Virtual classroom interaction method and system based on knowledge representation and reasoning

Abstract

The invention provides a virtual classroom interaction method and system based on knowledge representation and reasoning, and the method comprises the steps: obtaining a student interaction data flow in a power system training virtual classroom, carrying out the knowledge granularity decomposition of the student interaction data flow, and generating a knowledge tuple set containing concept nodes and relation edges, and inputting the knowledge tuple set into an inference engine to execute multi-dimensional inference processing, generating a student knowledge mastering state vector and a learning demand prediction vector, executing interaction content generation processing based on the knowledge mastering state vector and the learning demand prediction vector, generating classroom interaction response data with adaptability, and sending the classroom interaction response data to the student. And feeding back the classroom interaction response data to a virtual classroom interaction interface to form a dynamic interaction process of the student learning state and classroom feedback. According to the invention, the pertinence and practicability of virtual classroom interaction are effectively enhanced, so that the learning effect of students in power system training is improved.

Description

Technical Field

[0001] The present invention relates to the field of virtual classrooms, and in particular to a virtual classroom interaction method and system based on knowledge representation and reasoning. Background Art

[0002] With the development of virtual teaching technology, virtual classrooms are gradually being introduced in power system training to improve the efficiency of knowledge transfer and skills training. These methods simulate power system operation scenarios in a digital interactive environment, providing students with an online learning platform for knowledge presentation, simulated operations, and question-and-answer sessions. Currently, virtual classroom interaction methods typically respond to students' immediate questions or pre-set course content. For example, they return fixed knowledge points based on student-entered questions or push standardized operation demonstrations according to the course progress. However, power system knowledge is highly specialized (e.g., equipment principles and troubleshooting must be strictly linked), operational specifications are highly constrained (e.g., simulation experiments must follow specific procedures), and the learning process is highly logical (e.g., the progression from equipment fundamentals to operational characteristics). Existing interaction methods focus only on a single dimension of information (e.g., question content or operation results), failing to comprehensively analyze students' knowledge mastery, operational standards, and logical coherence. This results in a mismatch between interactive content and students' actual learning needs, hindering effective improvement in the relevance and learning outcomes of power system training. Therefore, developing virtual classroom interaction methods that accurately capture students' learning status and dynamically adapt interactive content has become a research priority in the digital transformation of power system training. Summary of the Invention

[0003] The present invention provides a virtual classroom interaction method and system based on knowledge representation and reasoning.

[0004] In a first aspect, an embodiment of the present invention provides a virtual classroom interaction method based on knowledge representation and reasoning, the method comprising: Obtaining a student interaction data stream in a power system training virtual classroom, wherein the student interaction data stream includes time-sequentially arranged question texts, simulation operation records, and historical conversation logs; Performing knowledge granularity decomposition processing on the student interaction data stream to generate a knowledge tuple set including concept nodes and relationship edges, wherein the concept nodes correspond to basic concepts of the power system, and the relationship edges reflect the logical associations between concepts and the contextual dependencies of student interaction behaviors; Input the knowledge tuple set into the inference engine to perform multi-dimensional inference processing to generate a student knowledge mastery state vector and a learning needs prediction vector. The knowledge mastery state vector includes indicators of concept understanding depth, operational standardization, and logical coherence, and the learning needs prediction vector includes content supplementation direction and ability improvement direction. Performing interactive content generation processing based on the knowledge mastery state vector and the learning demand prediction vector to generate adaptive classroom interactive response data, wherein the interactive response data includes knowledge completion information and ability training tasks; The classroom interaction response data is fed back to the virtual classroom interaction interface to form a dynamic interaction process between students' learning status and classroom feedback.

[0005] In a second aspect, an embodiment of the present invention provides a computer system, including: a memory storing a computer program; The processor is used to load the computer program to implement the virtual classroom interaction method based on knowledge representation and reasoning as described above.

[0006] The virtual classroom interaction method based on knowledge representation and reasoning provided by the present invention can comprehensively capture the students' knowledge expression, operation behavior and dialogue logic characteristics in the learning process by obtaining the student interaction data stream (including question text, simulation operation records and historical dialogue logs) in the power system training virtual classroom; the student interaction data stream is subjected to knowledge granularity decomposition processing to generate a knowledge tuple set containing concept nodes and relationship edges, wherein the concept nodes correspond to the basic concepts of the power system, and the relationship edges reflect the logical association between concepts and the context dependency of the students' interaction behavior. This processing method deeply integrates the inherent logic of the power system knowledge with the dynamic interaction characteristics of the students, forming an association structure between knowledge content and behavior patterns, and providing a structured information basis for subsequent analysis; the knowledge tuple set is input into the inference engine to perform multi-dimensional reasoning processing to generate the student knowledge tuple set. The knowledge mastery state vector and the learning demand prediction vector are combined, where the knowledge mastery state vector includes the depth of concept understanding, operational standardization and logical coherence indicators, and the learning demand prediction vector includes the content supplement direction and ability improvement direction. This multi-dimensional state modeling can comprehensively characterize the students' knowledge mastery level in power system training, avoiding the one-sidedness of focusing on a single dimension; based on the knowledge mastery state vector and the learning demand prediction vector, adaptive classroom interaction response data (including knowledge completion information and ability training tasks) are generated, and fed back to the virtual classroom interaction interface to form a dynamic interaction process, so that the interaction content can be dynamically adjusted according to the students' current learning status, which not only meets the needs of knowledge supplementation, but also fits the direction of ability improvement, effectively enhancing the pertinence and practicality of virtual classroom interaction, thereby improving students' learning effect in power system training. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 This is a flowchart of a virtual classroom interaction method based on knowledge representation and reasoning provided by an embodiment of the present invention.

[0008] Figure 2It is a schematic diagram of the composition of a computer system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0009] See also Figure 1 , Figure 1 A flowchart of a virtual classroom interaction method based on knowledge representation and reasoning provided by an embodiment of the present invention. The method can be executed by a computer system, including the following steps: Step S100: obtaining a student interaction data stream in a power system training virtual classroom, where the student interaction data stream includes time-sequentially arranged question texts, simulation operation records, and historical conversation logs.

[0010] The student interaction data stream is a collection of data generated during the interaction between students and the system in the power system training virtual classroom. This data is arranged in chronological order and can reflect the various behaviors and needs of students during the learning process. Question texts are questions raised by students in the virtual classroom. These questions often focus on relevant knowledge about power systems and include information such as equipment names, operating parameters, and fault phenomena. Simulation operation records are records of students' operations in the simulation environment of the virtual classroom, including equipment selection actions, parameter adjustment actions, and fault simulation actions. They record the process and steps of students' actual operations and reflect their practical ability and operating standards. Historical dialogue logs are records of past dialogues between students and the system, including student questions and system feedback. By analyzing these logs, we can understand the progressive process of students' knowledge and the trajectory of deepening their problems.

[0011] Capturing student interaction data streams can be achieved by setting up corresponding data acquisition modules within the virtual classroom system. For example, for question text, a data monitoring program can be set up in the student's question input box. When the student enters and submits the question, the program automatically saves the question text to a designated data storage location. For simulation operation records, the simulation system itself records each student operation and the corresponding timestamp. Arranging these records in chronological order yields the simulation operation log. For historical conversation logs, the system saves the content of each student question and system feedback, forming a conversation turn. Arranging these conversation turns in chronological order yields the historical conversation log.

[0012] Step S200: Perform knowledge granularity decomposition on the student interaction data stream to generate a knowledge tuple set containing concept nodes and relationship edges. The concept nodes correspond to basic concepts of the power system, and the relationship edges reflect the logical association between concepts and the contextual dependency of student interaction behaviors.

[0013] Knowledge granularity decomposition involves refining and analyzing the information in the student interaction data stream, breaking it down into smaller knowledge units and identifying the relationships between these units. Concept nodes are the basic elements in a knowledge tuple set, corresponding to fundamental concepts of the power system, such as generators, transformers, and power loads. These concepts are core components of the power system knowledge system. Relationship edges connect concept nodes, reflecting the logical connections between concepts, such as the binding relationship between devices and attributes and the triggering relationship between devices and behaviors. They also reflect the contextual dependencies of student interactions, such as the semantic progression between subsequent questions and previous questions in a conversation. The process of generating a knowledge tuple set requires comprehensive consideration of the question text, simulation operation records, and historical conversation logs in the student interaction data stream. By processing this data, key concepts and their relationships are extracted and organized into a directed graph, where nodes are concept nodes and edges are relationship edges, forming a knowledge tuple set. This knowledge tuple set more clearly demonstrates the knowledge structure and interaction logic involved in the student's learning process.

[0014] As an implementation manner, step S200 can be specifically implemented as the following steps S210 to S260: Step S210: semantic segmentation is performed on the question text in the student interaction data stream to extract semantic segments containing power system terms, where the semantic segments include equipment names, operating parameters, and fault phenomena.

[0015] Semantic segmentation analyzes the question text and breaks it down into semantically specific segments. These segments contain key terms and information about the power system. Equipment names are the specific names of various devices in the power system, such as generators, circuit breakers, and busbars. They are the physical components of the power system. Operating parameters are indicators that describe the operating status of the equipment, such as voltage, current, and power, reflecting the equipment's working status and performance. Fault phenomena are the various abnormal conditions exhibited when equipment fails, such as short circuits, overloads, and power outages.

[0016] As an implementation manner, step S210 can be specifically implemented as the following steps S211 to S215: Step S211: Perform word segmentation and filtering processing on the question text, and retain professional vocabulary related to the power system training objectives as candidate terms.

[0017] Word segmentation and filtering involves segmenting the question text into words and selecting vocabulary relevant to the power system training objectives. Professional vocabulary, such as "transformer," "reactive power," and "relay protection," are terms with specific meanings within the power system field. These terms are crucial vehicles for power system knowledge. Candidate terms, the professional vocabulary that remains after word segmentation and filtering, serve as the foundation for subsequent semantic analysis.

[0018] Filtering can be performed using word segmentation tools such as NLTK (Natural Language Toolkit) or Jieba. Taking Jieba as an example, the question text is first input into the Jieba word segmentation function, which segments the text into individual words. The segmented words are then filtered using a predefined dictionary of power system professional vocabulary, retaining only those in the dictionary as candidate terms. For example, for the question text "What is the voltage of the transformer?", Jieba will segment it into "please ask," "transformer," "of," "voltage," "is," and "how much." After filtering through the dictionary, "transformer" and "voltage" are retained as candidate terms.

[0019] Step S212: Perform semantic role labeling on the candidate terms to identify the semantic function of each candidate term in the question. The semantic function includes subject function, attribute function, and behavior function, where the subject function corresponds to the device name, the attribute function corresponds to the operating parameters, and the behavior function corresponds to the protection mechanism.

[0020] Semantic role labeling involves determining the semantic role a candidate term plays in the question text, that is, its semantic function. The principal function refers to the fact that the candidate term represents specific equipment in the power system, such as generators and circuit breakers. These equipment are the basis for the operation of the power system. The attribute function refers to the fact that the candidate term describes the operating parameters of the equipment, such as voltage, current, and frequency. These parameters reflect the working status of the equipment. The behavioral function refers to the fact that the candidate term involves the protection mechanism or operational behavior of the power system, such as relay protection and switch operation. These functions ensure the safe operation of the power system.

[0021] Semantic role labeling can be performed using machine learning-based methods, such as the Conditional Random Field (CRF) model. First, a large amount of text data related to the power system is collected, and candidate terms are manually labeled with their semantic functions. This labeled data is then used to train the CRF model. After training, the candidate terms to be analyzed are input into the model, which outputs the semantic function of each candidate term. For example, for the candidate term "transformer," the model labels its semantic function as a subject function; for "voltage," the model labels its semantic function as an attribute function.

[0022] Step S213: performing adjacency recognition processing on candidate terms with semantic functional associations, and extracting direct semantic connections between terms. Direct semantic connections include binding relationships between devices and attributes, and triggering relationships between devices and behaviors.

[0023] Adjacency identification involves identifying direct connections between candidate terms with semantic functional associations. The device-attribute binding relationship refers to the correspondence between a device and the parameters describing its operating status. For example, the device-attribute binding relationship between "transformer" and "voltage" indicates that voltage is a parameter describing the transformer's operating status. The device-behavior triggering relationship refers to the fact that the device's operating status or operation triggers specific behaviors or protection mechanisms. For example, a "short-circuit fault" triggers a "relay protection action," demonstrating a triggering relationship between a device and a behavior.

[0024] Adjacency identification can be performed using either rule-based or machine learning methods. Rule-based methods determine adjacency between terms based on predefined rules. For example, consider the rule "If a term for a main function is immediately followed by a term for an attribute function, with no other key delimiters in between, then a device-attribute binding relationship is considered to exist between them." Machine learning methods can use a graph neural network (GNN) model, constructing graph-structured data using candidate terms as nodes and relationships between terms as edges. The GNN model then learns and infers the graph to identify direct semantic connections between terms.

[0025] Step S214: Context verification is performed on the direct semantic connection, and effective semantic connections that conform to the power system knowledge system are screened out in combination with the overall semantic orientation of the question text.

[0026] Contextual validation is a further examination and screening of the extracted direct semantic connections to ensure that they conform to the knowledge system of the power system and the overall semantics of the question text. Valid semantic connections are those that have practical meaning in the power system knowledge system and are consistent with the semantics of the question text. For example, in a question about transformer fault analysis, the connection between "transformer" and "short-circuit fault" is valid because transformers in power systems may suffer from short-circuit faults, and it is consistent with the semantics of the question. However, if an unreasonable connection occurs, such as the connection between "transformer" and "car engine failure", it does not conform to the power system knowledge system and the question semantics and needs to be screened.

[0027] Context verification can be achieved by querying the power system knowledge base. The knowledge base stores various knowledge and rules of the power system, such as the normal operating parameter range of the equipment, fault types and processing methods. The direct semantic connection is compared with the knowledge in the knowledge base. If the connection is consistent with the knowledge in the knowledge base and conforms to the overall semantic direction of the question text, the connection is considered valid; otherwise, it is filtered out. For example, for the connection between "transformer" and "voltage is too high", the normal voltage range of the transformer is queried in the knowledge base. If the voltage value mentioned in the current question exceeds this range and is consistent with the fault analysis semantics of the question, the connection is considered valid.

[0028] Step S215: combining the candidate terms and their corresponding valid semantic connections to generate semantic segments containing power system terms.

[0029] A semantic fragment is a knowledge unit composed of candidate terms and valid semantic connections between them. It fully expresses a specific concept or situation in the power system. By combining candidate terms with valid semantic connections, scattered knowledge elements can be integrated to form meaningful semantic information. For example, combining "transformer" and "voltage" with the binding relationship between equipment and attributes can produce a semantic fragment such as "transformer voltage," which clearly expresses a transformer operating parameter.

[0030] Semantic fragments can be generated using simple string concatenation and data structure combination. Candidate terms are arranged in the order they appear in the question text. Based on the type of valid semantic connection, corresponding connectives or symbols are added between the terms to form a complete semantic fragment. For example, for the device-attribute binding relationship of "transformer" and "voltage," the semantic fragment "transformer voltage" can be generated. Semantic fragments are then stored in a suitable data structure, such as a list or dictionary, for subsequent processing and use.

[0031] Step S220: performing behavior sequence decomposition processing on the simulation operation record to extract the correspondence between the operation action and the operation object. The correspondence includes the binding relationship between the device selection action and the target device, and the constraint relationship between the parameter adjustment action and the parameter range.

[0032] Behavior sequence decomposition is the process of analyzing and decomposing the operational behaviors in the simulation operation records, breaking them down into individual operational actions, and finding the corresponding relationship between these actions and the operational objects. Operational actions are specific behaviors performed by students in simulation operations, such as selecting equipment, adjusting parameters, simulating failures, etc. The operational object is the target of the operational action, such as specific equipment, parameters, etc. The binding relationship between the device selection action and the target device means that when students select a certain device for operation, it is clear that the device is the target of the operation. The constraint relationship between the parameter adjustment action and the parameter range means that when adjusting parameters, the parameter adjustment value must be within the specified range, otherwise the operation may be invalid or cause equipment failure.

[0033] The behavior sequence decomposition process can be achieved by analyzing the simulation operation records line by line. First, the operation records are arranged in chronological order, and then each operation action is parsed to determine its operation type and operation object. For example, for the operation record "The transformer device was selected at 10:00", the operation action can be determined as "Select device" and the operation object is "Transformer", and a binding relationship between the device selection action and the target device is established. For parameter adjustment actions, it is necessary to refer to the relevant standards of the power system and the parameter setting requirements of the equipment to determine the reasonable range of the parameters. For example, for the voltage adjustment of the transformer, its voltage value must be within the specified rated voltage range. When students perform voltage adjustment operations, they check whether the adjustment value is within this range, thereby establishing a constraint relationship between the parameter adjustment action and the parameter range.

[0034] As an implementation method, step S220 is to perform behavior sequence decomposition processing on the simulation operation record to extract the corresponding relationship between the operation action and the operation object. Specifically, the following steps S221 to S225 can be implemented: Step S221: performing time axis alignment processing on the simulation operation records to generate an action event list arranged in the order of operation time.

[0035] Timeline alignment sorts the simulation operation records by operation time, giving them a clear chronological order. The action event list treats each operation in the simulation operation record as an event and arranges them in chronological order. Timeline alignment allows students to clearly see the order of their operations, facilitating analysis of their logic and flow.

[0036] Timeline alignment can be performed using a sorting algorithm, such as quick sort or merge sort. First, extract the timestamp information for each operation from the simulation operation record. Then, use a sorting algorithm to sort these timestamps and arrange the corresponding operations in the order of the sorted timestamps to generate an action event list. For example, there are three operations in the simulation operation record: Operation A is executed at 10:05, Operation B is executed at 10:02, and Operation C is executed at 10:08. After sorting, the order of the action event list is Operation B, Operation A, and Operation C.

[0037] Step S222: performing type identification processing on each action event in the action event list to distinguish between device operation type actions, parameter adjustment type actions, and fault simulation type actions.

[0038] Type identification processing determines the type of each action event in the action event list. Equipment operation actions refer to operations such as selecting, starting, and stopping equipment in the power system, such as selecting a transformer and starting a generator. These operations directly affect the operating status of the equipment. Parameter adjustment actions refer to adjustments to the operating parameters of the equipment, such as voltage, current, and power. These operations change the operating status of the equipment. Fault simulation actions simulate fault conditions in the power system, such as short circuit faults and overload faults, to test students' fault handling capabilities.

[0039] Type identification can be achieved by analyzing the descriptive information of action events. For example, the action event "circuit breaker selected" can be identified as a device operation action; "transformer voltage adjusted to 10kV" can be identified as a parameter adjustment action; and "short circuit fault simulated" can be identified as a fault simulation action. Machine learning classification algorithms, such as support vector machines (SVMs), can also be used. First, a large number of action event samples are collected and manually labeled. This labeled data is then used to train the SVM model. After training, the action event to be classified is input into the model, which then outputs its type.

[0040] Step S223: Performing logic dependency analysis on adjacent action events to identify constraints imposed by the preceding action on the subsequent action. The constraints include that parameter adjustment cannot be performed if the device is not started and recovery operations cannot be performed if the fault is not cleared.

[0041] Logical dependency analysis involves identifying the logical relationships between adjacent action events and determining the constraints imposed by preceding actions on subsequent actions. In power system operations, many operations have sequential and logical relationships. For example, parameter adjustments can only be performed after equipment startup, and recovery operations can only be performed after a fault has been cleared. Identifying these constraints helps standardize students' operational processes and improve operational accuracy and safety.

[0042] Logical dependency analysis can be achieved by establishing a logical rule base for power system operations. The rule base stores the logical relationships and constraints between various operations. When analyzing adjacent action events, the rule base is queried to determine whether the preceding action meets the execution conditions of the subsequent action. For example, if the preceding action is "transformer selected but not started", and the subsequent action is "adjusting the voltage of the transformer", the query to the rule base shows that the parameters cannot be adjusted if the equipment is not started, so the subsequent action does not meet the execution conditions. A knowledge graph-based method can also be used to construct the equipment, operations, and rules of the power system into a knowledge graph, and identify logical dependencies by reasoning in the graph.

[0043] Step S224: Perform pattern abstraction processing on continuous action events with the same operation target, and extract recurring action combinations as operation sub-patterns. The operation sub-patterns include the basic operation chain of equipment startup-parameter setting-operation monitoring, and the fault processing chain of fault triggering-phenomenon observation-troubleshooting.

[0044] Pattern abstraction involves identifying repetitive action combinations with the same operational objectives from continuous action events and abstracting them into operational sub-patterns. An operational sub-pattern summarizes and generalizes a series of operational actions, reflecting the typical process of power system operation. The basic operational chain of equipment startup, parameter setting, and operational monitoring is a common operational sub-pattern, describing the basic process from equipment startup to normal operation. The fault handling chain of fault triggering, phenomenon observation, and troubleshooting is another important operational sub-pattern, reflecting the general steps for handling power system faults.

[0045] Sequence mining algorithms, such as the Apriori algorithm, can be used to abstract patterns. First, the action event list is grouped according to the operation goal. Then, the consecutive action events within each group are analyzed to identify recurring action combinations. For example, in a student's operation record, the action combination "start the generator - set the generator voltage - monitor the generator power" appears repeatedly. Using a sequence mining algorithm, this can be extracted as an operation sub-pattern.

[0046] Step S225: Integrate the type information of the action event, the constraints of the logical dependency, and the combination rules of the operation sub-mode to generate a corresponding relationship between the operation action and the operation object.

[0047] Integration processing combines the action event type information, the logical dependency constraints, and the combination rules of the operation sub-patterns to form a correspondence between the action and the operation object. The action event type information clarifies the nature of the operation, the logical dependency constraints define the order of operations, and the combination rules of the operation sub-patterns provide a typical operation flow. By integrating this information, a more comprehensive description of the relationship between the action and the operation object can be achieved.

[0048] Data structures can be used to store and organize this information for integration. For example, a dictionary can be used to store the correspondence between operation actions and operation objects, where the dictionary key is the operation action and the value is a list containing information such as the operation object, action type, constraints, and operation sub-mode. For example, for the operation action "adjust transformer voltage", the corresponding dictionary value may include the operation object "transformer", the action type "parameter adjustment action", the constraint "transformer has been started", and the relevant operation sub-mode.

[0049] Step S230: Context chain construction is performed on the historical conversation logs to extract the semantic progressive relationship in the continuous conversation. The semantic progressive relationship includes the concept extension path and the problem deepening trajectory.

[0050] The context chain construction process is to analyze the semantic associations between consecutive conversations in the historical conversation log, construct context chains, and thus extract semantic progressive relationships. The semantic progressive relationship reflects the gradual deepening of students' knowledge and the gradual deepening of problems in the process of dialogue with the system. The concept extension path refers to the process of students' understanding of the concept of power system from basic concepts to derived concepts in the dialogue, such as extending from the basic concept of "generator" to "generator speed control system". The problem deepening trajectory refers to the process of students' questions developing from superficial problems to deep-seated problems, such as from "how to start the generator" to "possible faults and solutions during the generator startup process."

[0051] As an implementation manner, step S230 may be specifically implemented as the following steps S231 to S236: Step S231: Divide the historical dialogue log into turns to generate a dialogue turn set consisting of student questions and system feedback.

[0052] Turn partitioning involves dividing historical conversation logs into independent conversation turns based on the order of student questions and system feedback. A conversation turn collection is a summary of these conversation turns. Each conversation turn includes both a student's question and the system's feedback, reflecting a complete interaction process.

[0053] Turn division can be achieved by searching for delimiting markers in historical conversation logs. For example, symbols or keywords can be used in the logs to distinguish between student questions and system feedback, such as "student:" indicating the start of a student question and "system:" indicating the start of system feedback. These markers can be used to segment the log content into conversation turns. Natural language processing can also be used to analyze the semantics and tone of the text to determine whether it is a student question or system feedback, thereby determining turn division.

[0054] Step S232: Perform semantic vector conversion processing on each dialogue turn to generate a semantic vector reflecting the core content of the turn.

[0055] Semantic vector conversion converts the text content of a conversation turn into a vector format for subsequent similarity calculation and analysis. A semantic vector is a mathematical representation that reflects the core content and semantic information of a conversation turn. By converting a conversation turn into a semantic vector, the semantic information of the text can be quantified, making it easier for computers to process.

[0056] Pre-trained word vector models, such as Word2Vec or GloVe, can be used for semantic vector conversion. First, the text of the conversation turn is segmented and broken down into individual words. Then, each word is converted into a corresponding word vector using a pre-trained word vector model. Finally, the semantic vector for the entire conversation turn is obtained by performing a weighted average or other aggregation operation on these word vectors. For example, for the conversation turn "Student: What is the voltage of the transformer? System: The rated voltage of the transformer is 10kV," after word segmentation, the words "transformer," "voltage," "how much," "rated voltage," and "10kV" are obtained. The Word2Vec model is used to convert these words into word vectors, and then a weighted average is performed to obtain the semantic vector for the conversation turn.

[0057] Step S233: Calculate the cosine similarity between the semantic vectors of adjacent dialogue turns, and identify consecutive turns with close semantic associations as context chains.

[0058] Cosine similarity measures the cosine of the angle between two vectors and is used to determine their degree of similarity. In context chain construction, by calculating the cosine similarity between the semantic vectors of adjacent conversational turns, we can identify consecutive turns with close semantic connections. A context chain is a chain of consecutive conversational turns with close semantic connections, reflecting the progressive development of students' knowledge.

[0059] Step S234: Perform differential analysis on the semantic vectors in the context chain to extract the semantic changes from the initial question to the subsequent questions. The semantic changes include the expansion of the concept scope and the increase in the question depth.

[0060] Differential analysis processes analyze the semantic vectors in the context chain, calculating the differences between adjacent semantic vectors to extract semantic change. Semantic change reflects the evolution of students' knowledge during the conversation. Concept scope expansion refers to the increase in students' understanding of power system concepts from the initial question to subsequent questions, for example, from focusing solely on "generators" to focusing on "generator components and working principles." Question depth increase refers to the degree to which students' questions change from simple to complex and from superficial to in-depth, for example, from "how to operate a generator" to "safety hazards and preventive measures during generator operation."

[0061] Differential analysis can be performed by calculating the difference between adjacent semantic vectors. For example, for two adjacent semantic vectors V1 and V2 in a context chain, their difference DV = V2 - V1 is calculated. This difference vector is then analyzed to extract features associated with the expansion of concept scope and increased question depth. Machine learning methods, such as principal component analysis (PCA), can also be used to reduce the dimensionality of the difference vector and extract the main change features as the semantic change quantity.

[0062] Step S235: performing direction determination processing on the semantic change amount to determine whether the semantic progression path conforms to the logical structure of the power system knowledge system.

[0063] Direction determination involves determining whether the direction of semantic change conforms to the logical structure of the power system knowledge system. This knowledge system has inherent logical relationships, such as the hierarchical relationships between concepts and the sequential order of problem solving. If the semantic progression path conforms to this logical structure, the student's learning process is reasonable and effective; otherwise, there may be misunderstandings or errors.

[0064] Direction determination can be achieved by querying the power system knowledge base. The knowledge base stores various knowledge and logical rules of the power system. The semantic changes are compared with the logical structure in the knowledge base to determine whether the semantic progression path is reasonable. For example, if a student suddenly jumps from the concept of "generator" to "lightning protection measures for transmission lines" in a conversation without going through the relevant transition concepts, this semantic progression path may not conform to the logical structure of the power system knowledge system. Knowledge graphs can also be used to determine direction, mapping the semantic changes to the knowledge graph to see whether they progress along the logical relationship of the knowledge graph.

[0065] Step S236: Integrate the association information of the context chain, the numerical features of the semantic change amount, and the path direction of the semantic progression to generate a semantic progression relationship in the continuous dialogue.

[0066] Integration processing combines the contextual chain's association information, the numerical characteristics of semantic change, and the path direction of semantic progression to form a complete semantic progression relationship. The contextual chain's association information reflects the degree of semantic connection between consecutive conversations, the numerical characteristics of semantic change quantify the changes in students' knowledge, and the path direction of semantic progression ensures the rationality of semantic progression.

[0067] Data structures can be used to store and organize this information for integration. For example, a dictionary can be used to store semantic progression relationships, where the key is the conversation turn identifier and the value is a list containing contextual chain association information, numerical features of semantic changes, and the direction of the semantic progression path. This approach can fully represent the semantic progression relationships in continuous conversations, providing a foundation for subsequent analysis and application.

[0068] Step S240: taking the semantic segments, the correspondence between the operation actions and the operation objects, and the semantic progressive relationship as basic knowledge elements.

[0069] Basic knowledge elements are key information extracted from student interaction data streams, including semantic fragments, the correspondence between operational actions and operational objects, and semantic progression. Semantic fragments reflect students' understanding and focus of power system knowledge when asking questions, and include information such as equipment names, operating parameters, and fault symptoms. The correspondence between operational actions and operational objects reflects students' practical skills and operational standards in simulation operations, documenting their operational procedures and parameter adjustments for equipment. Semantic progression demonstrates the gradual deepening of knowledge and the deepening of questions students face during their conversations with the system, reflecting their learning process and development of thinking.

[0070] The purpose of using this information as basic knowledge elements is to be able to perform subsequent correlation calculations and cluster merging on it, thereby building a more complete knowledge system and providing a basis for accurately evaluating students' knowledge mastery status and predicting learning needs.

[0071] Step S250: performing correlation calculation on the basic knowledge elements to identify the conceptual correlation strength between semantic segments, the content matching degree between the operation actions and the semantic segments, and the temporal consistency between the semantic progressive relationship and the operation actions.

[0072] Association calculation involves analyzing the degree of association between basic knowledge elements and evaluating their relationships quantitatively. The strength of conceptual associations between semantic segments reflects the closeness between concepts within different semantic segments. For example, the strength of association between "transformer" and "voltage" is relatively high because voltage is a key operating parameter of a transformer. The content match between an action and a semantic segment refers to the degree of consistency between the action and the knowledge content involved in the semantic segment. For example, the action "adjusting the voltage of the transformer" has a high content match with the semantic segment "transformer voltage." The temporal consistency between semantic progression and action refers to whether the development of the semantic progression matches the chronological order of the action execution. For example, after students gradually gain a deeper understanding of generator fault handling knowledge during a conversation, they immediately perform generator fault simulation and handling operations in a simulation. In this case, the temporal consistency between the semantic progression and action is good.

[0073] There are several methods for calculating associations. For the conceptual association strength between semantic segments, the standard association strength distribution between concepts in the power system knowledge base can be queried and the matching deviation between the conceptual association strength in the semantic segment and the standard distribution can be calculated. For the content matching between operational actions and semantic segments, the historical execution frequency data of operational actions can be analyzed to calculate the correlation coefficient with the content matching of the semantic segments. For the temporal consistency between semantic progressive relationships and operational actions, the time span distribution of semantic progressive relationships in historical conversation logs can be analyzed to calculate the attenuation coefficient with the temporal consistency of operational actions and the interval between conversation turns. Finally, based on these calculation results, a dynamic association calculation model is constructed to perform weighted correction processing on the association of basic knowledge elements to obtain a comprehensive association value.

[0074] As an implementation manner, step S250 can be specifically implemented as the following steps S251 to S255: Step S251: obtaining a standard association strength distribution of concept nodes in the power system knowledge system, where the standard association strength distribution reflects the hierarchical association relationship between the core concepts and derived concepts of the power system.

[0075] The standard association strength distribution (SID) is a statistical distribution of the association strength between concept nodes in the power system knowledge system. It reflects the hierarchical relationship between core concepts and derived concepts. Core concepts, such as generators, transformers, and transmission lines, are the foundation and key concepts of the power system knowledge system. These concepts are the core support of the entire knowledge system. Derivative concepts, such as generator speed regulation systems and transformer winding losses, are developed based on core concepts and are closely related to core concepts.

[0076] Obtaining a standard association strength distribution can be achieved by analyzing and statistically analyzing the power system knowledge base. First, a power system knowledge graph is constructed, with concept nodes as nodes in the graph and associations between concepts as edges. Each edge is assigned a corresponding association strength value. Then, a statistical analysis is performed on the knowledge graph, calculating the association strength between each concept node and other nodes to obtain a standard association strength distribution. For example, in the knowledge graph, the association strength between "generator" and "generator rotor" is high, while the association strength with "transmission line insulator" is low. By statistically analyzing these association strength values, a standard association strength distribution can be obtained.

[0077] Step S252: Calculate the matching deviation value between the concept association strength between the semantic segments in the basic knowledge elements and the standard association strength distribution.

[0078] The matching deviation value is an indicator that measures the difference between the strength of conceptual associations between semantic segments in basic knowledge elements and the standard distribution of association strengths. A small matching deviation value indicates that the conceptual associations between semantic segments are consistent with the standard associations in the knowledge system, and students' understanding and application of concepts are relatively accurate. A large matching deviation value indicates that students' understanding may be biased or incomplete.

[0079] To calculate the matching deviation, we first extract the concept nodes from the semantic fragments of the basic knowledge elements and then calculate the association strength between these concept nodes based on the knowledge graph. Next, we compare the calculated association strength with the standard association strength distribution and calculate the difference between the two.

[0080] Step S253: extracting historical execution frequency data of the operation action in the simulation operation record, and calculating the correlation coefficient between the content matching degree between the operation action and the semantic segment and the operation execution frequency.

[0081] Historical execution frequency data refers to the statistical information about the number of times each action in the simulation operation record is executed within a certain period of time. It reflects the commonness of the action and the student's familiarity with it. The content match between the action and the semantic segment reflects the degree of fit between the knowledge involved in the action and the concepts contained in the semantic segment. The correlation coefficient measures the correlation between the content match between the action and the semantic segment and the frequency of action execution. A high correlation coefficient indicates a strong correlation between the frequency of action execution and the content match, meaning that students are more likely to frequently execute actions that have a high content match with the semantic segment. Conversely, a low correlation coefficient indicates a weaker correlation between the two.

[0082] Extracting historical execution frequency data for actions in simulation operation records can be achieved by statistically analyzing the records. First, the records are categorized by action type, and then the number of times each action was performed is counted. For example, over a period of time, a student performed the "select transformer" action 10 times, the "adjust transformer voltage" action 15 times, and so on. The degree of content match between the action and the semantic segment can be calculated by analyzing the action description and the semantic segment content. For example, the action "adjust transformer voltage" and the semantic segment "transformer voltage" can be considered highly content-matched and assigned a higher match score. To calculate the correlation coefficient, the Pearson correlation coefficient formula can be used. Assume that the content match between the action and the semantic segment is variable X, and the frequency of the action is variable Y. First, calculate the mean of X and Y, then calculate the difference between each data point and the mean. These differences are then multiplied and summed. Finally, the correlation coefficient is obtained by dividing by the product of the standard deviations of X and Y.

[0083] Step S254: Analyze the time span distribution of the semantic progressive relationship in the historical conversation log, and calculate the temporal consistency of the semantic progressive relationship and the operation action and the attenuation coefficient of the conversation turn interval.

[0084] The time span distribution of semantic progression refers to the distribution of the time span from the starting point to the end point of a semantic progression in historical conversation logs. It reflects the temporal characteristics of students' knowledge progression. For example, some students may achieve significant knowledge progression in a short period of time, while others may take longer. The temporal consistency between semantic progression and operational actions refers to whether the order of semantic progression development matches the order of operational execution. Ideally, students should perform related operations only after they have achieved a certain level of understanding and progression. The turn interval refers to the time interval between two consecutive conversation turns, which may affect the temporal consistency between semantic progression and operational actions. The decay coefficient measures the degree to which the temporal consistency between semantic progression and operational actions decreases as the turn interval increases.

[0085] Analyzing the time span distribution of semantic progression relationships in historical conversation logs can be achieved by performing timestamp analysis on the logs. First, determine the starting and ending conversation turns of the semantic progression relationship and record their timestamps. Then, calculate the time span. Statistically analyze the time spans of multiple semantic progression relationships to obtain the distribution of time spans. The temporal consistency between semantic progression relationships and operational actions can be calculated by comparing the key time points of the semantic progression with the execution time points of the operational actions. For example, if a student has a deep understanding of the semantic progression of generator fault handling knowledge in a conversation and then immediately performs a generator fault handling operation in a simulation, the temporal consistency is considered good. To calculate the decay coefficient, a mathematical model can be established that considers the relationship between conversation turn intervals and temporal consistency. For example, we can assume that temporal consistency decays exponentially with increasing conversation turn intervals. By fitting historical data, we can determine the decay coefficient.

[0086] Step S255: Construct a dynamic correlation calculation model based on the matching deviation value, correlation coefficient, and attenuation coefficient, perform weighted correction processing on the correlation of basic knowledge elements, and generate a comprehensive correlation value that reflects the knowledge system structure, operation behavior rules, and dialogue time characteristics.

[0087] The dynamic relevance calculation model is a mathematical model that comprehensively considers matching deviation, correlation coefficient, and attenuation coefficient to more accurately assess the relevance between basic knowledge elements. The matching deviation reflects the degree of alignment between the conceptual associations between semantic segments and the standard knowledge system. The correlation coefficient reflects the correlation between the matching degree of actions and semantic segment content and the frequency of action execution. The attenuation coefficient measures the degree to which the consistency between semantic progression and temporal consistency of actions is affected by the interval between conversational turns. By incorporating these factors into the model, it can comprehensively reflect the impact of the knowledge system structure, operational behavior patterns, and conversational temporal characteristics on the relevance of basic knowledge elements.

[0088] A linear weighted method can be used to construct a dynamic correlation calculation model. First, different weights are assigned to the matching deviation value, correlation coefficient, and attenuation coefficient, and the size of the weight is determined according to the importance of each factor in practical applications. Then, the value of each factor is multiplied by the corresponding weight, and the results are added to obtain a comprehensive correlation value. During the calculation process, the value of each factor can be normalized to ensure the stability and accuracy of the model. For example, for the matching deviation value, it is divided by the maximum possible deviation value and normalized to the range of 0 to 1. By weightedly correcting the correlation of basic knowledge elements through the dynamic correlation calculation model, a comprehensive correlation value that better reflects the actual situation can be obtained, providing a more reliable basis for subsequent clustering and merging processing.

[0089] Step S260: cluster and merge the basic knowledge elements according to the correlation calculation result to generate a knowledge tuple set with the power system concept as the node and the correlation strength matching consistency as the weight.

[0090] Cluster merging combines highly correlated basic knowledge elements based on correlation calculations to form more meaningful knowledge units. A knowledge tuple set is a collection of these merged knowledge units, with power system concepts serving as nodes and correlation strength matching consistency serving as edge weights. This consistency integrates information such as the strength of conceptual associations between semantic segments, the content matching between operational actions and semantic segments, and the temporal consistency between semantic progression and operational actions, providing a more comprehensive reflection of the relationships between knowledge.

[0091] Cluster merging can be performed using a clustering algorithm, such as a hierarchical clustering algorithm. First, a similarity matrix is constructed based on the correlation calculation results. The elements in the matrix represent the correlation between basic knowledge elements. Then, a hierarchical clustering algorithm is used to process the similarity matrix, gradually merging basic knowledge elements with high correlation into clusters. During this merging process, knowledge tuples are constructed using power system concepts as nodes and the consistency of correlation strength as edge weights. For example, the semantic fragment "transformer voltage" and the action "adjust transformer voltage" are highly correlated and are merged into a single knowledge tuple, with "transformer" as the node and the consistency of correlation strength as the edge weight. Through continuous merging, a set of knowledge tuples is ultimately generated, with power system concepts as nodes and the consistency of correlation strength as the weight. This set of knowledge tuples more clearly demonstrates the structure of power system knowledge and the knowledge connections during student learning.

[0092] Step S300: Input the knowledge tuple set into the inference engine to perform multi-dimensional reasoning processing to generate the student's knowledge mastery state vector and learning needs prediction vector. The knowledge mastery state vector includes the depth of concept understanding, operation standardization and logical coherence indicators, and the learning needs prediction vector includes the content supplement direction and ability improvement direction.

[0093] An inference engine is a system that analyzes and infers input knowledge. Based on certain rules and algorithms, it extracts valuable information from a collection of knowledge tuples. Multidimensional reasoning processing analyzes a collection of knowledge tuples from multiple perspectives, taking into account factors such as conceptual understanding, operational practice, and logical thinking. The student knowledge mastery state vector is a vector that comprehensively reflects the student's mastery of power system knowledge. The conceptual understanding depth index measures the student's understanding of power system concepts, the operational standardization index assesses the degree of standardization in simulation operations, and the logical coherence index examines the student's logical thinking ability during conversations and operations. The learning needs prediction vector predicts the knowledge content that students need to supplement and the abilities that they need to improve based on their knowledge mastery, providing targeted guidance for subsequent teaching.

[0094] After the knowledge tuple set is input into the inference engine, the inference engine processes the information in the knowledge tuple set. For the conceptual understanding depth indicator, the inference engine analyzes the coverage and correlation of the concept nodes in the knowledge tuple set and calculates the student's understanding of the power system concept. For the operational standardization indicator, the inference engine checks the matching rate of the operating actions in the simulation operation record with the standard operating procedures and the compliance rate of the constraints. For the logical coherence indicator, the inference engine evaluates the consistency of the semantic progressive relationship with the logical structure of the knowledge system and the rationality of the semantic change. Through the comprehensive calculation and analysis of these indicators, the student's knowledge mastery state vector is generated. At the same time, the inference engine analyzes the indicator items in the knowledge mastery state vector that are below the preset threshold, determines the direction of knowledge content that needs to be supplemented and the direction of operational links that need to be strengthened, and thus generates a learning needs prediction vector.

[0095] As an implementation manner, step S300 can be specifically implemented as the following steps S310 to S350: Step S310: Process the concept nodes and relationship edges in the knowledge tuple set through the concept understanding reasoning module of the reasoning engine, calculate the students' coverage and relevance of power system concepts, and generate a concept understanding depth index, where coverage is the ratio of the number of mastered concepts to the total number of concepts, and relevance is the ratio of the number of correct relevance relationships to the total number of relevance relationships.

[0096] The concept understanding reasoning module is a module within the reasoning engine specifically designed to process information related to concept understanding. It can perform in-depth analysis of concept nodes and relationship edges within a knowledge tuple set. Concept coverage reflects the extent to which students have mastered concepts within the power system. The number of mastered concepts refers to the number of concepts within the knowledge tuple set that students have already understood, and the total number of concepts refers to the number of all concepts involved in the power system training objectives. Concept relevance reflects the extent to which students understand the relationships between concepts. The number of correct relationships refers to the number of relationships within the knowledge tuple set that conform to the power system knowledge system, and the total number of relationships refers to the number of all relationships within the knowledge tuple set. The concept understanding depth index combines concept coverage and relevance to more comprehensively reflect students' understanding of power system concepts.

[0097] As an implementation manner, step S310 can be specifically implemented as the following steps S311 to S315: Step S311: extract all concept nodes in the knowledge tuple set, count their number as the number of mastered concepts, and obtain the standard concept set corresponding to the power system training target, and count their number as the total number of concepts.

[0098] A concept node is an element in the knowledge tuple set that represents the basic concepts of the power system. Extracting all concept nodes can be achieved by traversing the knowledge tuple set. During the traversal process, the concept nodes in each knowledge tuple are extracted and deduplicated to avoid repeated counting. The number of concepts mastered reflects the number of concepts that students have been exposed to and understood during the learning process. The standard concept set corresponding to the power system training objectives is a set of concepts determined according to the training syllabus and teaching requirements. It covers all concepts that students should master during the training process. Counting the number of standard concept sets can be completed by counting the set.

[0099] For example, a knowledge tuple set contains multiple knowledge tuples, such as ("transformer," "voltage," association), ("generator," "power," association), etc. Through traversal, we extract concept nodes such as "transformer," "voltage," "generator," and "power." After removing duplicates, we count the number of these nodes as four, which we use as the number of concepts mastered. The standard concept set corresponding to the power system training objective contains 10 concepts, such as "transformer," "voltage," "generator," "power," "transmission line," and "relay protection." These 10 concepts are counted as 10, which we use as the total number of concepts.

[0100] Step S312: Calculate the ratio of the number of concepts that have been mastered to the total number of concepts as the concept coverage.

[0101] Concept coverage is an indicator that measures the scope of a student's mastery of power system concepts. It is calculated as the ratio of the number of concepts mastered to the total number of concepts. A larger ratio indicates a broader range of concepts mastered by the student; conversely, a smaller ratio indicates a narrower range of concepts mastered by the student.

[0102] Using the number of concepts mastered and the total number of concepts obtained in step S311, calculate the concept coverage according to the formula: concept coverage = number of concepts mastered / total number of concepts. For example, if the number of concepts mastered is 4 and the total number of concepts is 10, the concept coverage is 4 / 10 = 0.4.

[0103] Step S313: extract all relationship edges in the knowledge tuple set, and count the number of correct association relationships that conform to the power system knowledge system.

[0104] Relationship edges are elements that connect concept nodes within a knowledge tuple set and reflect the logical associations between concepts. Extracting all relationship edges can be done by traversing the knowledge tuple set and extracting the relationship edges within each knowledge tuple. Correct relationships that conform to the power system knowledge system are those that are consistent with the actual knowledge and logic of the power system. For example, the device-attribute binding relationship between "transformer" and "voltage" is a correct relationship.

[0105] After extracting the relationship edges, they need to be compared with the knowledge in the power system knowledge base to determine whether they conform to the knowledge system. For example, for the relationship edge "transformer-generates-voltage," a query to the knowledge base shows that a transformer is a device that changes voltage, not generates it, so this relationship edge does not conform to the knowledge system. However, for "transformer-has-voltage," it does conform to the knowledge system. The number of relationship edges that conform to the knowledge system is counted to determine the number of correct associations.

[0106] Step S314: Count the total number of relationship edges in the knowledge tuple set, and calculate the ratio of the number of correct association relationships to the total number of relationship edges as the concept association degree.

[0107] The total number of edges is the number of all edges in the knowledge tuple set, calculated by counting the extracted edges. Concept relevance measures a student's understanding of the relationships between concepts. It is calculated as the ratio of the number of correct relationships to the total number of edges. A larger ratio indicates a more accurate understanding of the relationships between concepts.

[0108] The number of correct association relationships and the total number of relationship edges obtained by counting in step S313 are used to perform calculations according to the formula: concept association degree = number of correct association relationships / total number of relationship edges.

[0109] Step S315: Integrate the concept coverage and concept relevance to generate a concept understanding depth index.

[0110] The depth of concept understanding index comprehensively reflects students' understanding of power system concepts. It integrates concept coverage and concept relevance. Concept coverage reflects the extent of a student's mastery of a concept, while concept relevance reflects their understanding of the relationships between concepts. By combining these two indicators, a more comprehensive assessment of students' conceptual understanding can be achieved. A weighted average approach can be used to integrate concept coverage and concept relevance.

[0111] Step S320: The operation specification reasoning module of the reasoning engine processes the correspondence between the operation actions and the operation objects in the knowledge tuple set, calculates the matching rate of the operation sub-pattern and the standard operation process and the constraint compliance rate, and generates an operation standardization index, where the matching rate is the ratio of the number of matching sub-patterns to the total number of sub-patterns, and the compliance rate is the ratio of the number of actions that meet the constraints to the total number of actions.

[0112] The operational specification reasoning module is a module within the inference engine that processes operational specification-related information. It analyzes the correspondence between operational actions and operational objects in a knowledge tuple set. Operational subpatterns are recurring action combinations extracted from simulation operation records, and standard operating procedures are the correct operational procedures specified in power system simulation experiments. The matching rate reflects the degree to which a student's operational subpatterns align with the standard operating procedures. The number of matching subpatterns refers to the number of operational subpatterns that match subpatterns in the standard operating procedures, while the total number of subpatterns refers to the total number of operational subpatterns in the knowledge tuple set. Constraints are logical dependencies between operational actions, such as inability to adjust parameters if a device is not powered on. The compliance rate measures the degree to which students adhere to these constraints during operation. The number of actions that meet the constraints refers to the number of actions in the operation records that meet the constraints, while the total number of actions refers to the total number of operational actions in the knowledge tuple set. The operational standardization index combines the matching rate and the compliance rate to comprehensively assess the student's standardization in simulation operations.

[0113] When processing a knowledge tuple set through the operational standardization inference module, the module first extracts the operational subpatterns within the knowledge tuple set and compares them with the subpatterns in the standard operational procedures. The matching subpatterns and the total number of subpatterns are counted, and the matching rate is calculated using the formula: Matching rate = Number of matching subpatterns / Total number of subpatterns. Simultaneously, the module extracts the logical dependency constraints within the knowledge tuple set, counts the number of actions that meet the constraints and the total number of actions, and calculates the compliance rate using the formula: Compliance rate = Number of actions that meet the constraints / Total number of actions. Finally, the matching rate and compliance rate are combined to generate an operational standardization index.

[0114] As an implementation manner, step S320 can be specifically implemented as the following steps S321 to S325: Step S321: extract the operation sub-patterns in the knowledge tuple set, count their number as the total sub-pattern number, obtain the standard operation process corresponding to the power system simulation experiment, extract the standard sub-patterns therein, and count their number as the matching sub-pattern number.

[0115] Operational subpatterns are repetitive action combinations with the same operational objectives, abstracted from simulation operation records. Operational subpatterns can be extracted by analyzing the correspondence between operational actions and operational objects in a knowledge tuple set to identify recurring action sequences. The total number of subpatterns reflects the total number of operational subpatterns formed by students during simulation operations. The standard operating procedures corresponding to the power system simulation experiment are verified correct operating procedures that contain a series of standard subpatterns. Standard subpatterns within the standard operating procedures are extracted and counted to obtain the number of matching subpatterns.

[0116] For example, the knowledge tuple set includes operation subpatterns such as "start the generator - set the generator voltage - monitor the generator power" and "start the transformer - adjust the transformer oil temperature - check the transformer operating status." These subpatterns are counted as five, representing the total number of subpatterns. The standard operation procedure includes standard subpatterns such as "start the generator - set the generator voltage - monitor the generator power" and "start the transformer - check the transformer oil level - adjust the transformer voltage." Two of these subpatterns match the student's operation subpatterns, and these subpatterns are counted as the number of matching subpatterns.

[0117] Step S322: Calculate the ratio of the number of matching sub-patterns to the total number of sub-patterns as the operation sub-pattern matching rate.

[0118] The Operation Sub-Pattern Matching Rate is an indicator that measures the degree to which a student's operation sub-patterns conform to the standard operation procedures. It is calculated as the ratio of the number of matching sub-patterns to the total number of sub-patterns. The larger the ratio, the closer the student's operation sub-patterns are to the standard operation procedures.

[0119] Using the number of matching subpatterns and the total number of subpatterns counted in step S321, the operation subpattern matching rate is calculated according to the formula: Operation Subpattern Matching Rate = Number of Matching Subpatterns / Total Number of Subpatterns. For example, if the number of matching subpatterns is 2 and the total number of subpatterns is 5, the operation subpattern matching rate is 2 / 5 = 0.4.

[0120] Step S323: extracting the logical dependency constraints in the knowledge tuple set, counting the number of operation actions that meet the constraints as the number of actions that meet the constraints, and counting the total number of action events in the knowledge tuple set.

[0121] Logical dependency constraints represent the logical relationships between operational actions. For example, parameter adjustment cannot be performed if the device is not started, and recovery operations cannot be performed until a fault is cleared. Logical dependency constraints can be extracted from a knowledge tuple set by analyzing the correspondence between operational actions and operational objects to identify the constraints. The number of actions meeting the constraints is the number of actions in the operation record that satisfy these constraints, while the total number of action events is the total number of all operational actions in the knowledge tuple set.

[0122] When counting the number of actions that meet constraints, each action needs to be checked to determine whether it meets the constraints. For example, for the action "adjust the voltage of the transformer," it is necessary to check whether the transformer has been started. If it has, the action meets the constraints; otherwise, it does not. By checking all actions, the number of actions that meet the constraints is counted. At the same time, all actions in the knowledge tuple set are counted to obtain the total number of action events.

[0123] Step S324: Calculate the ratio of the number of actions that meet the constraint conditions to the total number of action events as the constraint condition compliance rate.

[0124] The constraint compliance rate measures the degree to which students adhere to the logical dependency constraints between actions during the operation process. It is calculated as the ratio of the number of actions that meet the constraints to the total number of action events. The larger the ratio, the more students adhere to the constraints during the operation.

[0125] The number of actions that meet the constraint conditions and the total number of action events obtained by counting in step S323 are used to perform calculations according to the formula: constraint compliance rate = number of actions that meet the constraint conditions / total number of action events.

[0126] Step S325: Integrate the operation sub-pattern matching rate and the constraint condition compliance rate to generate an operation standardization index.

[0127] The operational standardization metric combines the operational sub-pattern matching rate and the constraint compliance rate to comprehensively assess the student's standardization during simulation operations. The operational sub-pattern matching rate reflects the degree to which the student's operational procedures conform to the standard operating procedures, while the constraint compliance rate reflects the student's adherence to the logical constraints during the operation. A weighted average approach can be used to integrate the operational sub-pattern matching rate and the constraint compliance rate.

[0128] Step S330: Process the semantic progressive relationship in the knowledge tuple set through the logical coherence reasoning module of the reasoning engine, calculate the degree of consistency between the semantic progressive path and the logical structure of the knowledge system and the rationality of the semantic change, and generate a logical coherence index, where the degree of consistency is the ratio of the length of the consistent path to the total path length, and the rationality is the ratio of the number of changes that conform to the law of knowledge deepening to the total number of changes.

[0129] The logical coherence reasoning module is a module within the inference engine that processes information related to logical coherence. It analyzes the semantic progression relationships within a set of knowledge tuples. The semantic progression path is the path along which students progress through knowledge during the conversation, and the logical structure of the knowledge system represents the inherent logical relationships within the power system knowledge system. The degree of consistency reflects the degree of conformity between the semantic progression path and the logical structure of the knowledge system. The consistency path length refers to the length of the portion of the semantic progression path that aligns with the logical structure of the knowledge system, and the total path length is the total length of the semantic progression path. Semantic change refers to the amount of knowledge change from the initial question to the subsequent questions. Rationality measures whether the semantic change conforms to the laws of knowledge deepening. The number of semantic changes that conform to the laws of knowledge deepening refers to the number of semantic changes that conform to the laws of knowledge deepening, and the total number of changes refers to the total number of semantic changes in the set of knowledge tuples. The logical coherence index combines consistency and rationality to assess students' logical thinking abilities during conversation and learning.

[0130] When processing a knowledge tuple set through the logical coherence reasoning module, the semantic progressive paths within the knowledge tuple set are first extracted and compared with the standard logical structure of the power system knowledge system. The length of the matching paths and the total path length are counted, and the degree of consistency is calculated according to the formula: consistency = matching path length / total path length. Simultaneously, the semantic changes within the knowledge tuple set are extracted to determine whether they conform to the knowledge deepening rules. The number of matching changes and the total number of matching changes are counted, and the rationality is calculated according to the formula: rationality = number of matching changes / total number of matching changes. Finally, the degree of consistency and rationality are integrated to generate a logical coherence index.

[0131] As an implementation manner, step S330 may be specifically implemented as the following steps S331 to S335: Step S331: extract the semantic progressive path in the knowledge tuple set, count its length as the total path length, obtain the standard logical structure of the power system knowledge system, and count the number of rounds in the semantic progressive path that are consistent with the standard structure as the matching path length.

[0132] The semantic progressive path is the trajectory of student knowledge progression extracted from historical conversation logs. Extracting the semantic progressive path can be achieved by analyzing the context chain and semantic change. The total path length is the total length of the semantic progressive path, which can be determined by counting the number of conversation turns contained in the path. The standard logical structure of the power system knowledge system is the inherent logical relationship of the power system knowledge, such as the hierarchical relationship of concepts, the order of problem solving, etc. Count the number of turns in the semantic progressive path that are consistent with the standard structure, that is, find the conversation turns in the semantic progressive path that conform to the logical structure of the knowledge system, and use their number as the length of the matching path.

[0133] Step S332: Calculate the ratio of the matching path length to the total path length as the logical structure matching degree.

[0134] Logical structure consistency is an indicator that measures the degree to which the semantic progression path matches the logical structure of the knowledge system. It is calculated as the ratio of the length of the matching path to the total path length. The larger the ratio, the more consistent the semantic progression path is with the logical structure of the knowledge system.

[0135] The matching path length and the total path length obtained in step S331 are used to perform calculations according to the formula: logical structure matching degree = matching path length / total path length.

[0136] Step S333: extracting the semantic changes in the knowledge tuple set, counting the number of changes that conform to the knowledge deepening law as the reasonable number of changes, and counting the total number of semantic changes in the knowledge tuple set.

[0137] Semantic change is extracted from differential analysis of semantic vectors in the context chain and reflects the evolution of students' knowledge during the conversation. Changes consistent with knowledge deepening are those in semantic change that align with the logical development and knowledge deepening requirements of the power system knowledge system, such as the expansion from simple to complex concepts and the in-depth study from superficial to deeper issues. Counting the number of changes consistent with knowledge deepening and the total semantic change can be achieved by analyzing and assessing the semantic changes in the knowledge tuple set one by one.

[0138] After extracting semantic changes, they are compared with the knowledge deepening rules in the power system knowledge base to determine whether they conform to the rules. For example, the semantic change "from focusing on the basic structure of the generator to focusing on the principles of generator fault diagnosis" conforms to the knowledge deepening rules; however, if there is no reasonable transition from focusing on the generator to suddenly focusing on the insulators of the transmission line, it does not conform to the knowledge deepening rules. By examining all semantic changes, the number of changes that conform to the rules and the total number of semantic changes are counted.

[0139] Step S334: Calculate the ratio of the number of reasonable changes to the total number of semantic changes as the rationality of the semantic change.

[0140] The rationality of semantic change is an indicator that measures whether the amount of semantic change conforms to the laws of knowledge deepening. It is calculated by the ratio of the number of reasonable changes to the total number of semantic changes. The larger the ratio, the more consistent the semantic change is with the laws of knowledge deepening.

[0141] Using the reasonable change amount number and the total semantic change amount number obtained by counting in step S333, calculation is performed according to the formula: semantic change rationality = reasonable change amount number / total semantic change amount number.

[0142] Step S335: Integrate the logical structure consistency and the semantic change rationality to generate a logical coherence index.

[0143] The logical coherence index combines logical structure consistency and semantic change rationality, providing a comprehensive assessment of students' logical thinking abilities during conversation and learning. Logical structure consistency reflects the degree to which the semantic progression path aligns with the logical structure of the knowledge system, while semantic change rationality reflects whether the semantic change conforms to the principles of knowledge deepening. A weighted average approach can be used to integrate logical structure consistency and semantic change rationality.

[0144] Step S340: normalize the concept understanding depth index, operation standardization index, and logical coherence index to generate a student knowledge mastery state vector.

[0145] Normalization involves adjusting the values of different indicators to have the same dimension and range, facilitating comprehensive analysis and comparison. The conceptual understanding depth index, operational standardization index, and logical coherence index each reflect students' mastery of power system knowledge from different perspectives, but their numerical ranges and meanings may differ. Normalization eliminates these differences, making these indicators comparable. The student knowledge mastery state vector comprehensively reflects the student's knowledge mastery, encompassing information on multiple aspects, including conceptual understanding, operational practice, and logical thinking.

[0146] Normalization can be performed using a variety of methods, such as the minimum-maximum normalization method. First, the minimum and maximum values of each indicator are determined. Each indicator value is then transformed using the minimum-maximum normalization formula. After normalization, the conceptual understanding depth indicator, operational standardization indicator, and logical coherence indicator are combined into a vector to generate the student's knowledge mastery status vector.

[0147] As an implementation manner, step S340 can be specifically implemented as the following steps S341 to S345: Step S341: Calculate the first covariance value of the concept understanding depth index and the operation standardization index to reflect the degree of influence of concept understanding on operation standardization.

[0148] Covariance is a statistic that measures the linear relationship between two variables. The first covariance value measures the correlation between the depth of conceptual understanding indicator and the operational standardization indicator, reflecting the impact of conceptual understanding on operational standardization. A positive first covariance value indicates that an increase in the depth of conceptual understanding may lead to an improvement in operational standardization; a negative first covariance value indicates a possible negative correlation between the two; and a value close to 0 indicates a weak linear relationship between the two.

[0149] Step S342: Calculate the second covariance value between the operational standardization index and the logical coherence index to reflect the promoting effect of operational practice on logical coherence.

[0150] The second covariance value measures the correlation between the operational standardization index and the logical coherence index, reflecting the impact of operational practice on logical coherence. Similar to the first covariance value, a positive second covariance value indicates that improved operational standardization may promote enhanced logical coherence; a negative second covariance value indicates a negative correlation; and a value close to zero indicates a weak linear relationship between the two.

[0151] Step S343: Calculate the third covariance value between the logical coherence index and the conceptual understanding depth index to reflect the reinforcing effect of logical cohesion on conceptual mastery.

[0152] The third covariance value measures the correlation between the logical coherence index and the depth of conceptual understanding index, reflecting the impact of logical coherence on conceptual mastery. A positive third covariance value indicates that increased logical coherence contributes to improved depth of conceptual understanding; a negative third covariance value suggests the opposite relationship; and a value close to 0 indicates a weak linear correlation between the two.

[0153] Step S344: constructing an indicator association matrix based on the first covariance value, the second covariance value, and the third covariance value. The indicator association matrix is used to quantify the mutual influence weights between the indicators.

[0154] The indicator correlation matrix is a square matrix whose elements are composed of the first covariance value, the second covariance value, and the third covariance value. It is used to quantify the mutual influence weights between the depth of conceptual understanding indicator, the operational standardization indicator, and the logical coherence indicator. The diagonal elements of the matrix are usually set to 1, indicating that each indicator has a weight of 1.

[0155] Step S345: Perform weighted normalization processing on the concept understanding depth index, operation standardization index, and logical coherence index according to the indicator association matrix to generate a student knowledge mastery state vector that includes the synergistic relationship between indicators. The values of each dimension of the state vector are dynamically adjusted as the influence weights between indicators change.

[0156] Weighted normalization is based on normalization and weights each indicator according to the weights in the indicator association matrix. First, the conceptual understanding depth indicator, operational standardization indicator, and logical coherence indicator are normalized to a minimum-maximum value range of 0 to 1. Then, the normalized indicators are weighted and summed according to the weights in the indicator association matrix. The weighted indicator values are combined into a vector to generate a student knowledge mastery state vector that reflects the synergistic relationship between indicators. Because the weights in the indicator association matrix reflect the mutual influence between indicators, the values of each dimension of the state vector are dynamically adjusted as the influence weights between indicators change.

[0157] Step S350: Analyze the indicator items in the knowledge mastery state vector that are lower than the preset threshold, determine the direction of knowledge content that needs to be supplemented and the direction of operation links that need to be strengthened, and generate a learning demand prediction vector.

[0158] The preset threshold is a standard value set in advance to judge whether the student's knowledge mastery level is qualified. By analyzing the indicator items in the knowledge mastery state vector that are lower than the preset threshold, the deficiencies in the student's knowledge mastery can be found. The direction of knowledge content that needs to be supplemented refers to the situation where the concept understanding depth index is low, and the knowledge content that students need to learn further is determined. For example, if the concept understanding of power system fault handling in the concept understanding depth index is poor, then the relevant fault handling knowledge needs to be supplemented. The direction of operation links that need to be strengthened refers to the situation where the operation standardization index is low, and the operation links that students need to strengthen training are determined. For example, if the operation standardization index for equipment startup and parameter adjustment is poor, then the training of these operation links needs to be strengthened.

[0159] The learning needs prediction vector is a vector that reflects students' learning needs. It contains information such as the knowledge content that needs to be supplemented and the operational aspects that need to be strengthened. When analyzing the knowledge mastery state vector, each indicator is compared with a preset threshold. If an indicator is below the threshold, the corresponding knowledge content or operational aspect is considered part of the learning need. For example, if the depth of concept understanding indicator is below the threshold and the main issue is insufficient understanding of the concept of power system relay protection, "relay protection knowledge supplementation" is considered as the knowledge content that needs to be supplemented; if the operational standardization indicator is below the threshold and the main issue is poor standardization of fault simulation operations, "fault simulation operation reinforcement" is considered as the operational aspect that needs to be strengthened. This information is combined into a learning needs prediction vector.

[0160] Step S400: Execute interactive content generation processing based on the knowledge mastery state vector and the learning demand prediction vector to generate adaptive classroom interactive response data, which includes knowledge completion information and ability training tasks.

[0161] Interactive content generation processing generates classroom interactive response data tailored to students based on their knowledge mastery status and learning needs. The knowledge mastery state vector reflects the student's current level of mastery of power system knowledge, and the learning needs prediction vector indicates the knowledge content that the student needs to supplement and the operational links that need to be strengthened. Adaptive classroom interactive response data refers to response data that can provide targeted knowledge and training tasks based on the student's specific situation. Knowledge completion information is relevant knowledge content provided to make up for the deficiencies in students' knowledge mastery, and ability training tasks are designed to enhance students' practical and problem-solving abilities.

[0162] Based on the knowledge mastery state vector and the learning demand prediction vector, the system extracts relevant information from the corresponding knowledge base and task library. For lower indicator items in the knowledge mastery state vector, the corresponding knowledge content, such as basic definitions, related concepts, typical application cases, etc., is extracted from the power system knowledge base according to the direction determined by the learning demand prediction vector to generate knowledge completion information. For operational links that need to be strengthened in the learning demand prediction vector, appropriate training tasks are selected from the ability training task library, such as comprehensive application questions, cross-module collaboration questions, system-level analysis questions, etc., to generate ability training tasks. Finally, the knowledge completion information and ability training tasks are integrated to generate adaptive classroom interaction response data.

[0163] As an implementation manner, step S400 can be specifically implemented as the following steps S410 to S460: Step S410: For the concept understanding depth index in the knowledge mastery state vector, if the index is lower than a first preset threshold, the basic definition, related concepts and typical application cases of the corresponding concept are extracted from the power system knowledge base to generate knowledge completion information.

[0164] The first preset threshold is a standard value used to determine whether the depth of concept understanding is satisfactory. When the depth of concept understanding indicator is lower than the first preset threshold, it indicates that the student's understanding of certain power system concepts is insufficient. The power system knowledge base is a database that stores various power system knowledge, including basic definitions of concepts, related concepts, and typical application cases. Basic definitions are basic explanations of power system concepts, related concepts are other concepts related to the concept, and typical application cases are examples of the application of the concept in actual power systems.

[0165] When the depth of conceptual understanding index falls below the first preset threshold, the system identifies the specific concepts involved in the index. For example, if the depth of conceptual understanding of "transformer" is low, the system extracts the basic definition of "transformer" from the power system knowledge base, such as "a transformer is a device that uses the principle of electromagnetic induction to change AC voltage"; related concepts, such as "winding" and "iron core"; and typical application cases, such as "in power transmission, transformers are used to increase or decrease voltage to reduce transmission losses." This information is organized and combined to generate knowledge completion information about "transformer."

[0166] Step S420: For the operation standardization index in the knowledge mastery state vector, if the index is lower than the second preset threshold, the standard process, common errors and correction methods of the corresponding operation are extracted from the simulation operation standard library to generate operation standard reinforcement information.

[0167] The second preset threshold is a standard value used to determine whether the operational standardization is qualified. When the operational standardization index is lower than the second preset threshold, it indicates that the student's standardization in certain simulation operations is insufficient. The simulation operation standard library is a database that stores power system simulation operation standards and specifications. It contains standard procedures, common errors, and corrective measures for various operations. Standard procedures are the correct operating steps specified in power system simulation experiments. Common errors are errors that students are prone to making during operation, and corrective measures are solutions to these errors.

[0168] When the operational standardization index falls below a second preset threshold, the system identifies the specific operations involved. For example, if the operational standardization for a generator start-up operation is low, the system extracts standard procedures for this operation from a standard library of simulated operations, such as "check all generator parameters for normal operation, close the circuit breaker, start the prime mover, and gradually increase the excitation current." Common errors, such as "starting without checking parameters or increasing the excitation current too quickly," are also identified, along with corrective measures, such as "carefully check all parameters before starting and increase the excitation current at the prescribed rate." This information is organized and combined to generate enhanced operational standardization information for the generator start-up operation.

[0169] Step S430: For the logical coherence index in the knowledge mastery state vector, if the index is lower than the third preset threshold, the logical deduction process, associated verification method and typical problem chain of the corresponding knowledge module are extracted from the knowledge system logic library to generate logical connection reinforcement information.

[0170] The third preset threshold is a preset standard value used to measure whether the logical coherence meets the standard. When the logical coherence index is lower than this threshold, it means that the student is deficient in logical thinking and knowledge connection. The knowledge system logic library is a database that specifically stores the logical structure and deduction rules of power system knowledge, which contains the logical deduction process, association verification method and typical problem chain of each knowledge module. The logical deduction process shows the reasoning steps from basic concepts to complex conclusions. The association verification method is used to test whether the association between knowledge is reasonable. The typical problem chain is a series of interrelated questions that can guide students to deeply understand the logical relationship of knowledge.

[0171] When the logical coherence index falls below the third preset threshold, the system locates the specific knowledge module involved. For example, if the logical coherence of the knowledge module "Power System Fault Analysis" is poor, the system extracts the module's logical deduction process from the knowledge system's logic library, such as the process of deriving the fault type and location from the fault phenomenon through electrical principles and mathematical models; associated verification methods, such as using fault recording data to verify the accuracy of fault analysis results; and typical problem chains, such as "What changes will occur in current and voltage when a fault occurs? How can these changes be used to determine the fault type? What are the different treatment methods for different fault types?" This information is integrated to generate logical coherence reinforcement information for "Power System Fault Analysis."

[0172] Step S440: Based on the content supplement direction in the learning demand prediction vector, combined with the knowledge completion information, operation specification reinforcement information, and logical connection reinforcement information, content-adapted guided explanation data is generated.

[0173] The content supplementation directions in the learning needs prediction vector clearly identify areas of knowledge students need to further study. Knowledge completion information, operational specification reinforcement information, and logical connection reinforcement information provide students with resources to address knowledge gaps and enhance their abilities from different perspectives. Content-adapted guided explanation data organically combines this information and presents it to students in a guiding manner, helping them better understand and master the knowledge.

[0174] Based on the content supplementation direction, the system selects relevant knowledge supplementation information, operational specification reinforcement information, and logical connection reinforcement information. For example, if the content supplementation direction is "power system relay protection," the system selects basic definitions, related concepts, and typical application cases of relay protection from the previously generated knowledge supplementation information; extracts standard procedures, common errors, and correction methods for relay protection device operation from the operational specification reinforcement information; and obtains the logical deduction process, associated verification methods, and typical problem chains of the relay protection knowledge module from the logical connection reinforcement information. This information is then organized according to a certain logical sequence, for example, first introducing the basic concepts of relay protection, then explaining the operational specifications, and finally guiding students to a deeper understanding through logical deduction and typical problems. During this organization process, guiding language is used, such as "Let's first understand the basic principles of relay protection, which are crucial for subsequent operation and analysis. Next, let's look at what standards should be followed in actual operation...", thereby generating content-appropriate guided explanation data.

[0175] Step S450: Based on the ability improvement direction in the learning demand prediction vector, comprehensive application questions, cross-module collaboration questions and system-level analysis questions that are connected with the mastered knowledge are extracted from the ability training task library to generate ability-adapted training task data.

[0176] The competency improvement directions in the learning needs prediction vector indicate the types of competencies students need to focus on improving. The competency training task library is a database of various training tasks, including comprehensive application questions, cross-module collaboration questions, and system-level analysis questions of varying difficulty levels and types. These questions are related to various knowledge modules of the power system. Linking training tasks to existing knowledge means that they can expand and deepen students' existing knowledge, helping them apply what they've learned to solve real-world problems and enhance their overall abilities.

[0177] Based on the direction of capability improvement, the system will select appropriate questions from the capability training task library. For example, if the capability improvement direction is "comprehensive fault handling ability", the system will select comprehensive application questions that are connected with the knowledge of power system fault analysis that has been mastered, such as "It is known that a certain power system has experienced a fault with abnormal voltage fluctuations. Please combine the knowledge you have learned to analyze the possible causes of the fault and propose corresponding treatment measures"; cross-module collaboration questions, such as "In a power system that includes power generation, transmission and distribution modules, a complex fault has occurred. You need to coordinate the knowledge of each module to diagnose and repair the fault"; system-level analysis questions, such as "Evaluate and optimize the fault emergency plan for the entire power system." These selected questions will be sorted and arranged to generate capability-adapted training task data.

[0178] Step S460: Integrate the guided explanation data and the training task data into content and perform format adaptation processing to generate adaptive classroom interaction response data.

[0179] Content integration involves organically integrating guided explanation data and training task data, making them interconnected and complementary in content. Format adaptation involves adjusting the format of the integrated content to meet the interactive requirements of the virtual classroom and facilitate student access and understanding. Adaptive classroom interaction response data is integrated and adapted to precisely meet students' learning needs.

[0180] During the content integration process, the system will reasonably match the guided explanation data and training task data based on the logical structure of knowledge and the learning patterns of students. For example, after explaining the guided content of a certain knowledge module, related training tasks will be arranged immediately to allow students to consolidate what they have learned in a timely manner. At the same time, the content will be optimized and adjusted to avoid repeated or contradictory information. In terms of format adaptation processing, the integrated content will be converted into a format suitable for virtual classroom presentation, such as text, pictures, videos, etc. If the guided explanation data contains complex principle derivations, it may be presented in the form of pictures and texts; for training task data, it will be presented in the form of a clear list of questions and answer requirements. Through content integration and format adaptation processing, adaptive classroom interaction response data is ultimately generated to provide students with a coherent and clear learning experience.

[0181] Step S500: Feedback the classroom interaction response data to the virtual classroom interaction interface, forming a dynamic interaction process between students' learning status and classroom feedback.

[0182] The virtual classroom interactive interface is a platform for students to interact with the system. Feeding classroom interaction response data to this interface allows students to receive timely feedback on their learning progress. The dynamic interaction between student learning status and classroom feedback is a continuous and mutually influential process. A student's learning status is reflected in the knowledge mastery state vector and learning needs prediction vector. Based on this information, the system generates classroom interaction response data and feeds it back to the student. After receiving feedback, students learn and practice, and their learning status changes. This new learning status is then collected and analyzed by the system, generating new classroom interaction response data, and this cycle repeats.

[0183] When feeding back classroom interaction response data to the virtual classroom interactive interface, the system will display the data in an intuitive and easy-to-understand manner. For example, knowledge completion information can be presented in the form of a text box, pop-up window, or special page, and ability training tasks can be presented in the form of a task list. Students can answer questions and submit them directly on the interface. At the same time, the system will provide corresponding navigation and operation instructions to facilitate students to browse and use feedback information. After students study and operate in the virtual classroom, the system will collect new interaction data from students in real time, including new questions, simulation operation records, and dialogue logs, re-perform knowledge representation and reasoning, update the knowledge mastery state vector and learning demand prediction vector, and generate new classroom interaction response data, thus forming a dynamic and continuously optimized learning cycle, enabling students to gradually improve their mastery of power system knowledge and practical ability through continuous feedback and learning.

[0184] It is understandable that the various algorithms involved in the above-mentioned introductions of the embodiments of the present invention, such as the cosine distance algorithm, the Pearson correlation coefficient algorithm, the clustering algorithm, etc., can all be learned from the relevant content in the prior art. In order to save space, they will not be expanded too much in the embodiments of the present invention. In addition, when implementing the scheme of the present invention, those skilled in the art can supplement the details according to the common knowledge in this field. For example, according to the common knowledge in this field, normalization can be used to eliminate dimensional conflicts before feature fusion, interpolation can be used to eliminate dimensional differences, and thresholds can be reasonably set based on historical data, experience or business scenario requirements. The model can be trained based on a general model training method, and the number of layers in the model structure can be set based on actual needs, the activation function can be selected, etc. The present invention will no longer provide redundant introductions to the overly detailed implementation process.

[0185] See also Figure 2 , Figure 2This is a schematic diagram of the structure of a computer system provided in an embodiment of the present invention. The computer system includes at least a processor 101, a communication interface 102, and a memory 103. The processor 101, communication interface 102, and memory 103 may be connected via a bus or other means. The processor 101 (also known as the Central Processing Unit (CPU)) is the computing and control core of the computer system, capable of parsing various instructions within the computer system and processing various data within the computer system. The communication interface 102 may optionally include a standard wired interface or a wireless interface (such as Wi-Fi, a mobile communication interface, etc.), which can be used to send and receive data under the control of the processor 101. The communication interface 102 may also be used for data transmission and interaction within the computer system. The memory 103 is a storage device in the computer system for storing programs and data. It is understood that the memory 103 herein may include both the built-in memory of the computer system and, of course, the extended memory supported by the computer system. The memory 103 provides storage space, which stores the computer system's operating system, but this is not limited to this in the present invention.

[0186] In one embodiment, the processor 101 executes the virtual classroom interaction method based on knowledge representation and reasoning provided in the above embodiment of the present invention by running the computer program in the memory 103 .

Claims

1. A virtual classroom interaction method based on knowledge representation and reasoning, characterized by: The method comprises: Obtaining a student interaction data stream in a power system training virtual classroom, wherein the student interaction data stream includes time-sequentially arranged question texts, simulation operation records, and historical conversation logs; Performing knowledge granularity decomposition processing on the student interaction data stream to generate a knowledge tuple set including concept nodes and relationship edges, wherein the concept nodes correspond to basic concepts of the power system, and the relationship edges reflect the logical associations between concepts and the contextual dependencies of student interaction behaviors; Input the knowledge tuple set into the inference engine to perform multi-dimensional inference processing to generate a student knowledge mastery state vector and a learning needs prediction vector. The knowledge mastery state vector includes indicators of concept understanding depth, operational standardization, and logical coherence, and the learning needs prediction vector includes content supplementation direction and ability improvement direction. Performing interactive content generation processing based on the knowledge mastery state vector and the learning demand prediction vector to generate adaptive classroom interactive response data, wherein the interactive response data includes knowledge completion information and ability training tasks; The classroom interaction response data is fed back to the virtual classroom interaction interface to form a dynamic interaction process between students' learning status and classroom feedback.

2. The virtual classroom interaction method based on knowledge representation and reasoning according to claim 1, characterized in that: The step of performing knowledge granularity decomposition on the student interaction data stream to generate a knowledge tuple set containing concept nodes and relationship edges includes: Performing semantic segmentation processing on the question text in the student interaction data stream to extract semantic segments containing power system terminology, wherein the semantic segments include device names, operating parameters, and fault phenomena; Performing behavior sequence disassembly processing on the simulation operation record to extract the correspondence between the operation action and the operation object, wherein the correspondence includes the binding relationship between the device selection action and the target device, and the constraint relationship between the parameter adjustment action and the parameter range; Performing context chain construction on the historical conversation logs to extract semantic progressive relationships in continuous conversations, wherein the semantic progressive relationships include concept extension paths and question deepening trajectories; The semantic fragments, the corresponding relationship between the operation actions and the operation objects, and the semantic progressive relationship are used as basic knowledge elements; Calculating the association degree of the basic knowledge elements to identify the conceptual association strength between semantic segments, the matching degree between the operation actions and the content of the semantic segments, and the temporal consistency between the semantic progression relationship and the operation actions; The basic knowledge elements are clustered and merged according to the correlation calculation result to generate a knowledge tuple set with power system concepts as nodes and correlation strength matching consistency as weights.

3. The virtual classroom interaction method based on knowledge representation and reasoning according to claim 2, characterized in that: The step of performing semantic segmentation on the question text in the student interaction data stream to extract semantic segments containing power system terms includes: Performing word segmentation and filtering processing on the question text, and retaining professional vocabulary related to the power system training objectives as candidate terms; Performing semantic role labeling on the candidate terms to identify the semantic function of each candidate term in the question, wherein the semantic function includes a subject function, an attribute function, and a behavior function, wherein the subject function corresponds to a device name, the attribute function corresponds to an operating parameter, and the behavior function corresponds to a protection mechanism; Performing adjacency recognition processing on candidate terms with semantic functional associations to extract direct semantic connections between terms, including binding relationships between devices and attributes and triggering relationships between devices and behaviors; Performing context verification on the direct semantic connection, combining the overall semantic orientation of the question text, and screening out valid semantic connections that conform to the power system knowledge system; The candidate terms and their corresponding valid semantic connections are combined to generate semantic segments containing power system terms.

4. The virtual classroom interaction method based on knowledge representation and reasoning according to claim 2, characterized in that: The performing behavior sequence disassembly processing on the simulation operation record to extract the correspondence between the operation action and the operation object includes: Performing time axis alignment processing on the simulation operation record to generate an action event list arranged in the order of operation time; Performing type identification processing on each action event in the action event list to distinguish between device operation actions, parameter adjustment actions, and fault simulation actions; Perform logical dependency analysis on adjacent action events to identify constraints imposed by preceding actions on subsequent actions. These constraints include the inability to adjust parameters if the device is not started and the inability to perform recovery operations if the fault is not cleared. Perform pattern abstraction on continuous action events with the same operation target, extracting recurring action combinations as operation sub-patterns. The operation sub-patterns include a basic operation chain for equipment startup parameter setting and operation monitoring, and a fault handling chain for observing and troubleshooting fault triggering phenomena. The type information of the action event, the constraint conditions of the logical dependency, and the combination rules of the operation sub-mode are integrated to generate a corresponding relationship between the operation action and the operation object.

5. The virtual classroom interaction method based on knowledge representation and reasoning according to claim 2, characterized in that: The process of constructing a context chain on the historical conversation log to extract semantic progressive relationships in the continuous conversation includes: Dividing the historical conversation log into turns to generate a conversation turn set consisting of student questions and system feedback; Perform semantic vector conversion on each conversation turn to generate a semantic vector that reflects the core content of the turn; Calculate the cosine similarity between the semantic vectors of adjacent conversation turns and identify consecutive turns with close semantic connections as context chains; Performing differential analysis on the semantic vectors in the context chain to extract semantic changes from the initial question to subsequent questions, wherein the semantic changes include the expansion of concept scope and the increase in question depth; Performing direction determination processing on the semantic change amount to determine whether the semantic progression path conforms to the logical structure of the power system knowledge system; The association information of the context chain, the numerical characteristics of the semantic change amount, and the path direction of the semantic progression are integrated to generate a semantic progression relationship in a continuous dialogue.

6. The virtual classroom interaction method based on knowledge representation and reasoning according to claim 1, characterized in that: The step of inputting the knowledge tuple set into the inference engine to perform multi-dimensional inference processing to generate a student knowledge mastery state vector and a learning needs prediction vector includes: Processing the concept nodes and relationship edges in the knowledge tuple set through the concept understanding reasoning module of the reasoning engine, calculating the students' coverage and relevance of power system concepts, and generating a concept understanding depth index, where coverage is the ratio of the number of concepts mastered to the total number of concepts, and relevance is the ratio of the number of correct relevance relationships to the total number of relevance relationships; The operation specification reasoning module of the inference engine processes the correspondence between the operation actions and the operation objects in the knowledge tuple set, calculates the matching rate of the operation sub-pattern and the standard operation process and the constraint compliance rate, and generates an operation standardization index, wherein the matching rate is the ratio of the number of matching sub-patterns to the total number of sub-patterns, and the compliance rate is the ratio of the number of actions that meet the constraint conditions to the total number of actions; Processing the semantic progressive relationship in the knowledge tuple set through the logical coherence reasoning module of the reasoning engine, calculating the degree of consistency between the semantic progressive path and the logical structure of the knowledge system and the rationality of the semantic change, and generating a logical coherence index, wherein the degree of consistency is the ratio of the length of the consistent path to the total path length, and the rationality is the ratio of the number of changes that conform to the law of knowledge deepening to the total number of changes; Normalizing the concept understanding depth index, the operation standardization index, and the logical coherence index to generate a student knowledge mastery state vector; Analyze the index items in the knowledge mastery state vector that are lower than the preset threshold, determine the direction of knowledge content that needs to be supplemented and the direction of operation links that need to be strengthened, and generate a learning demand prediction vector.

7. The virtual classroom interaction method based on knowledge representation and reasoning according to claim 6 is characterized in that: The concept understanding reasoning module of the reasoning engine processes the concept nodes and relationship edges in the knowledge tuple set and calculates the coverage and relevance of the students' power system concepts, including: Extracting all concept nodes from the knowledge tuple set, counting their number as the number of mastered concepts, and obtaining a standard concept set corresponding to the power system training objective, counting their number as the total number of concepts; Calculate the ratio of the number of concepts mastered to the total number of concepts as concept coverage; Extracting all relationship edges from the knowledge tuple set and counting the number of correct association relationships that conform to the power system knowledge system; Counting the total number of relationship edges in the knowledge tuple set, and calculating the ratio of the number of correct association relationships to the total number of relationship edges as the concept association degree; The concept coverage and the concept association are integrated to generate a concept understanding depth index.

8. The virtual classroom interaction method based on knowledge representation and reasoning according to claim 6, characterized in that: The operation specification reasoning module of the reasoning engine processes the correspondence between the operation actions and the operation objects in the knowledge tuple set, and calculates the matching rate and constraint compliance rate between the operation sub-mode and the standard operation process, including: Extracting the operation sub-patterns in the knowledge tuple set, counting the number of the operation sub-patterns as the total number of sub-patterns, and obtaining the standard operation process corresponding to the power system simulation experiment, extracting the standard sub-patterns therein, and counting the number of the standard sub-patterns as the number of matching sub-patterns; Calculating the ratio of the number of matching sub-patterns to the total number of sub-patterns as the operation sub-pattern matching rate; Extracting logical dependency constraints from the knowledge tuple set, counting the number of operation actions that meet the constraints as the number of actions that meet the constraints, and counting the total number of action events in the knowledge tuple set; Calculating the ratio of the number of actions that meet the constraint condition to the total number of action events as the constraint condition compliance rate; Integrating the operation sub-pattern matching rate and the constraint condition compliance rate to generate an operation standardization index; The processing of the semantic progressive relationship in the knowledge tuple set by the logical coherence reasoning module of the reasoning engine and the calculation of the degree of consistency between the semantic progressive path and the logical structure of the knowledge system and the rationality of the semantic change amount include: Extracting the semantic progressive path from the knowledge tuple set, counting its length as the total path length, obtaining the standard logical structure of the power system knowledge system, and counting the number of rounds in the semantic progressive path that are consistent with the standard structure as the matching path length; Calculating the ratio of the matching path length to the total path length as the logical structure matching degree; Extracting semantic changes in the knowledge tuple set, counting the number of changes that conform to the knowledge deepening law as the reasonable number of changes, and counting the total number of semantic changes in the knowledge tuple set; Calculating the ratio of the reasonable change amount to the total semantic change amount as the semantic change rationality; The logical structure consistency and the semantic change rationality are integrated to generate a logical coherence index.

9. The virtual classroom interaction method based on knowledge representation and reasoning according to claim 1, characterized in that: The performing of interactive content generation processing based on the knowledge mastering state vector and the learning demand prediction vector to generate adaptive classroom interactive response data includes: For the concept understanding depth index in the knowledge mastery state vector, if the index is lower than a first preset threshold, extracting the basic definition, related concepts and typical application cases of the corresponding concept from the power system knowledge base to generate knowledge completion information; For the operation standardization index in the knowledge mastery state vector, if the index is lower than a second preset threshold, extracting the standard process, common errors and correction methods of the corresponding operation from the simulation operation standard library to generate operation standard reinforcement information; For the logical coherence index in the knowledge mastery state vector, if the index is lower than a third preset threshold, extracting the logical deduction process, associated verification method, and typical problem chain of the corresponding knowledge module from the knowledge system logic library to generate logical coherence reinforcement information; Based on the content supplement direction in the learning demand prediction vector, combined with the knowledge completion information, the operation specification reinforcement information, and the logical connection reinforcement information, content-adapted guided explanation data is generated; Based on the ability improvement direction in the learning needs prediction vector, comprehensive application questions, cross-module collaboration questions, and system-level analysis questions that are connected with the mastered knowledge are extracted from the ability training task library to generate ability-adapted training task data; The guided explanation data and the training task data are integrated with each other in content and adapted in format to generate adaptive classroom interaction response data.

10. A computer system, characterized in that: include: a memory storing a computer program; A processor, configured to load the computer program to implement the virtual classroom interaction method based on knowledge representation and reasoning according to any one of claims 1 to 9.

Images