in a power system based on a knowledge graph according to claim 2 is characterized in that: Cleaning noise information of sensitive data in power systems includes: Clean HTML tags, extra spaces, and special characters in sensitive data in the power system.

4. A method for detecting a negative list of sensitive words in a power system based on a knowledge graph according to any one of claims 1 to 3, characterized in that: The deep learning model is the BiLSTM-CRF model.

5. The method for detecting a negative list of sensitive words in a power system based on a knowledge graph according to claim 4 is characterized in that: The training steps of a deep learning model are: Establish a sensitive word library; Collect a negative list of historical power system data, annotate the historical power system data according to the sensitive word library, and construct a training set; Build a BiLSTM-CRF model; The training set is input into the BiLSTM-CRF model for training, and the model with the highest recognition accuracy is output as the trained deep learning model.

6. The method for detecting a negative list of sensitive words in a power system based on a knowledge graph according to claim 1, characterized in that: The step S3 further includes: constructing a knowledge graph; wherein the construction of the knowledge graph specifically includes the following steps: Extract graph entities based on sensitive data word sequences and define the characteristic attributes of graph entities; Construct a power system data knowledge graph framework, where the power system data knowledge graph framework includes: power equipment name, data type and operation behavior.

7. The method for detecting a negative list of sensitive words in a power system based on a knowledge graph according to claim 1, characterized in that: It also includes: step S5, feeding back the judgment result to the constructed knowledge graph to update the iterative knowledge graph.

8. A knowledge graph-based power system sensitive word negative list detection system, characterized by: Applying the method according to any one of claims 1 to 7, comprising: The data preprocessing module is used to preprocess the sensitive data of the power system to be identified and obtain a sensitive data word sequence; the sensitive word detection module is used to use the pre-trained deep learning model to identify and predict sensitive words in the sensitive data of the power system and obtain sensitive word data; The knowledge graph reasoning module is used to input sensitive data word sequences into the knowledge graph for processing and obtain the reasoning relationship between graph entities; The result judgment module is used to determine whether the sensitive data of the power system belongs to the negative list based on the reasoning relationship and sensitive word data. If so, the sensitive data of the power system is marked with a negative list label; otherwise, the sensitive data of the power system is marked with a non-negative list label.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium includes a stored program, wherein, when the program is running, the device where the computer-readable storage medium is located is controlled to execute a method for detecting a negative list of sensitive words in a power system based on a knowledge graph as described in any one of claims 1 to 7.

10. A processor, characterized in that: The processor is used to run a program, wherein when the program is run, it executes a method for detecting a negative list of sensitive words in a power system based on a knowledge graph as described in any one of claims 1 to 7.

Images