Illusion suppression method and device for power transmission line fault diagnosis model

Abstract

This application discloses a method and apparatus for suppressing hallucinations in transmission line fault diagnosis models, relating to the field of power line fault diagnosis technology. The method includes: upon receiving a fault diagnosis command, performing intent analysis on the command; and, if the intent of the command is deemed ambiguous, clarifying the diagnostic object based on preset question-and-answer constraints; extracting real data and procedural basis corresponding to the diagnostic object from a preset real-time information database and a preset knowledge base, respectively; performing physical logic verification on the real data based on a judgment model to obtain the constraints that the model diagnosis must adhere to; and inputting the real data, procedural basis, and constraints into a preset fault diagnosis model, enabling the model to perform causal inference on the real data based on the constraints and procedural basis to obtain the diagnostic result. This method suppresses hallucinations of high confidence but low reliability caused by ambiguous boundaries of the diagnostic object, incomplete parameters, and violations of physical logic.

Description

Technical Field

[0001] This application relates to the field of power line fault diagnosis technology, and in particular to a method and apparatus for suppressing hallucinations in power transmission line fault diagnosis models. Background Technology

[0002] With the deepening of the digital transformation of the power grid, large model technology has been widely introduced into line fault diagnosis to build intelligent diagnostic methods.

[0003] However, intelligent diagnostic methods based on large models generally suffer from model illusion in practical applications. That is, when dealing with scenarios involving multiple lines and multiple devices, the boundaries of the diagnostic objects identified by the large model are blurred. Furthermore, the large model is prone to mixing in irrelevant line parameters, historical abnormal data, or missing key parameters during the data retrieval process. In addition, when diagnosing faults in power lines, the large model is prone to violating common sense about power operation. Therefore, the diagnostic conclusions generated by the large model have high confidence but low reliability. Summary of the Invention

[0004] The main objective of this application is to provide a method and apparatus for suppressing hallucinations in a fault diagnosis model for transmission lines, aiming to solve the technical problem that the hallucination suppression model generates diagnostic conclusions with high confidence but low reliability.

[0005] To achieve the above objectives, this application proposes a method for hallucination suppression in transmission line fault diagnosis models, the method comprising: When a fault diagnosis instruction is received, the intent of the fault diagnosis instruction is analyzed, and when the intent of the fault diagnosis instruction is determined to be ambiguous, the diagnostic object is clarified based on preset question and answer prompt constraints. The real data and procedural basis corresponding to the diagnostic object are extracted from the preset real-time information database and the preset knowledge base, respectively. Based on the analysis model, physical and logical verification is performed on the real data to obtain the constraints that the model diagnosis needs to comply with. The real data, the procedural basis, and the constraints are input into a preset fault diagnosis model, which then performs causal deduction on the real data based on the constraints and the procedural basis to obtain a diagnosis result.

[0006] In one embodiment, the step of performing physical and logical verification on the real data based on the judgment model to obtain the constraints that the model diagnosis needs to comply with includes: Based on the analysis model, the real data is organized into a chain of evidence corresponding to power failure scenarios; Based on the mapping rules between the preset expert judgment criteria and the model reasoning logic, the physical consistency and procedural compliance of the evidence chain are verified to obtain the verification results. If the verification result is a verification failure, the structured constraint template corresponding to the mapping rule is invoked to generate constraint conditions for anchoring the inference path of the fault diagnosis model.

[0007] In one embodiment, before the step of performing physical consistency and procedural compliance checks on the evidence chain based on the mapping rules between preset expert judgment criteria and model reasoning logic to obtain the check results, the method further includes: Obtain expert judgment criteria in the field of wire fault diagnosis; The expert judgment criteria are analyzed as a formal reasoning process that includes observation conditions, causal inference steps, and compliance boundaries. Based on the formal reasoning process, structured data slots and instruction slots are constructed. The data slots are used to fill in real data, and the instruction slots are used to guide the model to reason step by step and to verify compliance. The variables in the expert judgment criteria are bound to the data slots, and the logical relationships in the expert judgment criteria are bound to the instruction slots to obtain mapping rules.

[0008] In one embodiment, the step of inputting the real data, the procedural basis, and the constraints into a preset fault diagnosis model, and enabling the fault diagnosis model to perform causal inference on the real data based on the constraints and the procedural basis to obtain a diagnostic result, includes: The constraints are input into a preset fault diagnosis model, which then determines the physical and logical boundaries of the reasoning. The real data and the procedural basis are input into the fault diagnosis model, so that the fault diagnosis model, within the physical and logical boundaries, performs semantic association and logical organization reasoning on the real data based on the reasoning path corresponding to the procedural basis, and obtains the diagnosis result.

[0009] In one embodiment, the step of extracting real data and procedural basis corresponding to the diagnostic object from a preset real-time information database and a preset knowledge base, respectively, includes: Retrieve physical parameters associated with the diagnostic object from a preset real-time information database; Retrieve the applicable procedural guidelines from the knowledge base, the procedural guidelines including the types of parameters necessary for fault diagnosis and the integrity requirements; Based on the integrity requirements and the parameter type, the physical parameters are verified to obtain the verification results; If the verification result indicates that the physical parameters are complete and conform to the acquisition specifications, then the physical parameters are extracted as the actual data of the diagnostic object, and the procedure basis is extracted.

[0010] In one embodiment, the step of extracting real data and procedural basis corresponding to the diagnostic object from a preset real-time information database and a preset knowledge base, respectively, further includes: If the real-time information database does not contain physical parameters associated with the diagnostic object, or the knowledge base does not contain procedural guidelines applicable to the diagnostic object, or the verification result indicates that the physical parameters are incomplete, a missing item alert will be triggered, and the extraction of real data and procedural guidelines will be suspended.

[0011] In one embodiment, the step of receiving a fault diagnosis instruction, performing intent analysis on the fault diagnosis instruction, and clarifying the diagnostic object based on preset question-and-answer prompt constraints when the intent of the fault diagnosis instruction is determined to be ambiguous, further includes: Upon receiving a fault diagnosis instruction, the instruction is semantically parsed based on a preset power industry prompt word framework to extract the question intent from the instruction. Determine whether the intent of the question is complete and unambiguous; If the question intent is incomplete or ambiguous, then push equipment information and power grid topology sub-graph related to the fault diagnosis command to the user to guide the user to complete the question intent; In response to the user's completed question intent and confirmation feedback, the question intent is normalized into a uniquely determined physical device object and its corresponding spatiotemporal boundary; Based on the physical device object and the spatiotemporal boundary, the diagnostic object is identified.

[0012] In one embodiment, the intent of the inquiry includes: device identification, spatiotemporal range, and fault phenomenon elements.

[0013] Furthermore, to achieve the above objectives, this application also proposes a hallucination suppression device for a transmission line fault diagnosis model, the hallucination suppression device for the transmission line fault diagnosis model comprising: The question-and-answer constraint module is used to analyze the intent of a fault diagnosis instruction when it is received, and to clarify the diagnosis object based on preset question-and-answer prompt constraints when the intent of the fault diagnosis instruction is determined to be ambiguous. The extraction module is used to extract real data and procedural basis corresponding to the diagnostic object from a preset real-time information database and a preset knowledge base, respectively. The verification module is used to perform physical and logical verification on the real data based on the judgment model to obtain the constraints that the model diagnosis needs to comply with. The diagnostic module is used to input the real data, the procedural basis, and the constraints into a preset fault diagnosis model, so that the fault diagnosis model can perform causal inference on the real data based on the constraints and the procedural basis to obtain a diagnostic result.

[0014] Furthermore, to achieve the above objectives, this application also proposes an illusion suppression device for a transmission line fault diagnosis model, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the illusion suppression method for a transmission line fault diagnosis model as described above.

[0015] One or more technical solutions proposed in this application have at least the following technical effects: Upon receiving a fault diagnosis command, the intent of the command is first analyzed. If the intent is ambiguous, the diagnostic object is clarified based on preset question-and-answer prompts. Then, real data and procedural basis corresponding to the diagnostic object are extracted from the preset real-time information database and the preset knowledge base, respectively. The real data is physically and logically verified by the analysis model to generate the constraints that the model diagnosis must comply with. Finally, the real data, procedural basis and constraints are input into the preset fault diagnosis model. Under the dual constraints of common sense in power operation and professional procedures, the model performs causal inference on the real data to obtain the diagnostic result. This suppresses the illusion of high confidence but low reliability caused by the ambiguous boundaries of the diagnostic object, the inclusion of irrelevant or missing key parameters in the retrieval, and the violation of common sense in power operation in multi-line and multi-equipment interconnected scenarios. In other words, this application ensures that the fault diagnosis model is based solely on real-time data and authoritative procedures that are strongly related to the current diagnostic object through a closed-loop process of intent clarification, data focus, logical verification, and constraint reasoning. This avoids interference from irrelevant historical anomalies, redundant line parameters, or missing key measurements on the diagnostic process. Furthermore, the constraints generated by the judgment model are explicitly embedded in the physical laws of the power system, forcing the fault diagnosis model to output causal chains that conform to common sense in operation. Ultimately, the credibility of the diagnostic conclusions is improved without sacrificing the semantic understanding capabilities of the large model. Attached Figure Description

[0016] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating an embodiment of the hallucination suppression method for a transmission line fault diagnosis model provided in this application. Figure 2 This is a flowchart illustrating Embodiment 2 of the hallucination suppression method for a transmission line fault diagnosis model provided in this application; Figure 3 This application provides a flowchart for Embodiment 3 of the hallucination suppression method for a transmission line fault diagnosis model. Figure 4 This is a flowchart illustrating Embodiment 4 of the hallucination suppression method for a transmission line fault diagnosis model provided in this application; Figure 5 This is a schematic diagram of the module structure of the hallucination suppression device for a transmission line fault diagnosis model according to an embodiment of this application; Figure 6 This is a schematic diagram of the equipment structure of the hardware operating environment involved in the illusion suppression method for the transmission line fault diagnosis model in the embodiments of this application.

[0019] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0020] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0021] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0022] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or an electronic device or power transmission line fault diagnosis device capable of performing the above functions. The following description uses a power transmission line fault diagnosis device as an example to illustrate this embodiment and the subsequent embodiments.

[0023] Based on this, embodiments of this application provide a method for hallucination suppression in a transmission line fault diagnosis model, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the illusion suppression method for a transmission line fault diagnosis model according to this application.

[0024] In this embodiment, the hallucination suppression method for the transmission line fault diagnosis model includes steps S10~S40: Step S10: When a fault diagnosis instruction is received, the intent of the fault diagnosis instruction is analyzed, and when the intent of the fault diagnosis instruction is determined to be ambiguous, the diagnosis object is clarified based on the preset question and answer prompt constraints. It should be noted that fault diagnosis commands are natural language instructions or structured query commands issued by maintenance personnel, monitoring systems, or automation platforms, requesting fault analysis and location of a specific device or area in the power system. Intent analysis is the process of semantically parsing the fault diagnosis command to identify the implicit or explicit diagnostic target, spatiotemporal scope, and fault phenomenon description. Intent ambiguity occurs when a fault diagnosis command lacks key elements, contains ambiguities, or has multiple interpretations, making it impossible to uniquely determine the diagnostic object. Question-and-answer prompt constraints are a set of interactive guidance rules pre-built into the transmission line fault diagnosis equipment, constructed based on power industry expertise. These rules automatically generate structured questions or push relevant information when user input is incomplete, thus constraining the user to complete the intent. The diagnostic object is a uniquely identified physical entity and its corresponding spatiotemporal boundaries, determined after intent clarification.

[0025] Understandably, fault diagnosis commands issued by maintenance personnel in emergency situations often contain incomplete descriptions, unclear equipment designations, or vague symptom descriptions. Directly inputting such ambiguous commands into the fault diagnosis model can easily lead to speculative and divergent outputs due to a lack of clear objectives. By first analyzing the intent of the received fault diagnosis command and proactively triggering a pre-defined question-and-answer constraint mechanism when the intent is deemed ambiguous, the system guides the user to accurately complete key elements such as equipment identification, spatiotemporal range, and fault symptoms. This normalizes the ambiguous natural language request into a uniquely defined physical device object and its spatiotemporal boundaries, preventing the fault diagnosis model from blindly reasoning without a clear objective. Furthermore, it improves the efficiency and accuracy of human-computer interaction, ensuring that all subsequent diagnostic processes revolve around a clear and unambiguous diagnostic object, fundamentally avoiding location illusions caused by input ambiguity.

[0026] Step S20: Extract the real data and procedural basis corresponding to the diagnostic object from the preset real-time information database and the preset knowledge base, respectively; It should be noted that the real-time information database is a time-series database storing power system operating status data. The knowledge base is a structured database storing power industry regulations, standards, typical fault cases, and equipment parameters. Real data is the raw or pre-processed operating data extracted from the real-time information database, corresponding to the identified diagnostic object within a specified time and space range, and has passed integrity and standardization checks. The procedural basis is the normative content retrieved from the knowledge base, applicable to the current diagnostic object, including fault judgment logic, threshold standards, and processing procedures.

[0027] Understandably, since the accuracy of fault diagnosis highly depends on the authenticity and authority of the input information, relying solely on a single data source would cause the fault diagnosis model's reasoning to lose its objective anchor or compliance standards, easily leading to illusions that are detached from reality or violate standards. By simultaneously extracting real data and procedural basis that strictly correspond to the clearly identified diagnostic object from a pre-set real-time information database and knowledge base, the model effectively prevents fictitious diagnostic results caused by missing or distorted information. This is because real data reflects the current actual operating state of the equipment, while procedural basis provides industry-recognized technical standards and processing logic.

[0028] Step S30: Perform physical and logical verification on the real data based on the judgment model to obtain the constraints that the model diagnosis needs to comply with; It should be noted that the judgment model is a lightweight rule engine or a small-scale neural symbolic hybrid model trained based on expert judgment criteria. It is used to perform physical consistency verification and procedural compliance checks, including dedicated judgment models for lightning strikes, wildfires, and icing. Physical logic verification, based on the fundamental physical laws of power systems and equipment operating mechanisms, verifies the rationality of the real data and identifies abnormal patterns that violate physical laws. Constraints are hard boundary conditions generated by the judgment model to limit the inference path of the fault diagnosis model.

[0029] Understandably, even with real data, it may contain noise, anomalies, or logical inconsistencies caused by sensor malfunctions. If a fault diagnosis model directly uses unverified real data for free reasoning, its output may still violate fundamental physical laws such as energy conservation and circuit topology, creating a typical illusion. Therefore, an independent judgment model is specifically designed to perform physical-logical consistency and procedural compliance checks on real data. Based on this, it generates specific, executable constraints, essentially setting up a safety barrier for the subsequent fault diagnosis model. That is, through the preliminary physical-logical verification, abstract expert experience is transformed into concrete constraints that the fault diagnosis model can understand, anchoring the model's reasoning path within the real laws of the physical world and fundamentally suppressing the generation of erroneous conclusions that violate common sense and mechanisms.

[0030] Step S40: Input the real data, the procedural basis, and the constraints into a preset fault diagnosis model, so that the fault diagnosis model can perform causal inference on the real data based on the constraints and the procedural basis to obtain a diagnosis result.

[0031] It should be noted that the fault diagnosis model is an intelligent diagnostic engine built on a large language model, possessing causal reasoning and natural language generation capabilities. Its reasoning process is strictly limited by external constraints and procedural rules. Causal deduction, under given constraints, combines procedural logic to gradually trace back possible root causes of faults from observed data.

[0032] Understandably, while advanced AI technologies such as large language models possess powerful pattern recognition and text generation capabilities, their inherent black-box nature makes them prone to sacrificing factual accuracy in pursuit of fluency during open-domain reasoning. By using verified real data, authoritative procedural guidelines, and hard constraints generated by the judgment model as input, the fault diagnosis model is forced to simultaneously satisfy physical feasibility and procedural compliance during reasoning. In other words, by constructing a constraint-driven causal reasoning paradigm, the intelligent advantages of the fault diagnosis model can be leveraged within a safe and reliable framework. The final diagnostic results, following a clear causal chain and procedures, possess high interpretability and the reliability required for engineering practice.

[0033] This embodiment provides a method for suppressing illusions in a fault diagnosis model for transmission lines. Upon receiving a fault diagnosis command, the method first performs intent analysis on the command. If the intent is ambiguous, the diagnostic object is clarified based on preset question-and-answer prompts. Then, real data and procedural basis corresponding to the diagnostic object are extracted from a preset real-time information database and a preset knowledge base, respectively. The real data is then physically and logically verified by the analysis model to generate the constraints that the model diagnosis must comply with. Finally, the real data, procedural basis, and constraints are input into the preset fault diagnosis model. Under the dual constraints of common sense in power operation and professional procedures, the model performs causal inference on the real data to obtain the diagnostic result. This method suppresses the illusion of high confidence but low reliability caused by ambiguous boundaries of the diagnostic object, irrelevant or missing key parameters in the retrieval, and violation of common sense in power operation scenarios involving multiple lines and multiple devices. In other words, this application ensures that the fault diagnosis model is based solely on real-time data and authoritative procedures that are strongly related to the current diagnostic object through a closed-loop process of intent clarification, data focus, logical verification, and constraint reasoning. This avoids interference from irrelevant historical anomalies, redundant line parameters, or missing key measurements on the diagnostic process. Furthermore, the constraints generated by the judgment model are explicitly embedded in the physical laws of the power system, forcing the fault diagnosis model to output causal chains that conform to common sense in operation. Ultimately, the credibility of the diagnostic conclusions is improved without sacrificing the semantic understanding capabilities of the large model.

[0034] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to that in Embodiment 1 above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 2Step S30 also includes steps S01 to S03: Step S01: Based on the judgment model, organize the real data into an evidence chain corresponding to the power failure scenario; Step S02: Based on the preset mapping rules between expert judgment criteria and model reasoning logic, perform physical consistency and procedural compliance verification on the evidence chain to obtain the verification results; Step S03: If the verification result is a verification failure, the structured constraint template corresponding to the mapping rule is invoked to generate constraint conditions for anchoring the inference path of the fault diagnosis model.

[0035] It should be noted that power failure scenarios are a set of failure types that occur frequently in power system operation and have standardized characteristic patterns. The evidence chain is a structured sequence of events organized according to time sequence and logical correlation of real data, serving as a set of observable evidence to characterize whether a typical power failure scenario has occurred. Expert judgment criteria are a set of technical rules summarized by experts in the power system field to determine the type and cause of failures. Model reasoning logic is the implicit or explicit reasoning mechanism within the fault diagnosis model used to generate diagnostic conclusions. Mapping rules are a set of structured rules that formalize the expert judgment criteria and establish correspondences with key nodes in the model reasoning logic. Physical consistency verification is a rationality test of the evidence chain based on the fundamental physical laws of the power system to determine whether it violates objective physical laws. Procedure compliance verification is a compliance review of the occurrence order, threshold range, and combination logic of each event in the evidence chain based on current power industry technical regulations, countermeasure requirements, or equipment operation specifications. The verification result is a comprehensive output of physical consistency verification and procedure compliance verification, including verification pass and verification failure. The structured constraint template is a predefined constraint generation framework that corresponds one-to-one with the mapping rules. It is used to dynamically instantiate into specific hard constraints when verification fails. Anchoring the inference path involves injecting hard constraints to limit the state space that the fault diagnosis model can explore during causal deduction. This forces the inference process to be constrained to a physically feasible and procedurally compliant subset, preventing it from diverging into illusory regions.

[0036] Furthermore, the mapping rules upon which the model relies can be non-static and fixed, but rather support online learning and iterative optimization. After each diagnostic task, the model's output is compared with the scheduler's manual review conclusion. If the model is found to still output incorrect conclusions under all constraints, the rule optimization engine is triggered to analyze the root cause of the error and automatically generate new formal reasoning process fragments. After expert review, these fragments are merged into the existing mapping rule base, enabling the continuous evolution of the constraint system.

[0037] Understandably, by directly inputting raw, real data into the fault diagnosis model, the model may only focus on the statistical correlation between data points, ignoring the underlying physical causal mechanisms. This could lead to seemingly reasonable but actually illusory conclusions that contradict the operating mechanisms of the power system. By first organizing real data into a chain of evidence corresponding to typical fault scenarios, subsequent verification can be conducted within a context with a clear semantic structure, thereby improving the relevance and efficiency of the verification.

[0038] Understandably, while expert judgment criteria are authoritative, they are difficult to directly utilize by black-box models (large language models). By constructing mapping rules between expert judgment criteria and model reasoning logic, and decomposing these rules into data slots and instruction slots, formal embedding of domain knowledge is achieved. This allows abstract expert experience to be precisely applied to key decision points in model reasoning, ensuring that the fault diagnosis model is constrained by expert experience when reasoning about faults based on real data, thus avoiding illusions caused by the model's divergent thinking.

[0039] Understandably, simply issuing a warning after a verification failure without intervening in the model's inference will not prevent hallucinations. By automatically invoking the corresponding structured constraint template upon verification failure, and dynamically generating hard constraints to anchor the inference path of the fault diagnosis model, it is equivalent to setting an inviolable red line for the model. This fundamentally constrains the inference process within the compliant framework of the real laws of the physical world and industry regulations, thereby improving the credibility of the diagnostic results.

[0040] It should be noted that the mapping rules can support multi-granularity nested structures to achieve fine-grained constraints on complex fault evolution paths.

[0041] Optionally, when the verification result is "verification failed" and the original output of the model is finally confirmed to be wrong by manual review, the power transmission line fault diagnosis equipment can automatically record the failure mode and use symbolic regression or rule mining algorithms to summarize new mapping rule fragments from historical failure cases. After expert review, the rules are incrementally updated to the rule base to realize the self-learning evolution of the constraint system.

[0042] In practical implementation, when the judgment model verifies the fault hypothesis of "insulator flashover", it can also actively query the lightning location system data in the meteorological database. If there are high-density lightning strike records at the time and location of the fault, the initial confidence weight of the fault type is increased and the leniency of the constraints is adjusted accordingly, thereby realizing adaptive diagnosis driven by environmental perception.

[0043] Optionally, prior to step S02, the hallucination suppression method for the transmission line fault diagnosis model further includes: Obtain expert judgment criteria in the field of wire fault diagnosis; The expert judgment criteria are analyzed as a formal reasoning process that includes observation conditions, causal inference steps, and compliance boundaries. Based on the formal reasoning process, structured data slots and instruction slots are constructed. The data slots are used to fill in real data, and the instruction slots are used to guide the model to reason step by step and to verify compliance. The variables in the expert judgment criteria are bound to the data slots, and the logical relationships in the expert judgment criteria are bound to the instruction slots to obtain mapping rules.

[0044] It should be noted that observation conditions are the prerequisite data states or environmental characteristics that must be met before fault diagnosis can be performed. The causal inference step is an ordered sequence of reasoning that starts from observable phenomena and, based on physical mechanisms or procedural logic, gradually derives the root cause of the fault. Compliance boundaries are the implicit or explicit numerical thresholds, time windows, logical combination restrictions, and other compliance requirements in the expert judgment criteria. The inference process transforms unstructured expert judgment criteria into a structured, computable inference model with clear inputs, processing logic, and outputs, including three core elements: observation conditions, causal inference steps, and compliance boundaries. Data slots are placeholders or variable containers pre-defined in the formal inference process for receiving and storing real monitoring data. Each slot corresponds to a specific physical quantity and has clear data type, unit, and time alignment requirements. Instruction slots are structured units pre-defined in the formal inference process for carrying inference control instructions and compliance verification logic. Each instruction slot contains an executable judgment statement or verification function to guide the model to advance the inference step-by-step according to expert logic. Variable binding is the process of establishing a one-to-one correspondence between the names of physical quantities described in the expert judgment criteria and the specific data slots in the formal reasoning process, ensuring that real data can be accurately filled into the correct positions in the reasoning process. Logical relationship binding is the process of mapping the causal, conditional, temporal, or threshold relationships contained in the expert judgment criteria to the corresponding instruction slots, forming logical expressions that can be parsed and executed by the computing engine.

[0045] Understandably, by resolving expert judgment criteria into a formal reasoning process that includes observation conditions, causal inference steps, and compliance boundaries, the structure and computability of expert knowledge are realized, laying the foundation for subsequent collaboration with fault diagnosis models.

[0046] Understandably, if expert criteria are only input as textual prompts into the fault diagnosis model, the model may still overlook its key logical details. By constructing structured data slots and instruction slots, and binding the variables and logical relationships in the expert criteria to the corresponding slots, expert knowledge can be transformed from vague background information into explicit control signals and verification nodes driving model reasoning. This allows for the rational utilization of expert experience, enabling the fault diagnosis model to deduce faults based on physical rules and actual procedures, thereby reducing illusions at the deduction level.

[0047] Optionally, the formal reasoning process can support dynamic branch pruning. When a data slot cannot be filled due to communication interruption, instruction slots that depend on that slot can be automatically skipped, and alternative criterion paths can be activated, thereby improving availability in scenarios where some data is missing.

[0048] Furthermore, step S40 also includes: The constraints are input into a preset fault diagnosis model, which then determines the physical and logical boundaries of the reasoning. The real data and the procedural basis are input into the fault diagnosis model, so that the fault diagnosis model, within the physical and logical boundaries, performs semantic association and logical organization reasoning on the real data based on the reasoning path corresponding to the procedural basis, and obtains the diagnosis result.

[0049] It should be noted that the physical logical boundary is a physically feasible region defined by the aforementioned constraints, which the fault diagnosis model must not traverse during the reasoning process. The reasoning path is the sequence of ordered reasoning steps that transforms the procedural basis into an executable sequence for the model. Semantic association is the process by which the fault diagnosis model identifies the semantic relationships between different physical quantities in real data and maps them to the fault mode semantic space. Logical organization reasoning, guided by both the physical logical boundary and the procedural reasoning path, involves the structured integration of semantic association results to form a diagnostic conclusion with a causal chain, traceability, and verifiability.

[0050] Understandably, if real data and procedural criteria are directly input into an unconstrained fault diagnosis model, the model might ignore physical inconsistencies in pursuit of linguistic fluency or pattern matching. By first inputting constraints to determine the physical and logical boundaries of the reasoning, the generation of unrealistic assumptions can be fundamentally prevented, thus avoiding the creation of illusions.

[0051] In practice, a small model based on expert experience is used as a "referee" to force the large model to perform logical reasoning within the scope of physical laws.

[0052] Specifically, the concurrent analysis and verification of expert experience mini-models involves calling pre-trained specialized analysis mini-models for lightning strikes, wildfires, and icing to perform physical-level logical verification of the evidence chain. Each mini-model, based on built-in expert judgment criteria, performs a second round of verification of fault characteristics and outputs preliminary diagnostic opinions with physical and logical support.

[0053] Anchoring mapping between semantic space and physical logic: Establishing mapping rules between expert judgment results and the generation logic of the large model. Through strong constraint instructions, the verification conclusion of the small model is set as the "logical boundary" of the large model, which forces the large model not to deviate from the physical qualitative results given by the small model during the reasoning process.

[0054] Evidence-driven logical organization and diagnostic conclusions: Within the pre-set logical boundaries, the large model connects fragmented chains of evidence into a deductive process with causal relationships, and combines it with the knowledge base of power regulations to finally arrive at a diagnostic result with physical interpretability, effectively avoiding logical illusions that violate common sense about electricity.

[0055] Based on the first and second embodiments of this application, the same or similar content as the above embodiments in the third embodiment of this application can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 3 Step S20 also includes steps S21 to S24: Step S21: Retrieve physical parameters associated with the diagnostic object from a preset real-time information database; Step S22: Retrieve the applicable procedural basis from the knowledge base, the procedural basis including the parameter types and integrity requirements necessary for fault diagnosis; Step S23: Based on the integrity requirements and the parameter type, verify the physical parameters and obtain the verification results; Step S24: If the verification result shows that the physical parameters are complete and meet the acquisition specifications, then the physical parameters are extracted as the real data of the diagnostic object, and the procedure basis is extracted.

[0056] It should be noted that physical parameters are electrical or state quantities that can be collected by sensors or monitoring devices and are related to the diagnostic object within a specified spatiotemporal range. The parameter type is the minimum dataset category specified in the procedure for supporting specific fault diagnosis. Integrity requirements are the minimum acceptable standards for physical parameters in terms of temporal continuity, spatial coverage, sampling quality, and synchronization accuracy, used to determine whether the acquired data is sufficient to support a reliable diagnosis.

[0057] Understandably, in power fault diagnosis, if incomplete or non-compliant data is used for analysis, even the most advanced model algorithms are prone to yielding erroneous or even dangerous conclusions. By retrieving physical parameters and procedural basis from real-time information databases and knowledge bases respectively, and performing automated verification based on the parameter types and integrity requirements clearly defined in the procedures, a strict data access mechanism is established at the front end of the diagnostic process, effectively preventing illusions caused by incomplete input data.

[0058] Understandably, by using procedures as reasoning rules and data quality acceptance criteria, task-driven data filtering is achieved. The real data extracted in this way meets the prerequisites for the diagnostic task, thereby improving the reliability of subsequent physical verification and causal inference.

[0059] In practical implementation, the integrity requirement can support dynamic weight evaluation. For example, when the zero-sequence current is missing due to CT failure, the system can reduce its weight in the "single-phase grounding" criterion and increase the substitution weight of the negative-sequence current or harmonic components. Combined with fuzzy logic, a "partially complete but diagnosable" verification result is generated, avoiding the interruption of the entire diagnostic process due to the failure of a single channel.

[0060] Furthermore, step S20 also includes: If the real-time information database does not contain physical parameters associated with the diagnostic object, or the knowledge base does not contain procedural guidelines applicable to the diagnostic object, or the verification result indicates that the physical parameters are incomplete, a missing item alert will be triggered, and the extraction of real data and procedural guidelines will be suspended.

[0061] It should be noted that the missing item reminder is a structured prompt message automatically generated by the power transmission line fault diagnosis equipment and pushed to the user terminal when data or rules are detected to be missing. It clearly indicates the type of missing item and suggests the correct action.

[0062] Understandably, in complex power systems, abnormal data acquisition or incomplete procedure coverage are common operating conditions. Continuing the diagnostic process without key inputs can easily lead to unfounded inferences based on incomplete information, resulting in illusory outputs that significantly deviate from reality and may even mislead dispatch decisions. By setting explicit criteria for missing information and triggering a missing information alert and halting extraction when any condition is met, the fault diagnosis model can be strictly prevented from filling in the missing information on its own.

[0063] In practice, real monitoring data is forcibly retrieved to replace the model's random speculation, ensuring the authenticity and reliability of the diagnostic basis.

[0064] Specifically, the automated extraction and alignment of multi-source heterogeneous data involves the transmission line fault diagnosis equipment using structured data such as current, voltage, and meteorological data from the centralized monitoring system, as well as unstructured information such as operation and maintenance records and manuals, based on the time and object determined from the fault diagnosis instructions. A spatiotemporal mapping algorithm is then used to uniformly associate this data with the fault location.

[0065] Dual-database verification and evidence extraction: Real-time monitoring values ​​are stored in a structured database, while power regulations and historical cases are stored in a vectorized knowledge base. The transmission line fault diagnosis equipment is forced to retrieve and match fault features from both databases, extracting "fault type confidence" and "core physical parameters" (such as current polarity, wind speed threshold, etc.) to ensure that all diagnostic evidence comes from database records.

[0066] Data Missing Item Alert and Deterministic Evidence Chain Generation: If key parameters cannot be found in the database, the power transmission line fault diagnosis equipment triggers a "data missing item alert" and notifies the user, strictly prohibiting the model from filling in the missing items automatically. Ultimately, the obtained real parameters are encapsulated into a structured evidence chain, serving as the sole truth reference for subsequent logical reasoning, preventing data fabrication.

[0067] Based on the above embodiments of this application, in the fourth embodiment of this application, the same or similar content as the above embodiments can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 4 Step S10 also includes steps S1 to S5: Step S1: When a fault diagnosis instruction is received, the fault diagnosis instruction is semantically parsed based on a preset power professional prompt word framework, and the question intent is extracted from the fault diagnosis instruction. Step S2: Determine whether the intent of the question is complete and unambiguous; Step S3: If the question intent is incomplete or ambiguous, push the device information and power grid topology sub-graph related to the fault diagnosis command to the user to guide the user to complete the question intent. Step S4: In response to the user's completed question intent and confirmation feedback, the question intent is normalized into a uniquely determined physical device object and its corresponding spatiotemporal boundary. Step S5: Based on the physical device object and the spatiotemporal boundary, identify the diagnostic object.

[0068] It should be noted that the power industry prompt framework is a pre-built, structured language model guidance system that includes a keyword library, semantic templates, and entity recognition rules for the power industry. This system is used to constrain and guide the semantic parsing process of fault diagnosis commands. The query intent is a set of core information extracted from the fault diagnosis command that represents the user's diagnostic needs, including three basic elements: equipment identification, spatiotemporal range, and fault phenomenon. Completeness means the query intent contains all the necessary elements sufficient to uniquely identify a diagnosable physical entity; unambiguity means each element is clearly stated, uniquely targeted, and free from multiple interpretations. For example, if any element is missing or has multiple possible interpretations, it is judged as "incomplete or ambiguous." Equipment information is a list of candidate equipment related to the keywords mentioned in the fault diagnosis command, including structured attributes such as equipment name, unique ID, voltage level, affiliated substation, and current operating status. The power grid topology sub-graph is a dynamically extracted local network diagram from the overall network topology model, containing candidate equipment and their electrical connections. It is used to visually display the location and relationships of equipment, assisting users in accurately locating target objects. The spatiotemporal boundary is the time window and spatial range bound to the physical device object. The time boundary is usually a few seconds before and after the time of the fault occurrence (e.g., 13:00±0.4); the spatial boundary is the device body and its direct electrical connection point.

[0069] Understandably, fault diagnosis commands input by power dispatchers or maintenance personnel in emergency situations are often brief, colloquial, and even ambiguous. Directly processing these commands by AI models could easily lead to misunderstandings due to input ambiguity, resulting in serious consequences. By introducing a semantic parsing mechanism based on a power industry-specific prompting framework and proactively assessing the completeness and ambiguity of the question's intent, the quality of user input is pre-evaluated. Furthermore, by pushing structured equipment information and visualized power grid topology sub-graphs to users, abstract information is transformed into clickable and verifiable options, significantly reducing the cognitive load of human-computer interaction and preventing errors in user-supplemented information. This avoids the fault diagnosis model experiencing illusions due to unclear diagnostic targets.

[0070] In practice, when a user inputs a vague or ambiguous diagnostic command, the power transmission line fault diagnosis device does not directly generate a conclusion, but instead locks the question-and-answer boundaries through a preset framework.

[0071] Specifically, the proposal intent parsing and element extraction process involves parsing the user's natural language input using a pre-defined framework of power industry-specific prompts and extracting key elements (such as line name, tower section, and time period). Semantic matching is then used to determine the completeness of the input information.

[0072] Ambiguity Identification and Proactive Questioning Guidance: If ambiguity is detected in the question / answer object (such as multiple line names being similar) or key elements are missing, the transmission line fault diagnosis equipment activates the feedback module. It guides the user to click or complete the key information by displaying a list of related devices in the topology diagram or popping up a confirmation option.

[0073] Diagnostic boundary locking and object alignment: Based on the confirmation results of user feedback, the power transmission line fault diagnosis equipment normalizes the query intent to specific physical devices and time-space axes, ensuring that all subsequent data retrieval and logical reasoning are locked within a uniquely defined object range, preventing location illusions caused by object ambiguity.

[0074] In practical implementation, to further improve the efficiency and accuracy of intent clarification, the question-and-answer prompt constraints are not limited to static lists, but can also dynamically integrate multimodal interaction capabilities. For example, when a user voice inputs "There's a problem at the wind farm," the transmission line fault diagnosis equipment can automatically identify geographical keywords, call GIS services to generate a local power grid topology heat map including the access lines of surrounding wind farms, and support the user to select via touch screen or confirm via voice, thereby accurately mapping the vague geographical description to specific device objects.

[0075] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the illusion suppression method for transmission line fault diagnosis models in this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0076] This application also provides a hallucination suppression device for a transmission line fault diagnosis model; please refer to... Figure 5 The hallucination suppression device for the transmission line fault diagnosis model includes: The question-and-answer constraint module 10 is used to perform intent analysis on the fault diagnosis instruction when a fault diagnosis instruction is received, and to clarify the diagnosis object based on preset question-and-answer prompt constraints when the intent of the fault diagnosis instruction is determined to be ambiguous. The extraction module 20 is used to extract real data and procedural basis corresponding to the diagnostic object from a preset real-time information database and a preset knowledge base, respectively. The verification module 30 is used to perform physical and logical verification on the real data based on the judgment model to obtain the constraints that the model diagnosis needs to comply with. The diagnostic module 40 is used to input the real data, the procedural basis, and the constraints into a preset fault diagnosis model, so that the fault diagnosis model can perform causal inference on the real data based on the constraints and the procedural basis to obtain a diagnostic result.

[0077] Optionally, the verification module 30 is further configured to organize the real data into an evidence chain corresponding to the power fault scenario based on the judgment model; perform physical consistency and procedural compliance verification on the evidence chain based on the preset mapping rules between expert judgment criteria and model reasoning logic, and obtain verification results; if the verification result is a verification failure, then call the structured constraint template corresponding to the mapping rules to generate constraint conditions for anchoring the reasoning path of the fault diagnosis model.

[0078] Optionally, the verification module 30 is further configured to acquire expert judgment criteria in the field of wire fault diagnosis; parse the expert judgment criteria into a formal reasoning process that includes observation conditions, causal inference steps, and compliance boundaries; based on the formal reasoning process, construct structured data slots and instruction slots, wherein the data slots are used to fill in real data, and the instruction slots are used to guide the model's step-by-step reasoning and compliance verification points; bind the variables in the expert judgment criteria to the data slots, and bind the logical relationships in the expert judgment criteria to the instruction slots to obtain mapping rules.

[0079] Optionally, the diagnostic module 40 is further configured to input the constraints into a preset fault diagnosis model, so that the fault diagnosis model determines the physical logical boundary of the reasoning; input the real data and the procedural basis into the fault diagnosis model, so that the fault diagnosis model, within the physical logical boundary, performs semantic association and logical organization reasoning on the real data based on the reasoning path corresponding to the procedural basis, and obtains the diagnostic result.

[0080] Optionally, the extraction module 20 is further configured to retrieve physical parameters associated with the diagnostic object from a preset real-time information database; retrieve procedural guidelines applicable to the diagnostic object from the knowledge base, the procedural guidelines including parameter types and integrity requirements necessary for fault diagnosis; verify the physical parameters based on the integrity requirements and parameter types to obtain verification results; if the verification results indicate that the physical parameters are complete and conform to the acquisition specifications, then the physical parameters are extracted as the actual data of the diagnostic object, and the procedural guidelines are extracted.

[0081] Optionally, the extraction module 20 is further configured to trigger a missing item reminder and stop the extraction of real data and procedural basis if the real-time information database does not contain physical parameters associated with the diagnostic object, or the knowledge base does not contain procedural basis applicable to the diagnostic object, or the verification result indicates that the physical parameters are incomplete.

[0082] Optionally, the question-and-answer constraint module 10 is further configured to, upon receiving a fault diagnosis instruction, perform semantic parsing on the fault diagnosis instruction based on a preset power professional prompt word framework, extract the question intent from the fault diagnosis instruction; determine whether the question intent is complete and unambiguous; if the question intent is incomplete or ambiguous, push equipment information and a power grid topology subgraph related to the fault diagnosis instruction to the user, guiding the user to complete the question intent; respond to the user's completed question intent and confirmation feedback, normalize the question intent into a uniquely determined physical device object and its corresponding spatiotemporal boundary; and, based on the physical device object and the spatiotemporal boundary, clarify the diagnosis object.

[0083] Optionally, the intent of the question includes: device identification, spatiotemporal range, and fault phenomenon elements.

[0084] The hallucination suppression device for transmission line fault diagnosis models provided in this application employs the hallucination suppression method for transmission line fault diagnosis models described in the above embodiments, and can solve the technical problem of hallucination suppression models generating diagnostic conclusions with high confidence but low reliability. Compared with the prior art, the beneficial effects of the hallucination suppression device for transmission line fault diagnosis models provided in this application are the same as those of the hallucination suppression method for transmission line fault diagnosis models provided in the above embodiments, and other technical features in the hallucination suppression device for transmission line fault diagnosis models are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0085] This application provides a hallucination suppression device for a transmission line fault diagnosis model. The hallucination suppression device for a transmission line fault diagnosis model includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the hallucination suppression method for the transmission line fault diagnosis model in the above embodiment 1.

[0086] The following is for reference. Figure 6The diagram illustrates a structural schematic of a hallucination suppression device suitable for implementing the transmission line fault diagnosis model in the embodiments of this application. The hallucination suppression device for the transmission line fault diagnosis model in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 6 The hallucination suppression device shown for a transmission line fault diagnosis model is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0087] like Figure 6 As shown, the hallucination suppression device for a transmission line fault diagnosis model may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the hallucination suppression device for the transmission line fault diagnosis model. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the hallucination suppression device for the transmission line fault diagnosis model to exchange data wirelessly or via wired communication with other devices. Although the figure shows a hallucination suppression device for the transmission line fault diagnosis model with various systems, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems can be implemented alternatively.

[0088] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0089] The hallucination suppression device for transmission line fault diagnosis models provided in this application employs the hallucination suppression method for transmission line fault diagnosis models described in the above embodiments, which can solve the technical problem of hallucination suppression models generating diagnostic conclusions with high confidence but low reliability. Compared with the prior art, the beneficial effects of the hallucination suppression device for transmission line fault diagnosis models provided in this application are the same as those of the hallucination suppression method for transmission line fault diagnosis models provided in the above embodiments, and other technical features in this hallucination suppression device for transmission line fault diagnosis models are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0090] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0091] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0092] The above description is only a part of the embodiments of this application and does not limit the scope of protection of this application. All equivalent structural transformations made under the technical concept of this application and using the content of this application specification and drawings, or direct / indirect applications in other related technical fields, are included in the scope of protection of this application.

Claims

1. A method for suppressing hallucinations in a transmission line fault diagnosis model, characterized in that, The method includes: When a fault diagnosis instruction is received, the intent of the fault diagnosis instruction is analyzed, and when the intent of the fault diagnosis instruction is determined to be ambiguous, the diagnostic object is clarified based on the preset question and answer prompt constraints. The real data and procedural basis corresponding to the diagnostic object are extracted from the preset real-time information database and the preset knowledge base, respectively. Based on the analysis model, physical and logical verification is performed on the real data to obtain the constraints that the model diagnosis needs to comply with. The real data, the procedural basis, and the constraints are input into a preset fault diagnosis model, which then performs causal deduction on the real data based on the constraints and the procedural basis to obtain a diagnosis result.

2. The hallucination suppression method for a transmission line fault diagnosis model as described in claim 1, characterized in that, The step of performing physical and logical verification on the real data based on the judgment model to obtain the constraints that the model diagnosis needs to comply with includes: Based on the analysis model, the real data is organized into a chain of evidence corresponding to power failure scenarios; Based on the mapping rules between the preset expert judgment criteria and the model reasoning logic, the physical consistency and procedural compliance of the evidence chain are verified to obtain the verification results. If the verification result is a verification failure, the structured constraint template corresponding to the mapping rule is invoked to generate constraint conditions for anchoring the inference path of the fault diagnosis model.

3. The hallucination suppression method for a transmission line fault diagnosis model as described in claim 2, characterized in that, Before the step of performing physical consistency and procedural compliance checks on the evidence chain based on the mapping rules between preset expert judgment criteria and model reasoning logic, and obtaining the check results, the method further includes: Obtain expert judgment criteria in the field of wire fault diagnosis; The expert judgment criteria are analyzed as a formal reasoning process that includes observation conditions, causal inference steps, and compliance boundaries. Based on the formal reasoning process, structured data slots and instruction slots are constructed. The data slots are used to fill in real data, and the instruction slots are used to guide the model to reason step by step and to verify compliance. The variables in the expert judgment criteria are bound to the data slots, and the logical relationships in the expert judgment criteria are bound to the instruction slots to obtain mapping rules.

4. The hallucination suppression method for a transmission line fault diagnosis model as described in claim 1, characterized in that, The step of inputting the real data, the procedural basis, and the constraints into a preset fault diagnosis model, so that the fault diagnosis model performs causal inference on the real data based on the constraints and the procedural basis to obtain a diagnosis result, includes: The constraints are input into a preset fault diagnosis model, which then determines the physical and logical boundaries of the reasoning. The real data and the procedural basis are input into the fault diagnosis model, so that the fault diagnosis model, within the physical and logical boundaries, performs semantic association and logical organization reasoning on the real data based on the reasoning path corresponding to the procedural basis, and obtains the diagnosis result.

5. The hallucination suppression method for a transmission line fault diagnosis model as described in claim 1, characterized in that, The steps of extracting real data and procedural basis corresponding to the diagnostic object from a preset real-time information database and a preset knowledge base, respectively, include: Retrieve physical parameters associated with the diagnostic object from a preset real-time information database; Retrieve the applicable procedural guidelines from the knowledge base, the procedural guidelines including the types of parameters necessary for fault diagnosis and the integrity requirements; Based on the integrity requirements and the parameter type, the physical parameters are verified to obtain the verification results; If the verification result indicates that the physical parameters are complete and conform to the acquisition specifications, then the physical parameters are extracted as the actual data of the diagnostic object, and the procedure basis is extracted.

6. The hallucination suppression method for a transmission line fault diagnosis model as described in claim 5, characterized in that, The step of extracting real data and procedural basis corresponding to the diagnostic object from a preset real-time information database and a preset knowledge base, respectively, further includes: If the real-time information database does not contain physical parameters associated with the diagnostic object, or the knowledge base does not contain procedural guidelines applicable to the diagnostic object, or the verification result indicates that the physical parameters are incomplete, a missing item alert will be triggered, and the extraction of real data and procedural guidelines will be suspended.

7. The hallucination suppression method for a transmission line fault diagnosis model as described in claim 1, characterized in that, The step of receiving a fault diagnosis instruction, performing intent analysis on the instruction, and clarifying the diagnostic object based on preset question-and-answer prompts when the intent of the instruction is determined to be ambiguous, further includes: Upon receiving a fault diagnosis instruction, the instruction is semantically parsed based on a preset power industry prompt word framework to extract the question intent from the instruction. Determine whether the intent of the question is complete and unambiguous; If the question intent is incomplete or ambiguous, then push equipment information and power grid topology sub-graph related to the fault diagnosis command to the user to guide the user to complete the question intent; In response to the user's completed question intent and confirmation feedback, the question intent is normalized into a uniquely determined physical device object and its corresponding spatiotemporal boundary; Based on the physical device object and the spatiotemporal boundary, the diagnostic object is identified.

8. The hallucination suppression method for a transmission line fault diagnosis model as described in claim 7, characterized in that, The intent of the question includes: device identification, spatiotemporal range, and fault phenomenon elements.

9. A hallucination suppression device for a transmission line fault diagnosis model, characterized in that, The device includes: The question-and-answer constraint module is used to analyze the intent of a fault diagnosis instruction when it is received, and to clarify the diagnosis object based on preset question-and-answer prompt constraints when the intent of the fault diagnosis instruction is determined to be ambiguous. The extraction module is used to extract real data and procedural basis corresponding to the diagnostic object from a preset real-time information database and a preset knowledge base, respectively. The verification module is used to perform physical and logical verification on the real data based on the judgment model to obtain the constraints that the model diagnosis needs to comply with. The diagnostic module is used to input the real data, the procedural basis, and the constraints into a preset fault diagnosis model, so that the fault diagnosis model can perform causal inference on the real data based on the constraints and the procedural basis to obtain a diagnostic result.

10. A hallucination suppression device for a transmission line fault diagnosis model, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the hallucination suppression method for a transmission line fault diagnosis model as described in any one of claims 1 to 8.

Images