Power grid dispatching adaptive evaluation question generation method and system based on semantic model

Abstract

The application provides a power grid dispatching adaptive evaluation question generation method and system based on a semantic model, relates to the technical field of power grid simulation intelligent training, and mainly comprises the following steps: obtaining dispatching regulation text data; constructing a dispatching knowledge skill point network model, combining the dispatching regulation text data, and forming a dispatching knowledge base; preprocessing the dispatching knowledge base, and generating a dispatching examination point network model with a key sentence; based on the dispatching examination point network model with the key sentence, extracting relevant knowledge points and the key sentences or key paragraphs connected therewith, and adaptively generating an evaluation question stem; generating an evaluation question distractor according to the knowledge point correlation of the dispatching examination point network model; and performing initial difficulty setting analysis on the question, and dynamically correcting the question difficulty according to the test result of a trainee. The application can exert the flexibility of the text technology, utilize the correlation of the knowledge points in the semantic model, and improve the diversity and depth of the question.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of power grid simulation intelligent training, and particularly relates to a power grid dispatching adaptive evaluation question generation method and system based on a semantic model. BACKGROUND

[0002] The statements in this section merely provide background information related to the present application and do not necessarily constitute the prior art.

[0003] In order to improve the efficiency and effectiveness of training and examination, the automatic question generation technology is applied to the power grid dispatcher training, which not only can solve many problems existing in the traditional method, but also can significantly increase the diversity and flexibility of the questions, so as to meet the higher level of evaluation requirements.

[0004] The existing automatic generation technology is mainly divided into two categories: text-based automatic generation technology and knowledge graph-based automatic generation technology.

[0005] The text-based question generation technology uses natural language processing and machine learning algorithms to extract text knowledge from electronic documents and corpora to generate questions that meet the test requirements. Early methods rely on templates and rules to extract key information from text through syntax rules. It also uses neural networks such as long short-term memory networks, convolutional neural networks, Transformer models, and BERT pre-training language models, but it needs to perform specific preprocessing on the input text, such as sentence simplification and semantic disambiguation. The generated questions can only be limited to the text itself, and cannot obtain information associated with the text at the same time, which limits the scope and depth of the questions.

[0006] The knowledge graph-based automatic generation technology uses knowledge graphs to generate educational test questions through big data and artificial intelligence technology, mainly focusing on building small-scale knowledge graphs and generating questions through simple rules. Knowledge graphs visually display knowledge concepts or entities (nodes) and their relationships (edges) through graphical means. However, the current technology mainly generates distractors for concept nouns, and still has limitations in processing sentences describing knowledge points, which limits the performance of the questions in difficulty and adaptability. SUMMARY

[0007] To overcome the shortcomings of the above-mentioned prior art, the present application provides a power grid dispatching adaptive evaluation question generation method and system based on a semantic model, which can combine the characteristics of the power grid dispatching field, combine the dispatching knowledge and skill point network model with the dispatching regulations, accident plans, safety operation guidance books and power grid dispatching professional knowledge, and fully utilize the relationship features of the knowledge and skill point network model as a structured semantic knowledge base and the rich connotation of the text to generate adaptive knowledge evaluation questions in the power grid dispatching field.

[0008] To achieve the above object, one or more embodiments of the present application provide the following technical solutions:

[0009] The first aspect of the present application provides a power grid dispatching adaptive evaluation question generation method based on a semantic model.

[0010] A power grid dispatching adaptive evaluation question and answer generation method based on a semantic model comprises:

[0011] Obtaining dispatching regulation text data;

[0012] Building a dispatching knowledge and skill point network model, combining the dispatching regulation text data to form a dispatching knowledge base;

[0013] Preprocessing the dispatching knowledge base to generate a dispatching examination point network model with key sentences;

[0014] Based on the dispatching examination point network model with key sentences, extracting relevant knowledge points and their key sentences or key paragraphs, and adaptively generating evaluation question stems;

[0015] Generating evaluation question distractors according to the knowledge point association relationship of the dispatching examination point network model;

[0016] Setting the initial difficulty of the question for analysis, and dynamically correcting the difficulty of the question according to the test results of the students.

[0017] The second aspect of the present application provides a power grid dispatching adaptive evaluation question generation system based on a semantic model.

[0018] A power grid dispatching adaptive evaluation question generation system based on a semantic model comprises:

[0019] The semantic model module is configured to obtain dispatching regulation text data;

[0020] Building a dispatching knowledge and skill point network model, combining the dispatching regulation text data to form a dispatching knowledge base;

[0021] Preprocessing the dispatching knowledge base to generate a dispatching examination point network model with key sentences;

[0022] The adaptive test question stem generation module is configured to extract relevant knowledge points and their key sentences or key paragraphs based on the dispatching examination point network model with key sentences, and adaptively generate evaluation question stems;

[0023] The distractor generation module is configured to generate evaluation question distractors according to the knowledge point association relationship of the dispatching examination point network model;

[0024] The title quality evaluation module is configured to perform initial difficulty setting analysis on the title, and to dynamically correct the title difficulty according to test results of the trainees.

[0025] The third aspect of the present application provides a computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps in the method according to the first aspect of the present application when executing the program.

[0026] The fourth aspect of the present application provides a computer readable storage medium having a program stored thereon, wherein the program is executable by a processor to implement the steps in the method according to the first aspect of the present application.

[0027] The above one or more technical solutions have the following beneficial effects:

[0028] The present application provides a power grid dispatching adaptive evaluation question generation method and system based on a semantic model, which can combine the characteristics of the power grid dispatching field, combine the dispatching knowledge skill point network model with dispatching regulations, accident plans, safety operation guidance books and power grid dispatching professional knowledge, fully utilize the relationship characteristics of the knowledge skill point network model as a structured semantic knowledge base and the rich connotation of the text, and generate adaptive knowledge evaluation questions in the power grid dispatching field. This method not only has the flexibility of text-based technology, but also uses the comprehensiveness and relevance of the knowledge graph to improve the diversity and depth of question generation.

[0029] In the present application, the knowledge point association relationship based on the dispatching examination point network model is adopted, a new pseudo-proposition construction method is designed, which is specially designed for judgment questions and selection questions, including key number replacement, comparison word replacement and positive and negative word replacement, aiming to more accurately generate interference items that can effectively confuse dispatchers. Through this association relationship-based strategy, we can improve the quality and educational value of the questions, not only test the dispatcher's mastery of basic knowledge, but also evaluate their understanding ability of the complex relationship between knowledge points. This method not only makes the questions more challenging, but also helps dispatchers to think and analyze more deeply when solving practical problems.

[0030] In the present application, personalized questions can also be generated by using intelligent algorithms based on the learning records and behavior data of students or employees, improving the relevance of learning and assessment. By using big data and real-time data analysis technology, the knowledge graph and generation algorithm are dynamically updated to ensure the timeliness and accuracy of question generation.

[0031] Advantages of additional aspects of the present application will be partially given in the following description, partially will become obvious from the following description, or will be understood by the practice of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0032] The accompanying drawings, which form a part of this specification, are included to provide a further understanding of the application and are incorporated in and constitute a part of this specification. The embodiments of the application, and their

[0033] Figure 1 A flow chart for constructing a dispatch examination point network model in the first embodiment of the application;

[0034] Figure 2 A flow chart for the blank filling question stem automatic generation method in the first embodiment of the application. DETAILED DESCRIPTION

[0035] It should be noted that the following detailed description is merely exemplary and is intended to provide further description of the application. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0036] It should be noted that the terms used herein are merely for the purpose of describing the specific embodiments and are not intended to limit the exemplary embodiments according to the application.

[0037] In the case of no conflict, the embodiments in the application and the features in the embodiments can be combined with each other.

[0038] Embodiment One

[0039] The embodiment discloses a power grid dispatch adaptive evaluation question generation method based on a semantic model, constructs a question generation framework for hanging key sentences based on a knowledge and skill point network model, as shown in Figure 1 To more clearly illustrate the embodiment, the power grid dispatch adaptive evaluation question generation method based on the semantic model can be specifically described as follows:

[0040] Step 1, obtaining dispatch regulation text data;

[0041] The obtained dispatch regulation text data is specifically, according to the data information of SCADA / EMS, OMS, PMS, and the like, dispatch rules and regulations, technical specifications, management regulations, training specifications, safety operation guidance, power dispatcher professional ability training specifications, and the like, relevant materials such as power grid dispatch related basic teaching materials, according to the knowledge and skill range required for each level of power dispatcher in the professional skill standard, the power grid dispatch related content is combed.

[0042] Step 2, constructing a dispatch knowledge and skill point network model, combining the dispatch regulation text data to form a dispatch knowledge base;

[0043] The specific steps for constructing the dispatch knowledge and skill point network model are as follows:

[0044] Step 2-1, according to the "Power Grid Dispatching Terminology Standard", the power industry vocabulary standard and the professional skill standard, form the key knowledge points and skill points.

[0045] Step 2-2, for the structured and semi-structured data of SCADA / EMS, OMS, PMS, and other systems, extract the knowledge and skill points contained in the device knowledge, power grid structure knowledge, and power grid business process, etc. according to the storage structure and relationship in the relational database, and build the association relationship between the knowledge and skill points to form the basic model of the dispatching knowledge and skill points network.

[0046] Step 2-3, for unstructured text data such as dispatching rules and regulations, technical specifications, management regulations, training specifications, and safety operation guidance, according to their text format and content characteristics, build their sub-knowledge and skill points and related relationships on the basis of the dispatching knowledge and skill points network model.

[0047] Step 2-4, use professional knowledge and basic knowledge textbooks such as power system analysis, power electronics technology, electrical engineering foundation, high voltage technology foundation, and electrical identification drawing foundation to extract basic concepts and principles related to dispatching knowledge points, and build relationships between knowledge points according to theoretical logic to complete the construction of the dispatching knowledge and skill points network model.

[0048] On this basis, further clarify the nodes and relationships in the knowledge and skill points network model. Each knowledge and skill point is a node in the network, which contains node type (knowledge point, skill point), node name, and node attribute (knowledge or skill field, such as power grid control, power grid abnormal handling, power grid operation, etc.; node occupation level, such as junior worker, intermediate worker, senior worker, technician, senior technician). For example, the "voltage abnormal handling method" node is a knowledge point, the node attribute field is power grid abnormal handling, and the node belongs to the occupation level of senior worker and technician. The "circuit breaker closing operation" node is a skill point, the node attribute field is power grid operation, and the node belongs to the occupation level of junior worker and intermediate worker. To facilitate adaptive learning navigation and recommendation, knowledge and skill points are mainly connected by three kinds of relationships, including inclusion relationship, leading relationship and related relationship.

[0049] 1) Inclusion relationship: indicates the subordinate relationship and parent-child relationship between knowledge and skill points and their child nodes. For example, the "circuit breaker operation" node is the parent node, which contains the child node "circuit breaker closing operation" and the child node "circuit breaker opening operation".

[0050] 2) Leading relationship: indicates the relationship between the prerequisite knowledge and skill points that must be understood before mastering a certain knowledge and skill point node. For example, the knowledge point "voltage abnormal handling precautions" is the prerequisite knowledge point of the knowledge point "voltage abnormal handling method".

[0051] 3) Correlation: It represents the relationship between knowledge and skill points with the same principle or the same result. For example, the knowledge point "mutual inductor" and the knowledge point "transformer" are connected by correlation because they have the same working principle. "Voltage collapse" and "frequency collapse" will cause a large-scale power system or local system blackout, so they are connected by correlation, which facilitates the extension and correlation learning of similar knowledge points by dispatchers.

[0052] Step 3, preprocessing the dispatch knowledge base to generate a dispatch examination point network model linked to key sentences.

[0053] The dispatch knowledge base is composed of a dispatch knowledge and skill point network model and dispatch procedure text data.

[0054] The specific steps of preprocessing the dispatch knowledge base to generate a dispatch examination point network model linked to key sentences are as follows:

[0055] Step 3-1, the dispatch knowledge and skill point network model is regarded as a list of triples. Preprocessing is to filter the knowledge point triples with questioning relationships to form an examination point network model, and delete unusable triples and skill points.

[0056] In this embodiment, the knowledge points with containing relationships in the dispatch knowledge and skill point network model and their subordinate "classification" relationship parallel knowledge points or "condition" relationship knowledge points and the knowledge points contained in the knowledge points (i.e. leaf knowledge points) are retained, and abstract knowledge and skill points that are difficult to quantify by computers are deleted. The leaf knowledge points of the knowledge points are divided into dispatch examination point network models required to be mastered by dispatchers of various levels according to the dispatch professional standard.

[0057] Step 3-2, preprocessing the dispatch procedure text data, dividing the dispatch procedures and other texts into key sentences and key paragraphs according to their semantics. Key sentences are mainly sentences containing facts and knowledge, and key paragraphs are multi-sentence description paragraphs of facts and related knowledge reasons, conditions, measures, operation steps, etc.

[0058] Step 3-3, jieba word segmentation is performed on the main sentences of key sentences and key paragraphs to segment out hot words.

[0059] Step 3-4, match the segmented hot words with the knowledge points in the dispatch examination point network model.

[0060] Step 3-5, link the key sentences and key paragraphs to the knowledge points with high matching degree as extension sentence nodes, and the relationship between the knowledge points is "mounting".

[0061] Step 3-6, use expert manual review technology to check the rationality of the linked key sentences and key paragraphs.

[0062] Step 3-7, generate a scheduling examination point network model of the hanging key sentence.

[0063] Step 4, based on the scheduling examination point network model of the hanging key sentence, extract relevant knowledge points and their hanging key sentences or key paragraphs, and adaptively generate test questions.

[0064] Step 4-1, adaptive generation of fill-in-the-blank question stems:

[0065] Fill-in-the-blank question stem types include: hollow description type and formula fill-in-the-blank type. This method can effectively evaluate the dispatcher's mastery of specific knowledge points.

[0066] In this embodiment, the specific generation method of the hollow description type stem is: by identifying the key sentences and key paragraphs hanging on the knowledge points in the scheduling examination point network model, the key matching items (such as knowledge points and key numbers) in them are hollowed out, thereby generating the stem.

[0067] The specific generation method of the formula fill-in-the-blank type stem is: by identifying the key sentences and key paragraphs hanging on the knowledge points, the variables or parameters in the formula involved in the key sentences / key paragraphs are hollowed out, thereby generating the stem.

[0068] In this embodiment, the difficulty of fill-in-the-blank questions lies in the excavation of the empty, which can generally consider using key sentence analysis. First, use jieba segmentation to identify whether it contains the name information of the key knowledge node or contains the key number to be used for hollowing. If it cannot be identified, use the TF-IDF text mining technology to find the three with higher frequency of key sentence occurrence as the hollowed-out options.

[0069] The specific implementation steps are as follows:

[0070] 1) First, give a key knowledge point K, based on the scheduling examination point network model, extract all the key sentences hanging on it to form a sentence pool.

[0071] 2) Randomly select a sentence S1 from the sentence pool as a stem option.

[0072] 3) Use the jieba segmentation method to segment S1.

[0073] 4) Perform segmentation matching to determine whether the key knowledge point name in S1 matches.

[0074] 5) If the segmentation matching is successful, the knowledge point name is empty, and step 9 is entered.

[0075] 6) If the segmentation matching fails, determine whether S1 contains a key number.

[0076] 7) If it contains, the key number is empty, and step 9 is entered.

[0077] 8)If not included, use the TF-IDF algorithm to analyze the weight of the hot words, take the top 3 words as empty.

[0078] 9) Remove the empty S1, as the stem, and the empty as the answer, fill in the blank question successfully.

[0079] Among them, TF-IDF is a common text mining technology, the full name of Term Frequency-Inverse Document Frequency (TF-IDF). It is a method for evaluating the importance of keywords in a document, commonly used in text classification, information retrieval and other fields. TF-IDF represents a document as a vector, where each element of the vector corresponds to a word. The value of the element represents the importance of the word in the document. The importance is calculated based on two indicators: the frequency of the word in the document (TF) and the number of documents in the document set that the word appears in (IDF).

[0080] TF (Term Frequency) represents the frequency of a word in a text. This number is usually normalized (usually word frequency divided by the total number of words in the article) to prevent it from being biased towards long files (the same word may have a higher word frequency in long files than in short files, regardless of the importance of the word). TF is represented by the following formula:

[0081] ;

[0082] where, represents the number of times a word appears in a document , is the frequency of the word in the document . However, some common words do not contribute to the test, and some words with low frequency may express the theme of the text, so using TF alone is not appropriate. The design of the weight must meet the following conditions: the stronger the word's ability to predict the theme, the greater the weight, and vice versa. For example, when performing text statistics, some words only appear in a few articles in the entire corpus. Such words have a great impact on the theme of the article, so the weight of these words should be designed to be larger. IDF is designed to do this.

[0083] IDF (Inverse Document Frequency) represents the prevalence of a keyword. This requires a corpus to simulate the language's usage environment. A higher IDF indicates that the term has good category discrimination ability, as fewer documents contain a particular term. The IDF of a specific term can be obtained by dividing the total number of documents by the number of documents containing that term, and then taking the logarithm of the quotient.

[0084] ;

[0085] in, Indicates the total number of texts. Indicates that the term is included. The number of documents.

[0086] TF-IDF refers to the high frequency of a word within a specific text and its low frequency across the entire corpus, resulting in a high-weighted TF-IDF. Therefore, TF-IDF tends to filter out common words and retain important ones. Its expression is as follows:

[0087] ;

[0088] Step 4-2: Adaptive generation of true / false question stems:

[0089] True / False questions are mainly used to test the dispatcher's understanding and cognition of a certain knowledge point. The question stem directly quotes the factual key statements / key paragraphs attached to the network model of the scheduling test point.

[0090] Meanwhile, to increase the number, complexity, and confusion of the questions, a distractor technique was employed, replacing key numbers, contrasting words, and affirmative and negative words to generate pseudo-propositions. These can be mainly categorized into the following five types:

[0091] True / False Questions: Directly judge the truthfulness of the statement or the conformity of the conditions.

[0092] Comparison and judgment questions: Compare the characteristics or applicable scenarios of two or more knowledge points.

[0093] Cause-and-effect judgment questions: assess the causal relationships between knowledge points.

[0094] Definition judgment question: Determine whether the statement accurately defines a certain knowledge point.

[0095] Operational judgment questions: These questions test the correctness of the operational steps.

[0096] Step 4-3: Adaptive generation of multiple-choice question stems:

[0097] Multiple choice questions are divided into two categories: single knowledge point identification and cross-knowledge point comprehensive comparison.

[0098] Step 4-3-1, Single Knowledge Point Discrimination Multiple Choice Questions, which examines the understanding and discrimination of a single knowledge point. Common question types can be divided into 6 categories according to the label types of knowledge points in the dispatch knowledge skill point network model (reason, condition, effect, processing method, application scenario, definition, phenomenon, etc.). The options are statements / words related to this knowledge point, and single or multiple choice questions can be generated.

[0099] In this embodiment, based on the key paragraphs mounted in the dispatch examination point model, the stem can be generated based on the main sentences in the paragraphs, and the options use the remaining conditional statements.

[0100] For example, "The conditions of a certain knowledge point include ()", and the options are the remaining conditional statements, which are directly used as the answers to multiple choice questions. For the structure of the examination point network model, the stem can be generated according to the association relationship, such as "A certain knowledge point includes ()", and the options are the knowledge point set of the next layer of the knowledge point, forming a multiple choice question.

[0101] Step 4-3-2, Cross-Knowledge Point Comprehensive Discrimination Multiple Choice Questions, which examines the discrimination between multiple knowledge points. The knowledge points involved in the options have certain commonalities such as "pioneer", "contain", "related" relationships, and the stem describes the related content of a certain knowledge point in the options. Therefore, multiple knowledge point discrimination type questions not only need to design four reasonable options, but also need to design appropriate stems so that only one option meets the description of the stem.

[0102] (1) Knowledge extraction: Extract relevant knowledge points and their association relationships, characteristic descriptions, and conditional descriptions from the dispatch examination point network model, and link the key sentences / paragraphs.

[0103] (2) Stem design: Generate the stem based on the extracted information to ensure that the stem accurately reflects the knowledge point and its discrimination content being examined.

[0104] (3) Option generation: Design options, including correct answers and distractors. Ensure that each option is related to the knowledge point described in the stem, but only one option is completely correct, and the distractors may be incorrect or partially correct. They can be generated by introducing similar but incorrect knowledge points or characteristics to increase the complexity and confusion of the question. These options can ensure in-depth examination of the examinees' understanding and discrimination ability of the knowledge points by involving "pioneer", "contain", "related" relationships, etc.

[0105] (4) Verification and optimization: Verify the generated questions to ensure their reasonableness and scientificity, and make necessary adjustments. Ensure that each option is reasonable in terms of semantics and logic, and has some relevance to the knowledge point described in the stem. At the same time, it is necessary to avoid designing obviously interfering options to increase the difficulty and challenge of the question.

[0106] Through these methods, the present application can automatically generate a rich and diverse set of evaluation questions, thereby improving the comprehensiveness and accuracy of knowledge evaluation.

[0107] Step 5, generating the interference item of the evaluation question according to the knowledge point association relationship of the scheduling test point network model.

[0108] In evaluating the quality of the question, the role of interference factors is crucial, especially in the design of true-false questions and multiple-choice questions. The quality of the interference item directly affects the difficulty and effectiveness of the question. If the interference item is not enough to confuse the dispatcher, they may easily identify the correct answer, which will seriously reduce the quality and reliability of the question.

[0109] The present application adopts the knowledge point association relationship based on the scheduling test point network model, designs a new pseudo-proposition construction method, which is specifically designed for true-false questions and multiple-choice questions, including key number replacement, comparison word replacement, and positive and negative word replacement, aiming to more accurately generate interference items that can effectively confuse dispatchers. Through this association relationship-based strategy, we can improve the quality and educational value of the question, not only testing the dispatcher's mastery of basic knowledge, but also assessing their understanding of the complex relationships between knowledge points. This method not only makes the question more challenging, but also helps dispatchers to think and analyze more deeply when solving practical problems. As shown in Table 1.

[0110] Table 1 Pseudo-proposition construction method example

[0111]

[0112] Step 5-1, key number replacement

[0113] Key number replacement is a commonly used question generation technique, especially when numbers in a sentence serve as the main component. These main components can include subject-predicate structures, verb-object structures, and their modifiers. When performing key number replacement, we usually use similar numbers to generate pseudo-propositions, thereby increasing the diversity and difficulty of the question.

[0114] In this embodiment, the specific method is: using the jieba word segmentation tool to extract key numbers from sentences. To ensure the effectiveness and logic of number replacement, it is necessary to consider whether the stem content contains implicit information about the key number, and exclude it in advance.

[0115] For example, "There are three basic requirements for electrical main wiring design: reliability, flexibility, and economy", the size of the number can be summarized from the information in the second half of the sentence, which does not interfere with the dispatcher, and should be identified in advance and removed.

[0116] For the selection of replacement numbers, the frequency of the numbers appearing in the key sentences and key segments of all knowledge points in the knowledge points is counted. For ordinary numbers, the results of ±1, *(-1), *2, / 2 are confused and replaced. For percentages, ±5%, ±10%, ±20% are added or subtracted based on the original number, or the numbers with higher frequency are extracted from multiple related knowledge points as replacement items or other options of the selection questions.

[0117] However, in the case of numbers related to voltage levels and other limitations, such as "500kV terminal iron tower" and other limited numbers and specific modifiers "three-phase", they need to be identified in advance and replaced according to the corresponding voltage level list and phase sequence list. If you want to increase the difficulty, you can also replace it according to the corresponding table of base voltage and rated voltage. In the case of numbers related to frequency values, the results of ±0.2 and ±0.5 are used for replacement. Finally, some unreasonable restrictions on the topic are removed by experts.

[0118] Step 5-2, comparison word replacement.

[0119] Comparison word replacement plays an important role in proposition design. When the main components of the sentence in the examination point network model contain comparison words, we can consider using their antonyms to generate pseudo propositions. In the power industry, we first need to sort out common comparison word combinations for subsequent comparison word replacement of propositions.

[0120] Table 2 shows some common comparison words and their antonyms, which form a contrast relationship with each other. However, when dealing with propositions that contain multiple comparison words, simply replacing one comparison word may sometimes result in an unreasonable overall sentence description, such as "Although the effectiveness of backup capacity and emergency response mechanisms is equally important to device life and maintenance frequency, the stability and reliability of the power grid may differ under high and low load conditions." At this time, simply replacing "high load" with "low load" and "stability" with "reliability" will only make the sentence an incorrect and unreasonable sentence. In this case, it is recommended to use natural language analysis to handle both parallel relationships and comparison word replacement, or ultimately check by experts to ensure the logical and semantic accuracy of the proposition.

[0121] Table 2: Common comparison word table for power grid dispatching

[0122]

[0123] Step 5-3, Affirmative / Negative Word Replacement:

[0124] Affirmative / negative word replacement is a common pseudo-proposition construction scheme. If the sentence is in the negative form, it is changed to the affirmative form, and vice versa. First, the form of the proposition core structure needs to be determined through dependency syntax analysis. When the proposition changes from the affirmative form to the negative form, it needs to be determined what kind of negative word to add and how to add the negative word. When the proposition changes from the negative form to the affirmative form, all the adverbs related to the negative word need to be deleted. Part of the affirmative sentence and negative sentence construction results are shown in Table 3.

[0125] Table 3 Affirmative / negative replacement construction results

[0126]

[0127] Step 6, initial difficulty setting analysis of the question from the question type factor, knowledge point factor, and interference strategy factor, and then dynamic correction of the question difficulty according to the test results of the general audience.

[0128] When designing the difficulty of the question for the dispatch professional field, considering the business characteristics, for the newly generated question, which has not been answered by the students, the difficulty of the question needs to be estimated according to the question properties, and then adjusted according to the feedback of the student's usage information.

[0129] 1) Difficulty factor analysis

[0130] The factors affecting the difficulty of the question are complex and diverse. This project analyzes the initial difficulty setting of the question from three aspects of question type factor, knowledge point factor, and interference strategy factor.

[0131] The question type is mainly divided into fill-in-the-blank, judgment, and selection. Fill-in-the-blank requires students to have a high degree of accuracy in knowledge mastery, and can directly distinguish the exact content of the question; judgment only needs to judge whether the description is consistent with the actual situation; selection needs to distinguish the description of knowledge points related to the content or different knowledge points, the more the number of distinctions needed, the more difficult the question. Therefore, in general, from the question type, judgment is simpler than selection, single-choice is simpler than multiple-choice, and fill-in-the-blank is the most difficult.

[0132] The knowledge point factor mainly considers the type and difficulty of the knowledge points covered in the question, as well as the number of knowledge points. Different knowledge points have different levels of difficulty. For some conceptual knowledge points, which do not require much reasoning or deduction, the difficulty is relatively low, and questions involving only these types of knowledge points are relatively simple. However, for some causal or measure-based knowledge points, which are relatively complex and require reasoning and deduction to understand the underlying principles, the difficulty is relatively high, and questions involving these types of knowledge points are naturally more difficult. Therefore, the difficulty of the question is positively correlated with the difficulty of the knowledge points. Secondly, the number of knowledge points covered in the question also affects the difficulty of the question. The more knowledge points a question covers, the more difficult it is, requiring the identification and judgment of multiple knowledge points.

[0133] Distractor strategies mainly include key number substitution, contrast word substitution, and affirmative / negative word substitution. Contrast word substitution and affirmative / negative word substitution primarily test the learner's ability to analyze knowledge descriptions. Key number substitution, however, is more disruptive, requiring learners to clearly identify the actual time of application, the correct frequency range, the correct voltage range, or the correct voltage level within specific concepts. This may be more difficult than the other two strategies.

[0134] 2) Difficulty Quantification Assessment Design

[0135] Based on the above summary of factors influencing difficulty, these factors are categorized into different levels according to their degree of influence and then quantified. This project uses a rating system for quantification. For example, the factor of question type is assigned values ​​of 1, 3, 5, 3, and 5 for "True / False Questions," "Single Knowledge Analytical Multiple Choice Questions," "Multiple Knowledge Point Analytical Multiple Choice Questions," and "Fill-in-the-Blank Questions," respectively. The quantification of other influencing factors is shown in Table 4.

[0136] Table 4 Quantification of Influencing Factors

[0137]

[0138] The initial total difficulty of the problem can then be measured by a weighted average of the factors influencing difficulty. The formula is as follows:

[0139] ;

[0140] Where d represents the initial difficulty value of the question, d i represents the difficulty value of the i-th difficulty factor, and n represents the total number of difficulty factors.

[0141] Based on the values ​​assigned in the influencing factor quantification table, where 1 represents easy, 3 represents medium, and 5 represents difficult. 2 and 4 represent adjacent judgments, being midpoints between 1-3 and 3-5, respectively, indicating medium-to-easy and medium-to-difficulty. When d is a score, the set of files represented by its closest integer digit is considered.

[0142] After the initial difficulty assessment of the question, it is necessary to dynamically correct the difficulty of the question according to the test results of the broad audience. Considering adopting the classical test theory, the answer conditions of all students N to the question are counted, the number of correct answers is R, and the probability of correct answer is calculated through the formula:

[0143]

[0144] The difficulty of the question is updated according to the real learning and answering conditions of the students. Assuming that the preset difficulty of the question is d, when the probability of the student answering the question correctly is greater than 75%, the difficulty of the question d is reduced by one level and set to d-1, when d is 1, the difficulty of the question is unchanged. When the probability of the student answering the question correctly is less than 25%, the difficulty of the question d is increased by one level and set to d+1, when d is 5, the most difficult, the difficulty of the question is unchanged. But when the probability of the student answering the question correctly is between 25% and 75%, it is considered that the preset difficulty of the question is reasonable, and it is not adjusted temporarily.

[0145] The difficulty of the question is an index for measuring the difficulty of the question in the standardized test system according to the provisions of the teaching goal, which reflects the characteristics of the question itself (such as the type of the question, the content of the question, etc.), and is also a key parameter in the intelligent teaching system. Determining the difficulty level of the question can reasonably recommend adaptive questions to dispatchers with different knowledge and skill levels. The difficulty design of the question can effectively support multiple intelligent functions such as automatic generation of questions and personalized question recommendation.

[0146] Embodiment two

[0147] The embodiment discloses a power grid dispatching adaptive evaluation question generation system based on a semantic model, comprising:

[0148] A semantic model module is configured to obtain dispatching regulation text data;

[0149] A dispatching knowledge and skill point network model is constructed, and the dispatching regulation text data is combined to form a dispatching knowledge base;

[0150] The dispatching knowledge base is preprocessed to generate a dispatching examination point network model connected with key sentences;

[0151] An adaptive test question stem generation module is configured to extract relevant knowledge points and their connected key sentences or key paragraphs based on the dispatching examination point network model connected with key sentences, and adaptively generate an evaluation question stem;

[0152] An interference term generation module is configured to generate an evaluation question interference term according to the knowledge point association relationship of the dispatching examination point network model;

[0153] ​The question quality evaluation module is configured to perform initial difficulty setting analysis on the question, and to perform dynamic correction on the question difficulty according to the test results of the trainees.

[0154] Embodiment three

[0155] The embodiment aims to provide a computer device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the program.

[0156] Embodiment four

[0157] The embodiment aims to provide a computer readable storage medium.

[0158] A computer readable storage medium, which stores a computer program, wherein the program is executed by a processor to perform the steps of the above method.

[0159] The steps involved in the device of the above embodiment correspond to the method embodiment one, and the specific embodiments can refer to the relevant description part of embodiment one. The term "computer readable storage medium" should be understood as including a single medium or multiple media of one or more instruction sets; it should also be understood as including any medium capable of storing, encoding or carrying instruction sets for execution by a processor and causing the processor to perform any method in the present application.

[0160] Those skilled in the art should understand that the above modules or steps of the present application can be realized by a general computer device, alternatively, they can be realized by program codes executable by a computing device, so that they can be stored in a storage device for execution by a computing device, or they can be made into individual integrated circuit modules, or a plurality of modules or steps among them can be made into a single integrated circuit module. The present application is not limited to any specific combination of hardware and software.

[0161] The above describes the specific embodiments of the present application in combination with the accompanying drawings, but is not a limitation on the protection scope of the present application. Those skilled in the art should understand that various modifications or variations made by those skilled in the art on the basis of the technical solutions of the present application without creative labor are still within the protection scope of the present application.

Claims

1. A power grid dispatching adaptive assessment question generation method based on a semantic model, characterized in that, The method comprises the following steps: acquiring dispatching regulation text data; constructing a dispatching knowledge skill point network model, combining the dispatching regulation text data to form a dispatching knowledge base; preprocessing the dispatching knowledge base to generate a dispatching examination point network model with key sentences connected; based on the dispatching examination point network model with key sentences connected, extracting relevant knowledge points and key sentences or key paragraphs connected thereto, and adaptively generating a test question stem; generating test question distractors according to the knowledge point association relationship of the dispatching examination point network model; initially setting the difficulty of the question and dynamically correcting the difficulty of the question according to the test results of the students; the specific steps of preprocessing the dispatching knowledge base to generate a dispatching examination point network model with key sentences connected are as follows: the dispatching knowledge base is composed of a dispatching knowledge skill point network model and dispatching regulation text data; the dispatching knowledge skill point network model is preprocessed to obtain a dispatching examination point network model; the specific steps of constructing a dispatching knowledge skill point network model are as follows: according to the power grid dispatching terminology specification, the power industry vocabulary standard and the professional skill standard, key knowledge points and skill points are formed; for the structured and semi-structured data of SCADA / EMS, OMS, PMS and signal protection systems, knowledge skill points contained in equipment knowledge, power grid structure knowledge and power grid business process content are extracted, the association relationship between knowledge skill points is constructed according to the storage structure and relationship in the relational database, and a dispatching knowledge skill point network basic model is formed; for the unstructured text data of dispatching rules and regulations, technical specifications, management regulations, training specifications and safety operation instruction books, according to the text format and content characteristics, the sub-knowledge skill points and related relationships are hierarchically constructed on the basis of the dispatching knowledge skill point network basic model; basic concept and principle knowledge points related to dispatching knowledge points are extracted by using the professional knowledge and basic knowledge textbooks of power system analysis, power electronics technology, electrical engineering foundation, high voltage technology foundation and electrical identification drawing foundation, and the relationship between knowledge points is constructed according to the theoretical logic relationship, so as to complete the construction of the dispatching knowledge skill point network model; each knowledge skill point is a node in the network, which contains node type, node name and node attribute; the knowledge skill points are mainly connected by three kinds of relationships, including inclusion relationship, leading relationship and correlation relationship; 1) inclusion relationship: indicating the parent-child relationship and subordinate relationship between knowledge skill points and their sub-nodes; 2) leading relationship: indicating the relationship between prerequisite knowledge skill points and knowledge skill points that must be understood before mastering the knowledge skill points; 3) correlation relationship: indicating the relationship between knowledge skill points with the same principle or the same result; self-adaptive generation of fill-in-the-blank question stems: the fill-in-the-blank question stem types include description type and formula fill-in-the-blank type; wherein, the specific generation method of the description type stem is as follows: by identifying the key sentences and key paragraphs connected to the knowledge points in the dispatching examination point network model, the key matching items are blanked out to generate the stem; The specific method for generating formula-filling questions is as follows: by identifying key statements and key paragraphs connected to knowledge points, the variables or parameters in the formulas involved in the key statements / key paragraphs are removed to generate the question stem; By analyzing key statements, firstly, jieba word segmentation is used to identify whether they contain the names of key knowledge nodes or key numbers for data extraction. If they cannot be identified, TF-IDF text mining technology is used to find the three most frequently occurring key statements as data extraction options. The specific implementation steps are as follows: 1) First, given a key knowledge point K, based on the scheduling test point network model, extract all the key statements attached to it to form a statement pool. 2) Randomly select statement S1 from the statement pool as the option in the question stem; 3) Use the jieba word segmentation method to segment S1; 4) Perform word segmentation and matching to determine whether the key knowledge point name matches in S1; 5) If word segmentation matching is successful, leave the knowledge point name as empty and proceed to step 9). 6) If word segmentation matching fails, determine whether S1 contains key numbers; 7) If it is contained, treat the key number as empty and proceed to step 9). 8) If not included, use the TF-IDF algorithm to perform weight analysis on the separated hot words, and take the 3 words with the highest weight as empty; 9) Remove the blank from S1 to form the question stem, and use the blank as the answer. The fill-in-the-blank question is now successfully created. True / False questions and their stems are generated adaptively. The true / false question stem directly quotes key factual statements / paragraphs attached to the network model of scheduling test points; Simultaneously, distractor techniques are employed to replace key numbers, contrasting words, and affirmative and negative words to generate pseudo-propositions; these are categorized into the following five types: True / False Questions: Directly judge the truthfulness of the statement or the fulfillment of the conditions; Comparison and judgment questions: Compare the characteristics or applicable scenarios of two or more knowledge points; Cause-and-effect judgment questions: assess the causal relationships between knowledge points; Definition judgment questions: Determine whether a statement accurately defines a certain knowledge point; Operational judgment questions: These questions test the correctness of the operational steps. Multiple-choice question stems are generated adaptively: Multiple choice questions are divided into two categories: single knowledge point identification and cross-knowledge point comprehensive comparison. Among them, the single knowledge point identification type multiple-choice questions are divided into 6 categories according to the label type of the knowledge point in the scheduling knowledge and skill point network model; the options are the number of sentences / words related to this knowledge point, generating single-choice or multiple-choice questions. For multiple-choice questions that require comprehensive analysis across multiple knowledge points, the knowledge points involved in the options have "leading", "inclusion", or "related" relationships, and the question stem describes the relevant content of a certain knowledge point in the options. Therefore, multiple knowledge point analysis questions not only require designing four reasonable options, but also designing a suitable question stem so that only one option satisfies the description in the question stem. (1) Knowledge extraction: Extract relevant knowledge points and their relationships, characteristic descriptions and condition descriptions from the scheduling test point network model, and connect them with key statements / key paragraphs; (2) Question design: Generate question stems based on the extracted information to ensure that the question stems accurately reflect the knowledge points being tested and the content to be distinguished; (3) Option generation: design options, including the correct answer and distractors; ensure that each option is related to the knowledge point described in the stem, but only one option is completely correct, and the distractors are incorrect or partially correct, generated by introducing similar but incorrect knowledge points or characteristics, increasing the complexity and confusion of the question; these options ensure in-depth examination of the examinee's understanding and discrimination ability of the knowledge points through the "pilot", "contains", "related" relationship; (4) Verification and optimization: verify the generated questions and ensure the reasonableness and scientificity of the questions and options, and optimize and adjust; ensure that each option is semantically and logically reasonable and has certain relevance with the knowledge point described in the stem; At the same time, avoid designing too obvious distractors to increase the difficulty and challenge of the question; The dispatch regulation text is divided into key sentences and key paragraphs according to its semantics; jieba segmentation is performed on the main sentences of the key sentences and key paragraphs to segment the hot words; Match the segmented hot words with the knowledge points in the dispatch test point network model; Hang the key sentences and key paragraphs to the knowledge points with high matching degree as extended sentence nodes; Reasonably check the hung key sentences and key paragraphs by expert manual checking technology; Generate a dispatch test point network model with hung key sentences; The adaptive generation of multiple-choice question stems is based on the key paragraphs hung in the dispatch test point model. The stem is generated based on the main sentences in the paragraph, and the options use the remaining conditional sentences; The interference item of the evaluation question includes key number replacement: first, extract the key numbers by using the jieba segmentation tool to segment the sentence; For the selection of replacement numbers, count the frequency of numbers appearing in the key sentences and key paragraphs hung in all knowledge points above and below the knowledge points in the test point, and extract the numbers with high frequency from multiple related knowledge points as replacement items or other options of the selection question; Based on the dispatch test point network model with hung key sentences, jieba segmentation is used to analyze the key sentences and identify whether they contain the name information of the key knowledge nodes or contain the key numbers used for fill-in-the-blank questions; Use TF-IDF text mining technology to find three key sentences with high frequency as blank options; TF: ； wherein, denotes a term in a document occurs, is a term in a document occurs; IDF: ； wherein, represents the number of all texts, represents the number of documents containing the word item ; TF-IDF: ； The interference item of the evaluation question mainly includes key number replacement, comparison word replacement, and positive and negative word replacement; Comparison word replacement: Positive and negative word replacement: 。 2. The adaptive assessment question generation method for power grid dispatching based on a semantic model according to claim 1, characterized in that, Preprocess the dispatch knowledge base, specifically: retain the knowledge points with containing relationship in the dispatch knowledge and skill point network model, and their subordinate "classification" relationship parallel knowledge points, or "condition" relationship knowledge points and leaf knowledge points contained in the knowledge points, and delete abstract knowledge and skill points that are difficult for computers to quantify, divide the leaf knowledge points of the knowledge points into dispatch test point network models required by dispatchers at all levels according to the dispatch professional standard; The dispatch regulation text is divided into key sentences and key paragraphs according to its semantics.

3. The adaptive assessment question generation method for power grid dispatching based on a semantic model according to claim 1, characterized in that, The initial difficulty setting analysis of the question is specifically performed by using a grade assignment method to quantify the difficulty influencing factors, and the initial total difficulty of the question is measured by weighted average of the difficulty influencing factors.

4. The adaptive assessment question generation method for power grid dispatch based on semantic model according to claim 1, characterized in that, The difficulty of the question is dynamically corrected according to the test results of the wide range of students, and the difficulty of the question is updated according to the real learning and answering conditions of the students by using the classical test theory.

5. A power grid dispatching adaptive assessment question generation system based on a semantic model, based on the power grid dispatching adaptive assessment question generation method based on a semantic model in any one of claims 1-4, characterized in that, It comprises: The semantic model module is configured to obtain dispatching regulation text data; A dispatching knowledge skill point network model is constructed, and a dispatching knowledge base is formed in combination with the dispatching regulation text data; The dispatching knowledge base is preprocessed to generate a dispatching examination point network model with key sentences connected; The adaptive test question stem generation module is configured to extract relevant knowledge points and key sentences or key paragraphs connected thereto based on the dispatching examination point network model with key sentences connected, and to adaptively generate a test question stem; The interference term generation module is configured to generate test question interference terms according to the knowledge point association relationship of the dispatching examination point network model; The question quality evaluation module is configured to perform initial difficulty setting analysis of the question, and to dynamically correct the difficulty of the question according to the test results of the students; The dispatching knowledge base is preprocessed to generate a dispatching examination point network model with key sentences connected, and the specific steps are as follows: The dispatching knowledge base is composed of a dispatching knowledge skill point network model and dispatching regulation text data; The dispatching knowledge skill point network model is preprocessed to obtain a dispatching examination point network model; The dispatching regulation text is divided into key sentences and key paragraphs according to its semantics; The key sentences and key paragraph main sentences are subjected to jieba word segmentation to cut out hot words therein; The cut-out hot words are matched with the knowledge points in the dispatching examination point network model; The key sentences and key paragraphs are connected to the knowledge points with high matching degrees as extended sentence nodes; The connected key sentences and key paragraphs are subjected to rationality verification by using an expert artificial review technique; A dispatching examination point network model with key sentences connected is generated; The adaptive generation of the multiple-choice question stem is based on the key paragraphs mounted in the dispatching examination point model, the stem is generated based on the main sentences in the paragraphs, and the options use the remaining conditional sentence; The test question interference terms include key number replacement: key numbers are extracted by using the jieba word segmentation tool to segment the sentences; for the selection of the replacement numbers, the frequencies of the numbers appearing in the key sentences and key paragraphs connected to all knowledge points above and below the knowledge point are counted, and the numbers with high frequencies are extracted from the multiple related knowledge points as replacement items or other options of the multiple-choice questions; Based on the dispatching examination point network model with key sentences connected, the key sentences are analyzed by using the jieba word segmentation to identify whether the name information of the key knowledge nodes or the key numbers are contained for use in the blank filling; TF-IDF text mining technology is used to find three key sentences with high frequencies as blank filling options; TF: ； wherein, denotes a term in a document occurs, is a term in a document occurs; IDF: ； wherein, denotes the number of all texts, denotes the number of documents containing the term . TF-IDF: ； The test question interference terms mainly include key number replacement, comparison word replacement, and positive and negative word replacement; Comparison word replacement: Positive and negative word replacements: 。 6. A computer apparatus comprising a memory, a processor, and a computer program stored on the memory and running on the processor, characterized in that, The processor, when executing the program, implements the steps of the method of any one of claims 1-4.

7. A computer-readable storage medium having stored thereon a computer program, characterized in that, The program, when executed by the processor, performs the steps of the method of any one of claims 1-4.

Images