Power equipment fault data tracing and analyzing system

Abstract

The invention discloses a power equipment fault data tracing and analyzing system, belongs to the technical field of power equipment fault diagnosis, and aims to solve the problems that the fusion efficiency is low and a diagnosis model is statically solidified when multi-source heterogeneous data is processed in the conventional power equipment fault diagnosis, and an effective reasoning and interaction mechanism is lacked when an uncertain fault phenomenon occurs. The system comprises a data interface and acquisition unit which is used for acquiring original data; the data standardization and material evidence quantification module is used for receiving the original data, processing the original data into digital material evidences with uniform structures and calculating the information amount of the digital material evidences; the causal knowledge base module is used for storing a causal suspicion network; the fault evidence chain reasoning engine searches a fault evolution path and outputs the path as a candidate fault evidence chain; the adaptive iteration control module is used for evaluating the certainty of a reasoning result and executing iteration control on the whole tracing and analysis process; and the diagnosis report and interaction module is used for generating and presenting a structured diagnosis report. The method is used for power equipment.

Description

Technical Field

[0001] This invention relates to a power equipment fault data tracing and analysis system, belonging to the field of power equipment fault diagnosis technology. Background Technology

[0002] Power equipment is the foundation for the safe and stable operation of a power system, making timely and accurate fault diagnosis crucial. With the increasing intelligence of power grids, the available equipment status data is becoming increasingly abundant, including massive amounts of real-time monitoring signals, discrete protection action records, and unstructured maintenance work orders, providing a data foundation for precise fault analysis.

[0003] However, in current fault diagnosis practices, effectively integrating data from diverse sources, types, and formats remains a technical challenge. Existing diagnostic methods often struggle to comprehensively analyze qualitative event information and quantitative time-series data within a unified framework. The diagnostic process frequently relies on the experience of domain experts, lacking standardized objective evidence metrics, which to some extent affects the accuracy and reproducibility of diagnostic results.

[0004] Furthermore, when dealing with complex faults, on-site monitoring information is often uncertain and incomplete, pointing to multiple potential causes simultaneously, leading to the diagnostic system outputting several similar conclusions. In such cases, existing technologies typically lack effective mechanisms to guide subsequent troubleshooting, forcing maintenance personnel to conduct extensive, somewhat indiscriminate, manual inspections and tests. This process is cumbersome, inefficient, and delays the optimal time for fault resolution.

[0005] More importantly, most current diagnostic systems rely on static fault knowledge bases or models. These knowledge bases are pre-defined at the system's initial construction and cannot self-improve during subsequent operation. When unexpected new fault modes occur, or when the fault propagation path contradicts existing knowledge, the fixed knowledge base cannot provide a reasonable explanation for the observed anomalies, leading to diagnostic failure. Furthermore, the system lacks the ability to learn and revise its knowledge system from the analysis of complex faults, limiting its long-term robustness and adaptability. Summary of the Invention

[0006] The purpose of this invention is to address the problems of low fusion efficiency, static and fixed diagnostic models, and lack of effective reasoning and interaction mechanisms when dealing with uncertain fault phenomena in existing power equipment fault diagnosis methods, which limit the degree of automation and accuracy of diagnosis. This invention provides a power equipment fault data tracing and analysis system.

[0007] The present invention discloses a power equipment fault data tracing and analysis system, which includes:

[0008] The data interface and acquisition unit are used to connect to multiple different types of external data sources to obtain raw data related to power equipment faults.

[0009] The data standardization and physical evidence quantification module is connected to the data interface and the acquisition unit. It is used to receive the raw data, process it into digital physical evidence with a unified structure, and calculate an amount of information for each digital physical evidence to characterize the degree to which the device state deviates from the normal operating state.

[0010] The causal knowledge base module is used to store a causal suspect network that represents the causal relationships of power equipment failures in a directed graph formal structure.

[0011] The fault evidence chain reasoning engine has its input terminals connected to the data standardization and physical evidence quantification module and the causal knowledge base module, respectively. It is used to search for fault evolution paths that can explain all current observation evidence with the minimum total explanation cost based on the received digital physical evidence set and the loaded causal suspect network, and output the path as a candidate fault evidence chain.

[0012] An adaptive iterative control module, connected to the fault evidence chain inference engine, is used to receive candidate fault evidence chains, evaluate the certainty of the inference results, and perform iterative control on the entire tracing and analysis process based on the evaluation results.

[0013] The diagnostic report and interaction module is connected to the fault evidence chain inference engine and the adaptive iteration control module, respectively. It is used to generate and present a structured diagnostic report, receive and present instructions from the adaptive iteration control module, and send user feedback information back to the adaptive iteration control module.

[0014] Preferably, the data standardization and physical evidence quantification module processes the original data in the following ways:

[0015] For continuous time-series signal data, noise suppression, time alignment, and standard fraction normalization are performed.

[0016] For discrete event data, events are converted into unique integer identifiers according to a preset event dictionary and aligned with the time base of the time-series signal data;

[0017] The processed data is encapsulated into digital evidence containing a unique identifier, timestamp, data source, and standardized numerical value.

[0018] Preferably, the process of determining the amount of information includes:

[0019] Based on the type of digital evidence, select a pre-built probability distribution model that describes the normal operating status of the device;

[0020] This model is used to calculate the probability of the standardized value of current digital evidence appearing.

[0021] Finally, information theory function operations are performed on the probability value to obtain the final information content.

[0022] Preferably, the causal suspect network is a directed graph structure;

[0023] The nodes represent the physical state or fault events of power equipment, and the directed edges between nodes represent the direct physical causal relationships between the nodes.

[0024] Each directed edge is associated with a conditional probability value that characterizes the strength of the causal relationship.

[0025] Preferably, the total explanation cost Path cost Cost of data fitting Addition constitutes:

[0026] ;

[0027] in, This indicates that the observed digital evidence set Under the premise of, candidate failure path Total explanation cost;

[0028] This represents a specific candidate failure path;

[0029] Indicate candidate failure paths The path cost is used to quantify the prior complexity of the occurrence of the candidate failure path itself;

[0030] Indicate candidate failure paths The data fitting cost is used to quantify the cost of observational evidence that the candidate failure path fails to explain;

[0031] The path cost is used to quantify the prior complexity of the candidate failure evidence chain itself, and is calculated as the sum of the negative log probabilities of all causal transitions on the chain:

[0032] ;

[0033] in, Indicate candidate failure paths The first node in the candidate path, i.e., the starting node or root cause node;

[0034] Represents the root cause node The prior probability of occurrence is obtained based on historical failure statistics or expert evaluation;

[0035] Represents logarithmic operations to the base 2;

[0036] Indicate candidate failure paths The total number of nodes included.

[0037] Indicate candidate failure paths The first in 1 node

[0038] Indicate candidate failure paths The first in The nth node, i.e., the nth node Each node is the direct predecessor node on the fault path;

[0039] Indicates in Under the condition that the node's state occurs, The conditional probability of a node's state occurring is stored in the causal suspected network, from... point to The weight value associated with that directed edge;

[0040] The data fitting cost is used to quantify the cost of observational evidence that the candidate fault evidence chain fails to explain, and is calculated as the sum of the information content of all digital evidence that is not explained by the chain.

[0041] Preferably, the fault evidence chain reasoning engine employs a heuristic best-first search algorithm to find the candidate fault evidence chain with the minimum total explanation cost in the causal suspect network.

[0042] The algorithm determines the priority of node expansion through an evaluation function whose value is the sum of the actual cumulative path cost from the starting node to the current node and the calculated cost of fitting the current data.

[0043] Preferably, the iterative control function of the adaptive iterative control module includes active evidence inquiry; this function is executed when it is determined that there are two or more candidate fault evidence chains, the total interpretation cost of which is lower than a preset acceptable threshold, and the cost difference between them is less than a preset ambiguity distinction threshold.

[0044] Preferably, the execution process of the proactive evidence inquiry function includes:

[0045] First, calculate the information entropy of the current candidate fault evidence chain set to quantify the uncertainty of the reasoning result;

[0046] Then, for each piece of evidence to be queried in a list of evidence to be queried, calculate the expected information gain that can be obtained after obtaining the evidence;

[0047] Finally, the evidence with the greatest information gain is selected as the key evidence to be queried, and a query instruction is generated.

[0048] Preferably, the iterative control function of the adaptive iterative control module further includes causal network counterfactual correction;

[0049] This function is executed when the total explanatory cost of all candidate fault evidence chains is higher than a preset contradiction determination threshold.

[0050] Preferably, the execution process of the causal network counterfactual correction function includes:

[0051] First, locate one or more stubborn digital pieces of evidence that cause excessively high data fitting costs;

[0052] Secondly, a new directed edge is generated as a candidate causal hypothesis. The source node of this new directed edge exists in the current lowest-cost candidate fault evidence chain, and the observable phenomenon associated with its target node can explain the stubborn digital evidence.

[0053] Finally, the candidate causal hypothesis is sent to the diagnostic report and interaction module for user evaluation, and an update operation is performed on the causal knowledge base module after receiving confirmation from the user.

[0054] This invention provides a power equipment fault data tracing and analysis system. It has the following beneficial effects:

[0055] 1. This invention, by setting up data standardization and physical evidence quantification modules, processes multi-source heterogeneous raw data into structured digital physical evidence carrying information. Combined with a fault evidence chain reasoning engine, it searches in the causal suspect network based on the principle of minimizing the total interpretation cost, thereby achieving accurate tracing of the root cause of the fault. By combining qualitative expert experience with quantitative operational data, the diagnostic process is based on a unified evidence measurement standard, avoiding the problems of difficult data fusion and strong subjectivity in the reasoning process in traditional methods, thus improving the accuracy and objectivity of fault diagnosis.

[0056] 2. This invention, by setting an adaptive iterative control module, executes an active evidence retrieval function when there is ambiguity in the diagnostic results. This function calculates and selects the evidence to be queried with the maximum information gain and initiates a purposeful interactive query to the user. This effectively solves the problem of difficulty in distinguishing multiple potential fault paths in complex fault scenarios, guides the system to actively obtain the most critical identifiable information, avoids the blindness of manual investigation, and improves the convergence speed and efficiency of the diagnostic process.

[0057] 3. This invention, through the causal network counterfactual correction function of the adaptive iterative control module, can locate key contradictory evidence when all candidate fault evidence chains cannot reasonably explain the observation evidence, and generate new causal relationship hypotheses for user confirmation, thereby updating the causal knowledge base module. This enables the system's fault knowledge base to have the ability to learn and evolve, allowing the system to discover new fault mechanisms from difficult faults that do not conform to existing knowledge. This overcomes the limitations of traditional diagnostic systems' knowledge models being fixed and unable to adapt to new fault modes, and enhances the system's long-term applicability and robustness. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of the functional module structure of the power equipment fault data tracing and analysis system described in this invention;

[0059] Figure 2 This is a flowchart illustrating the overall workflow for tracing and analyzing power equipment fault data.

[0060] Figure 3 This is a flowchart of the optimal path search algorithm. Detailed Implementation

[0061] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0062] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0063] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but this is not intended to limit the scope of the invention.

[0064] Example 1:

[0065] The following is combined Figures 1-3 This embodiment describes a power equipment fault data tracing and analysis system, which includes:

[0066] The data interface and acquisition unit are used to connect to multiple different types of external data sources to obtain raw data related to power equipment faults.

[0067] The data standardization and physical evidence quantification module is connected to the data interface and the acquisition unit. It is used to receive the raw data, process it into digital physical evidence with a unified structure, and calculate an amount of information for each digital physical evidence to characterize the degree to which the device state deviates from the normal operating state.

[0068] The causal knowledge base module is used to store a causal suspect network that represents the causal relationships of power equipment failures in a directed graph formal structure.

[0069] The fault evidence chain reasoning engine has its input terminals connected to the data standardization and physical evidence quantification module and the causal knowledge base module, respectively. It is used to search for fault evolution paths that can explain all current observation evidence with the minimum total explanation cost based on the received digital physical evidence set and the loaded causal suspect network, and output the path as a candidate fault evidence chain.

[0070] An adaptive iterative control module, connected to the fault evidence chain inference engine, is used to receive candidate fault evidence chains, evaluate the certainty of the inference results, and perform iterative control on the entire tracing and analysis process based on the evaluation results.

[0071] The diagnostic report and interaction module is connected to the fault evidence chain inference engine and the adaptive iteration control module, respectively. It is used to generate and present a structured diagnostic report, receive and present instructions from the adaptive iteration control module, and send user feedback information back to the adaptive iteration control module.

[0072] Furthermore, the data standardization and evidence quantification module processes the raw data in the following ways:

[0073] For continuous time-series signal data, noise suppression, time alignment, and standard fraction normalization are performed.

[0074] For discrete event data, events are converted into unique integer identifiers according to a preset event dictionary and aligned with the time base of the time-series signal data;

[0075] The processed data is encapsulated into digital evidence containing a unique identifier, timestamp, data source, and standardized numerical value.

[0076] Furthermore, the process of determining the amount of information includes:

[0077] Based on the type of digital evidence, select a pre-built probability distribution model that describes the normal operating status of the device;

[0078] This model is used to calculate the probability of the standardized value of current digital evidence appearing.

[0079] Finally, information theory function operations are performed on the probability value to obtain the final information content.

[0080] Furthermore, the causal suspect network is a directed graph structure;

[0081] The nodes represent the physical state or fault events of power equipment, and the directed edges between nodes represent the direct physical causal relationships between the nodes.

[0082] Each directed edge is associated with a conditional probability value that characterizes the strength of the causal relationship.

[0083] Furthermore, the total explanation cost Path cost Cost of data fitting Addition constitutes:

[0084] ;

[0085] in, This indicates that the observed digital evidence set Under the premise of, candidate failure path Total explanation cost;

[0086] This represents a specific candidate failure path;

[0087] Indicate candidate failure paths The path cost is used to quantify the prior complexity of the occurrence of the candidate failure path itself;

[0088] Indicate candidate failure paths The data fitting cost is used to quantify the cost of observational evidence that the candidate failure path fails to explain;

[0089] The path cost is used to quantify the prior complexity of the candidate failure evidence chain itself, and is calculated as the sum of the negative log probabilities of all causal transitions on the chain:

[0090] ;

[0091] in, Indicate candidate failure paths The first node in the candidate path, i.e., the starting node or root cause node;

[0092] Represents the root cause node The prior probability of occurrence is obtained based on historical failure statistics or expert evaluation;

[0093] Represents logarithmic operations to the base 2;

[0094] Indicate candidate failure paths The total number of nodes included.

[0095] Indicate candidate failure paths The first in 1 node

[0096] Indicate candidate failure paths The first in The nth node, i.e., the nth node Each node is the direct predecessor node on the fault path;

[0097] Indicates in Under the condition that the node's state occurs, The conditional probability of a node's state occurring is stored in the causal suspected network, from... point to The weight value associated with that directed edge;

[0098] The data fitting cost is used to quantify the cost of observational evidence that the candidate fault evidence chain fails to explain, and is calculated as the sum of the information content of all digital evidence that is not explained by the chain.

[0099] Furthermore, the fault evidence chain reasoning engine employs a heuristic best-first search algorithm to find the candidate fault evidence chain with the minimum total explanation cost in the causal suspect network.

[0100] The algorithm determines the priority of node expansion through an evaluation function whose value is the sum of the actual cumulative path cost from the starting node to the current node and the calculated cost of fitting the current data.

[0101] Furthermore, the iterative control function of the adaptive iterative control module includes proactive evidence inquiry; this function is executed when it is determined that there are two or more candidate fault evidence chains, the total interpretation cost of which is lower than a preset acceptable threshold, and the cost difference between them is less than a preset ambiguity distinction threshold.

[0102] Furthermore, the execution process of the proactive evidence inquiry function includes:

[0103] First, calculate the information entropy of the current candidate fault evidence chain set to quantify the uncertainty of the reasoning result;

[0104] Then, for each piece of evidence to be queried in a list of evidence to be queried, calculate the expected information gain that can be obtained after obtaining the evidence;

[0105] Finally, the evidence with the greatest information gain is selected as the key evidence to be queried, and a query instruction is generated.

[0106] Furthermore, the iterative control function of the adaptive iterative control module also includes causal network counterfactual correction;

[0107] This function is executed when the total explanatory cost of all candidate fault evidence chains is higher than a preset contradiction determination threshold.

[0108] Furthermore, the execution process of the causal network counterfactual correction function includes:

[0109] First, locate one or more stubborn digital pieces of evidence that cause excessively high data fitting costs;

[0110] Secondly, a new directed edge is generated as a candidate causal hypothesis. The source node of this new directed edge exists in the current lowest-cost candidate fault evidence chain, and the observable phenomenon associated with its target node can explain the stubborn digital evidence.

[0111] Finally, the candidate causal hypothesis is sent to the diagnostic report and interaction module for user evaluation, and an update operation is performed on the causal knowledge base module after receiving confirmation from the user.

[0112] In this invention, reference is made to the appendix. Figure 1 and attached Figure 2 This invention provides a power equipment fault data tracing and analysis system, which is deployed in a server or industrial control computer hardware environment with a processor, memory, and network communication interface. Functionally, the system includes: a data interface and acquisition unit 10, a data standardization and physical evidence quantification module 20, a causal knowledge base module 30, a fault evidence chain inference engine 40, an adaptive iterative control module 50, and a diagnostic report and interaction module 60.

[0113] The data interface and acquisition unit 10 is used to communicate and acquire data from multiple heterogeneous external data sources. Specifically, the data interface and acquisition unit 10 includes industry standard protocol interfaces such as OPC and Modbus for real-time time-series data, an SQL interface for historical databases, and a file parsing interface for parsing unstructured or semi-structured text (such as maintenance work orders and test reports). The data interface and acquisition unit 10 acquires time-series signal data, condition monitoring data, discrete event data, and static attribute data during the operation of power equipment from these data sources, and transmits the acquired raw data stream to the data standardization and physical evidence quantification module 20.

[0114] The data standardization and evidence quantification module 20 is connected to the data interface and acquisition unit 10 to receive raw data and process it into a unified, structured data unit suitable for subsequent causal inference. The data standardization and evidence quantification module 20 performs two core operations: the first is data standardization, which involves cleaning, aligning, and normalizing input data of different dimensions and formats, and encapsulating it into a preset digital evidence data structure containing fields such as unique identifiers, timestamps, data sources, and standardized values; the second is information quantification, where the module internally stores probability distribution models for various monitoring data under normal equipment operation. For each received standardized data value, the module calculates its information content based on the corresponding probability distribution model. The information content characterizes the degree to which the data value deviates from the normal state. After processing, the structured digital evidence set carrying the information content is output to the fault evidence chain inference engine 40.

[0115] The causal knowledge base module 30 is a storage unit for storing and managing causal knowledge of equipment failures. It is implemented as a graph database or a relational database. The causal knowledge base module 30 stores a causal suspect network. This causal suspect network is a directed graph structure where nodes represent the physical state or failure events of electrical equipment, directed edges represent direct physical causal relationships between nodes, and conditional probability values ​​characterizing the strength of these causal relationships are associated with the edges. The causal knowledge base module 30 provides an interface to the failure evidence chain inference engine 40 to read the entire or part of the causal suspect network and has a writable update interface for receiving verified new causal relationships from the adaptive iterative control module 50, thereby updating the network.

[0116] The fault evidence chain inference engine 40 is the core processing unit of the system, with its inputs connected to the data standardization and evidence quantification module 20 and the causal knowledge base module 30, respectively. When a fault event occurs, the fault evidence chain inference engine 40 obtains a set of digital evidence related to the fault event from the data standardization and evidence quantification module 20 and loads a causal suspect network from the causal knowledge base module 30. The core function of the fault evidence chain inference engine 40 is to search and identify one or more fault evolution paths, i.e., fault evidence chains, within the causal suspect network, based on a principle of minimizing total explanation cost. This total explanation cost consists of two parts: the prior probability cost of the path itself and the data fitting cost caused by the evidence that the path fails to explain. After the inference calculation is completed, the fault evidence chain inference engine 40 outputs one or more candidate fault evidence chains sorted by total explanation cost to the adaptive iterative control module 50 and the diagnostic report and interaction module 60.

[0117] The adaptive iterative control module 50 is connected to the fault evidence chain inference engine 40 and is used to handle uncertainties in the inference process. The adaptive iterative control module 50 receives the candidate fault evidence chain set output by the fault evidence chain inference engine 40.

[0118] When the adaptive iterative control module 50 determines that the current inference result is ambiguous (i.e., the interpretation costs of multiple candidate chains are similar), it initiates an active evidence query function. By comparing the differences between these candidate chains, it calculates and determines the key evidence to be queried that can eliminate ambiguity to the greatest extent, and generates a query command to send to the diagnostic report and interaction module 60. After receiving user input (i.e., new evidence) fed back by the diagnostic report and interaction module 60, the adaptive iterative control module 50 adds the new evidence to the current evidence set and instructs the faulty evidence chain inference engine 40 to re-execute the inference based on the updated evidence set.

[0119] When the adaptive iterative control module 50 determines that the current inference result is contradictory (i.e., the explanation cost of all candidate chains is too high), it activates the causal network counterfactual correction function. By analyzing physical evidence that cannot be reasonably explained, it generates one or more new, unverified causal relationship hypotheses and sends these unverified hypotheses to the diagnostic report and interaction module 60 for user evaluation. Only after receiving a user confirmation signal from the diagnostic report and interaction module 60 does the adaptive iterative control module 50 send an update command to the causal knowledge base module 30 to permanently correct the knowledge base.

[0120] The diagnostic report and interaction module 60 serves as the interface between the system and the user. It receives the final inference result from the fault evidence chain inference engine 40 and formats it into a human-readable structured diagnostic report containing the fault evolution path, a list of key evidence, and a confidence assessment. Simultaneously, the module receives active query commands or causal hypotheses to be verified from the adaptive iterative control module 50 and presents them to the user. The user can respond to query commands (e.g., inputting the result of an offline test) or provide feedback on new causal hypotheses through the interface of the diagnostic report and interaction module 60. The user-inputted information is fed back to the adaptive iterative control module 50 via the diagnostic report and interaction module 60 to drive a new round of inference until the inference results converge.

[0121] Reference Appendix Figure 1 and attached Figure 2 The input end and data interface of the data standardization and physical evidence quantification module 20 are connected to the acquisition unit 10 to receive multimodal raw device data; its output end is connected to the fault evidence chain inference engine 40 to provide a set of digital physical evidence that has been standardized and carries information weights.

[0122] The data standardization and evidence quantification module 20's internal processing flow is divided into a heterogeneous data preprocessing stage and an information quantification stage. In the heterogeneous data preprocessing stage, the data standardization and evidence quantification module 20 performs corresponding standardization operations on different types of data received from the data interface and acquisition unit 10. For continuous time-series signal data, such as voltage, current, and oil temperature, the data standardization and evidence quantification module 20 first uses a digital filter (e.g., a median filter or a low-pass filter) to suppress noise. Then, it resamples data with different sampling rates to a unified time base using methods such as linear interpolation or spline interpolation, achieving temporal alignment. Subsequently, the aligned data is standardized using the Z-score method, converting it into dimensionless values ​​with a mean of 0 and a standard deviation of 1 to eliminate the influence of different physical dimensions.

[0123] For discrete event data, such as the action records of protection devices or operation items in maintenance work orders, the data standardization and evidence quantification module 20 first converts the text descriptions or encoded events into unique integer identifiers based on a preset event dictionary. The event dictionary establishes a mapping relationship between all discrete states or operations of the device and integer identifiers. Subsequently, the timestamp of the event is extracted and aligned with the time base of the time-series signal data.

[0124] For static attribute data extracted from equipment ledgers or test reports, the data standardization and evidence quantification module 20 treats it as a constant feature value associated with a specific equipment identifier and binds it to the dynamic data of the equipment at the data structure level. All data that has undergone the above preprocessing is uniformly encapsulated into a standard data structure unit, namely digital evidence. The data structure unit contains fields such as a unique identifier, timestamp, data source, and standardized numerical values.

[0125] During the information quantification phase, the data standardization and physical evidence quantification module 20 calculates the information content for each standardized digital physical evidence. The information calculation process relies on pre-built probability distribution models of normal operation status for various types of monitoring data. These models are built offline in the data standardization and physical evidence quantification module 20 during the initial stage of system deployment, using a large amount of historical data confirmed to be from the period of normal equipment operation.

[0126] Specifically, for continuous variables (such as standardized temperature and vibration amplitude), their probability distribution model is a probability density function. The probability density function is constructed using kernel density estimation, employing a Gaussian kernel function. Through learning from a large number of normal sample points, the continuous probability distribution of the continuous variable under normal conditions is fitted. For discrete variables (such as integer identifiers of events), their probability distribution model is a probability mass function, constructed by statistically analyzing the frequency of each discrete event in a large number of normal operating samples.

[0127] When a new, standardized digital evidence enters this stage, the data standardization and evidence quantification module 20 selects a pre-built probability distribution model based on its data source or type. If the digital evidence's value is continuous, its probability density function is calculated at that value; if it is discrete, its corresponding occurrence frequency is directly looked up. This calculated probability value represents the likelihood of the digital evidence's value occurring under normal equipment operation. Finally, the data standardization and evidence quantification module 20 obtains the final information content of the digital evidence by taking the negative logarithm to base 2 of this probability value. The final information content is filled into the corresponding fields of the digital evidence data structure. After quantification, the digital evidence set containing complete information is transmitted to the fault evidence chain inference engine 40.

[0128] Reference Appendix Figure 1 and attached Figure 2 The reasoning process of the fault evidence chain reasoning engine 40 is carried out on a pre-built, structured causal knowledge model. The causal knowledge model is the causal suspect network stored in the causal knowledge base module 30.

[0129] The construction process of the causal suspicion network involves transforming the physical structure, operating mechanism, and failure mode knowledge of power equipment into a formalized directed graph structure. Nodes in the network are discretized representations of critical physical states, component health conditions, or external influencing factors of the equipment. Specifically, these nodes are defined through analysis of equipment design manuals, failure mode and effects analysis reports, and the knowledge of domain experts. For example, inter-turn short circuits in windings, cooler fan failures, or excessive moisture content in insulating oil can all be defined as independent nodes.

[0130] In a network, directed edges represent a direct causal relationship between the states of two nodes, conforming to engineering principles. The direction of an edge points from the cause node to the result node. For example, a directed edge can be created from a node with excessive moisture content in insulating oil to a node showing a decrease in the breakdown voltage of the insulating oil, indicating that the former is the direct cause of the latter. These causal relationships are determined based on explicit physical laws, publicly published research findings on equipment failure mechanisms, or generally accepted expert consensus within the industry.

[0131] Each directed edge is assigned a quantified weight, which is a conditional probability value representing the probability of the result node event occurring after the causal node event has occurred. This conditional probability value is determined based on statistical analysis of historical fault data, calculation results from equipment simulation models, or the experience-based knowledge base of domain experts. In this way, the entire causal suspect network not only describes the fault propagation path but also quantifies the probability of each path occurring.

[0132] The constructed causal suspect network is persistently stored in the causal knowledge base module 30 in the form of a graph data structure. One specific storage implementation uses a graph database, where each network node is stored as an independent node object, containing a unique identifier and a state description text. Each directed edge is stored as a relationship connecting two node objects, and this relationship itself contains an attribute used to store the conditional probability value. This storage method supports efficient graph traversal and adjacency query operations. The causal knowledge base module 30 provides a data query interface to the fault evidence chain inference engine 40, enabling the fault evidence chain inference engine 40 to efficiently traverse the network's topology and obtain attribute information for any node or edge when performing inference tasks.

[0133] Reference Appendix Figure 1 and attached Figure 2 After receiving the digital evidence set from the data standardization and evidence quantification module 20, the core task of the fault evidence chain reasoning engine 40 is to evaluate and rank all faulty paths in the causal suspect network loaded from the causal knowledge base module 30. This evaluation process is based on calculating the total explanatory cost for each candidate faulty path. The fault evidence chain reasoning engine 40 ultimately selects the path with the lowest total explanatory cost as the optimal fault evidence chain.

[0134] Specifically, for any candidate fault path in the causal suspected network And what is currently observed is from A collection of digital physical evidence Its total explanatory cost Path cost Cost of data fitting The two parts are added together to form:

[0135] ;

[0136] in, Representative of the observed digital evidence set Under the premise of, candidate failure path Total explanation cost; This represents a specific candidate failure path, defined as a path in the causal suspect network consisting of... A sequence formed by connecting nodes in sequence, i.e. ; Represents the set of all currently observed digital evidence, i.e. ; Representative candidate failure paths The path cost is used to quantify the prior complexity of the occurrence of the candidate failure path itself; Representative candidate failure paths The data fitting cost is used to quantify the cost of observational evidence that candidate failure paths fail to explain.

[0137] Part 1: Path Costs Used to quantify candidate failure paths The prior complexity or impossibility of the event itself is calculated as the sum of the negative log probabilities of all causal transitions along the path:

[0138]

[0139] in, Is it selecting the fault path? The first node is the starting node or root cause node of the fault path; Root cause node The prior probability of occurrence is obtained based on historical failure statistics or expert evaluation and is pre-stored in the causal knowledge base module 30 as... Node attributes; Is it selecting the fault path? The total number of nodes included; Is it selecting the fault path? The first in One node; Is it selecting the fault path? The Middle Each node, i.e. The node's direct predecessor node on the fault selection path; Is Under the condition that the node's state occurs, The conditional probability of a node's state occurring is stored in the causal suspected network, from... point to The weight value associated with that directed edge; It is a logarithmic operation with base 2; This indicates that a fault path will be selected. From the second node To the last node Each pair of adjacent nodes logarithm of conditional probability Add them all up.

[0140] Part Two: Data Fitting Costs Used to quantify in the assumed candidate failure path If true, the currently observed set of evidence In this context, how much of the evidence is unexpected, and how much of the data fitting cost equals the cost of all unselected candidate failure paths? The sum of information that can be explained from digital evidence:

[0141] ;

[0142] in, Represents the set of all currently observed digital evidence, i.e. , It is a set, representing in In the middle, the candidate fault paths can be A subset of all digital physical evidence inherently explained; It is the difference of two sets, and the result represents All unselected fault paths The stubborn evidence that was explained; It is a specific, uninterpreted digital evidence within the difference set; It is digital evidence The amount of information is calculated by the data standardization and evidence quantification module 20 during the data preprocessing stage, and is used as... One of the attributes; It is a summation function for all uninterpreted digital evidence. Information content Accumulate.

[0143] A digital piece of evidence Does it belong to The basis for judgment is: in the causal knowledge base module 30, each node Each device is pre-associated with a set of observable phenomena, which lists the phenomena that occur when the device is in a certain state. In this state, there is a high probability of monitoring anomalies (i.e., characteristics of digital evidence). If digital evidence... Features and candidate fault paths any node If any phenomenon in the set of associated observable phenomena matches, then Classified The physical meaning of this cost term is that if a path cannot explain multiple highly informative anomalies observed in the field, then the data fitting cost of that path is [amount missing]. It will increase significantly.

[0144] The faulty evidence chain reasoning engine 40 calculates the faulty evidence chain reasoning engine 40 by testing all candidate paths. Finally, the candidate failure path with the lowest cost value is determined as the analysis result.

[0145] Reference Appendix Figure 3 In the faulty evidence chain reasoning engine 40, in order to achieve the total interpretation cost The solution to minimize this problem involves finding the optimal chain of evidence of failure within the causal suspect network. This embodiment employs a heuristic best-first search algorithm.

[0146] The heuristic best-first search algorithm performs a forward search in a causal suspect network. The starting set of nodes for the search is all nodes in the causal suspect network that are defined as root causes (i.e., nodes without incoming edges). The goal of the search is to find a candidate fault path. End point of candidate fault path It can correspond to the observed major fault phenomena with high information content, and candidate fault paths It has the lowest total explanation cost.

[0147] The heuristic best-first search algorithm uses an evaluation function. To determine the priority of node expansion. Evaluation function. For each node expanded during the search process The calculation is performed, defined as the sum of the current actual cumulative path cost and the current data fit:

[0148] ;

[0149] in, It is a node The evaluation function value is used to sort the priority queue to determine which node to expand next; It starts from the starting node (root cause). Along some candidate fault paths Reach the current node The actual cumulative path cost; This assumes some candidate failure paths. Calculate the cost of fitting the current data if the condition is met.

[0150] The item is precisely set as the candidate failure path for that section. Path cost :

[0151] ;

[0152] in, It starts from the starting node (root cause). Along some candidate fault paths Reach the current node The actual cumulative cost; Root cause node The prior probability of occurrence is obtained based on historical failure statistics or expert evaluation and is pre-stored in the causal knowledge base module 30 as... Node attributes; These are some candidate failure paths. The first in One node; These are some candidate failure paths. The first in 1 node (i.e.) (predecessor node); From node arrive The conditional probability is obtained from the corresponding directed edges in the causal suspect network; It is a logarithmic operation with base 2; Indicates partial candidate fault paths In the middle, from the second node To the current node All single-step costs (i.e. , , ..., Add them all together.

[0153] Designed to fit the final data cost An acceptable estimate. Specifically, Set in some candidate failure paths Given the circumstances, all candidate fault paths that have not yet been included in this part. The sum of the information content of the explained digital evidence:

[0154] ;

[0155] in, Represents the set of all currently observed digital evidence, i.e. ; yes A subset containing all candidate failure paths. (i.e., from) arrive The digital evidence matched and explained by the set of observable phenomena associated with the node sequence; It is the difference of two sets, representing the difference between them. Digital evidence at the nodes that remains unexplained; It is a specific, uninterpreted digital evidence within the difference set; It is digital evidence The amount of information is calculated by the data standardization and evidence quantification module 20 during the data preprocessing stage, and is used as... One of the attributes; express function.

[0156] The faulty evidence chain reasoning engine 40 internally maintains a priority queue (open list), and the priority queue is based on... The values ​​are sorted in ascending order for all discovered but not expanded nodes. A heuristic best-first search algorithm starts from all root cause nodes. To begin, in each iteration, retrieve from the priority queue The node with the smallest value To extend this process, we need to examine all its successor nodes. For each successor node The faulty evidence chain reasoning engine 40 calculates its corresponding... and The value is then added to the priority queue.

[0157] When a node is defined as a target fault phenomenon (i.e., strongly correlated with high-information physical evidence) The search process terminates when a value is removed from the priority queue. Because... Admissibility, at this point from arrive The path is the optimal fault evidence chain we are looking for. Optimal chain of evidence for failure With the lowest total explanation cost .

[0158] Reference Appendix Figure 1 and attached Figure 2 The adaptive iterative control module 50 is connected to the faulty evidence chain inference engine 40. The adaptive iterative control module 50 receives the output of the faulty evidence chain inference engine 40, calculated based on the total interpretation cost. The candidate fault evidence chain set is sorted. The adaptive iterative control module 50 first performs the determination of ambiguous states. An ambiguous state is defined as: when there are multiple (two or more) paths in the candidate fault evidence chain set output by the fault evidence chain inference engine 40, their total interpretation cost is... All costs are below a preset acceptable cost threshold, and the cost difference between them is less than a preset ambiguity discrimination threshold. When the system is determined to be in an ambiguous state, the adaptive iterative control module 50 activates the active evidence inquiry function.

[0159] The goal of the proactive evidence eviction function is to eliminate ambiguity most efficiently by requesting additional evidence with the greatest information gain from the user. First, the adaptive iterative control module 50 quantifies the current set of candidate fault paths. Uncertainty. It includes all candidate fault paths that meet the conditions for determining ambiguous states. The adaptive iterative control module 50 first considers each candidate fault path... Total Explanation Cost Convert to its posterior probability This transformation is achieved through a normalized exponential function to ensure that paths with lower total costs are assigned higher probabilities. Subsequently, the current set of candidate failure paths is calculated. Information entropy :

[0160] ;

[0161] in, This represents the set of candidate fault paths that currently meet the conditions for determining an ambiguous state. Information entropy is used to quantify the degree of ambiguity of the current reasoning result; It is the set of candidate fault paths that currently meet the conditions for determining ambiguous states. ; It is the set of candidate fault paths that currently meet the conditions for determining ambiguous states. The total number of candidate fault paths included; Iterate from 1 to The index variable of the summation function; It is the set of candidate fault paths that currently meet the conditions for determining ambiguous states. The first in Candidate fault paths; Represents the set of all currently observed digital evidence, i.e. ; After observing all digital evidence sets In the case of candidate failure paths The posterior probability of its validity is given by It is obtained after normalization by the normalized exponential function; It is a logarithmic operation with base 2; express function.

[0162] Subsequently, the adaptive iterative control module iterates through the list of evidence to be questioned 50 times. The list of evidence to be questioned is based on... Candidate Fault Paths Generated through topological discrepancy analysis, this dataset contains the query evidence corresponding to all key nodes (i.e., bifurcation points) that can distinguish these candidate failure paths. For each piece of evidence to be queried in the list. The adaptive iterative control module 50 calculates its expected information gain. The expected information gain is defined as the difference between the current amount of information and the amount of information acquired. After that (i.e.) Observation results The difference between the expected posterior information (given):

[0163] ;

[0164] in, Represents evidence to be searched The expected information gain; This represents the set of candidate fault paths that currently meet the conditions for determining an ambiguous state. Information entropy is used to quantify the degree of ambiguity of the current reasoning result; Evidence to be searched The set of all observations (e.g., for a sensor inspection) ,That for Normal and Abnormal (Two results); It is a set One of the specific observation results; It is the observation result The prior probability of occurrence is estimated based on historical data or models; When the observation results were observed Next, the set of candidate failure paths The posterior information entropy, the method of calculating the posterior information entropy is the same as... Same, but applicable to Under this new evidence, the new path posterior probability obtained through Bayesian update To calculate; express function.

[0165] The adaptive iterative control module 50 calculates all the evidence to be queried in the list. of Values, and select evidence with the maximum information gain. , This refers to the key evidence identified as the most valuable and crucial for eliminating ambiguity. The adaptive iterative control module 50 will... The message is encapsulated as a query command and sent to the diagnostic report and interaction module 60 for presentation to the user. When the user provides feedback through the diagnostic report and interaction module 60... After the observation results (which constitute a new digital piece of evidence) are obtained, the new evidence is added to the digital evidence collection. In the middle. The adaptive iterative control module 50 then issues an instruction to the faulty evidence chain inference engine 40, causing it to use the updated digital evidence set. Re-execute the optimal path search. Repeat this iterative process until the ambiguity is eliminated (i.e., only one path remains). (significantly lower than all other paths).

[0166] Reference Appendix Figure 1 and attached Figure 2 In addition to handling ambiguous states, the adaptive iterative control module 50 also handles contradictory states in the reasoning results. After receiving the candidate fault evidence chain set, the adaptive iterative control module 50 first performs the judgment of contradictory states. When the fault evidence chain reasoning engine 40 outputs all candidate fault evidence chains, its total interpretation cost... When all values ​​exceed a preset contradiction judgment threshold, the adaptive iterative control module 50 determines that the system has entered a contradiction state.

[0167] The specific source of this high cost is the cost of data fitting. An excessively high value indicates the presence of one or more highly informative items. Stubborn digital evidence These digital evidences cannot be covered by the set of observable phenomena of any existing causal path in the causal knowledge base module 30, that is... .

[0168] After confirming the contradictory state, the adaptive iterative control module 50 activates the causal network counterfactual correction function. The causal network counterfactual correction function first analyzes... The composition is determined by identifying one or more stubborn digital pieces of evidence that contribute the most to the cost. .

[0169] Subsequently, the adaptive iterative control module 50 searches and generates one or more candidate causal hypotheses in the suspected causal network. Each hypothesis is represented by a proposed new directed edge that does not exist in the current network topology. .

[0170] The target node of the new directed edge The selection criteria are: the target node The set of observable phenomena pre-associated in the causal knowledge base module 30 must include evidence that is consistent with persistent physical evidence. The phenomenon of matching characteristics. That is, if If this condition occurs, then the stubborn physical evidence... It can be explained, thus Can be classified gather.

[0171] The source node of the new directed edge The selection criteria are: source node It is the optimal fault evidence chain that exists with the lowest total explanation cost (although still above the threshold). One of the upstream nodes, and Based on physical constraints (such as spatial proximity, energy transfer relationships, or temporal order) and the target node It has the rationale to establish a causal relationship.

[0172] The adaptive iterative control module 50 temporarily adds new directed edges to the local compute replica. (and assign it an initial conditional probability) Then, the fault evidence chain reasoning engine 40 is instructed to recalculate the candidate fault paths containing the new directed edges. Total Explanation Cost The adaptive iterative control module 50 can select to enable... The largest decrease and the drop below the contradiction determination threshold As the optimal candidate causal hypothesis.

[0173] Candidate causal hypotheses are not directly written into the causal knowledge base module 30. Instead, the adaptive iterative control module 50 stores this hypothesis (e.g., a description) in the causal knowledge base module 30. and (A textual description of the new causal relationship between them) is sent to the diagnostic report and interaction module 60.

[0174] The diagnostic report and interaction module 60 presents the candidate causal hypothesis to the user as a suggestion for further offline experiments, simulations, or expert evaluation. When the user confirms the new causal relationship through the diagnostic report and interaction module 60 (e.g., by inputting a confirmation command), the adaptive iterative control module 50 receives the confirmation signal and sends an update command to the causal knowledge base module 30, adding the new directed edge... and its determined conditional probability This process permanently adds the causal suspect network, enabling the network to evolve based on newly discovered knowledge.

[0175] Reference Appendix Figure 1 and attached Figure 2 The diagnostic report and interaction module 60 is the terminal for information exchange between the system and the user. The data input end of the diagnostic report and interaction module 60 is connected to the fault evidence chain inference engine 40 and the adaptive iterative control module 50; its data output end (used to provide user feedback) is connected to the adaptive iterative control module 50. The diagnostic report and interaction module 60 has two main functions: dynamic generation and display of diagnostic reports, and response and feedback of human-computer interaction commands.

[0176] In the dynamic generation and display function of the diagnostic report, the diagnostic report and interaction module 60 receives the optimal fault evidence chain as its final output from the fault evidence chain reasoning engine 40. And the total explanation cost corresponding to this path. The diagnostic report and interaction module 60 parses these calculation results and converts them into structured, human-readable reports.

[0177] The report specifically includes: First, a clearly presented chain of evidence for the best failure, presented in graphical or textual list format. That is, from the root cause node To the final failure phenomenon node The complete node sequence. Second, list the chain of evidence supporting the optimal failure. The key digital evidence, namely in the calculation At that time, it was classified as The set contains a high amount of information Third, provide a confidence assessment of the diagnosis, based on the posterior probability of the optimal path. (This probability value can be calculated by the adaptive iterative control module 50) Intermediate results It is determined by (obtaining).

[0178] The diagnostic report and interaction module 60 presents the report content to the user through a graphical user interface. In the graphical user interface, the user can view the topology of the fault evolution path and select any node on the path. or any key physical evidence To query its detailed attribute information or raw monitoring data.

[0179] In the human-computer interaction command response and feedback function, the diagnostic report and interaction module 60 is responsible for receiving and displaying two types of commands from the adaptive iterative control module 50.

[0180] The first type of instruction is the proactive evidence query instruction. This occurs when the adaptive iterative control module 50 determines the evidence to be queried with the maximum information gain in an ambiguous state. Subsequently, the diagnostic report and interaction module 60 will... (For example, a text instruction to check the status of circuit breaker number XX) is displayed on the interactive interface and provides corresponding input controls (e.g., drop-down menus or text boxes).

[0181] The second type of instruction is the causal network counterfactual correction instruction. This occurs when the adaptive iterative control module 50 generates candidate causal hypotheses under contradictory conditions. Subsequently, the diagnostic report and interaction module 60 will present candidate causal hypotheses (e.g., a system hypothetical state). With state (A causal relationship exists; please verify.) This is displayed on the interactive interface, along with confirmation or denial feedback buttons.

[0182] After the user responds to the command through the interactive interface, the diagnostic report and interaction module 60 is responsible for encapsulating the user's feedback information and sending it back. For the evidence query command, the user inputs the observation results. The diagnostic report and interaction module 60 are sent back to the adaptive iterative control module 50 to trigger a new round of inference. For causal hypothesis instructions, the user's confirmation signal is sent back to the adaptive iterative control module 50 by the diagnostic report and interaction module 60 to trigger a permanent update to the causal knowledge base module 30.

[0183] While the invention has been described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the invention. Therefore, it should be understood that many modifications can be made to the exemplary embodiments, and other arrangements can be designed without departing from the spirit and scope of the invention as defined by the appended claims. It should be understood that different dependent claims and features described herein can be combined in ways different from those described in the original claims. It is also understood that features described in conjunction with individual embodiments can be used in other described embodiments.

Claims

1. A power equipment fault data tracing and analysis system, characterized in that, It includes: The data interface and acquisition unit are used to connect to multiple different types of external data sources to obtain raw data related to power equipment faults. The data standardization and physical evidence quantification module is connected to the data interface and the acquisition unit. It is used to receive the raw data, process it into digital physical evidence with a unified structure, and calculate an amount of information for each digital physical evidence to characterize the degree to which the device state deviates from the normal operating state. The causal knowledge base module is used to store a causal suspect network that represents the causal relationships of power equipment failures in a directed graph formal structure. The fault evidence chain reasoning engine has its input terminals connected to the data standardization and physical evidence quantification module and the causal knowledge base module, respectively. It is used to search for fault evolution paths that can explain all current observation evidence with the minimum total explanation cost based on the received digital physical evidence set and the loaded causal suspect network, and output the path as a candidate fault evidence chain. An adaptive iterative control module, connected to the fault evidence chain inference engine, is used to receive candidate fault evidence chains, evaluate the certainty of the inference results, and perform iterative control on the entire tracing and analysis process based on the evaluation results. The diagnostic report and interaction module is connected to the fault evidence chain inference engine and the adaptive iteration control module, respectively. It is used to generate and present a structured diagnostic report, receive and present instructions from the adaptive iteration control module, and send user feedback information back to the adaptive iteration control module.

2. The power equipment fault data tracing and analysis system according to claim 1, characterized in that, The data standardization and evidence quantification module processes the raw data in the following ways: For continuous time-series signal data, noise suppression, time alignment, and standard fraction normalization are performed. For discrete event data, events are converted into unique integer identifiers according to a preset event dictionary and aligned with the time base of the time-series signal data; The processed data is encapsulated into digital evidence containing a unique identifier, timestamp, data source, and standardized numerical value.

3. The power equipment fault data tracing and analysis system according to claim 1, characterized in that, The process of determining the amount of information includes: Based on the type of digital evidence, select a pre-built probability distribution model that describes the normal operating status of the device; This model is used to calculate the probability of the standardized value of current digital evidence appearing. Finally, information theory function operations are performed on the probability value to obtain the final information content.

4. The power equipment fault data tracing and analysis system according to claim 1, characterized in that, The causal suspicion network is a directed graph structure; The nodes represent the physical state or fault events of power equipment, and the directed edges between nodes represent the direct physical causal relationships between the nodes. Each directed edge is associated with a conditional probability value that characterizes the strength of the causal relationship.

5. The power equipment fault data tracing and analysis system according to claim 1, characterized in that, The total explanation cost Path cost Cost of data fitting Addition constitutes: ； in, This indicates that the observed digital evidence set Under the premise of, candidate failure path Total explanation cost; This represents a specific candidate failure path; Indicate candidate failure paths The path cost is used to quantify the prior complexity of the occurrence of the candidate failure path itself; Indicate candidate failure paths The data fitting cost is used to quantify the cost of observational evidence that the candidate failure path fails to explain; The path cost is used to quantify the prior complexity of the candidate failure evidence chain itself, and is calculated as the sum of the negative log probabilities of all causal transitions on the chain: ； in, Indicate candidate failure paths The first node in the candidate path, i.e., the starting node or root cause node; Represents the root cause node The prior probability of occurrence is obtained based on historical failure statistics or expert evaluation; Represents logarithmic operations to the base 2; Indicate candidate failure paths The total number of nodes included. Indicate candidate failure paths The first in 1 node Indicate candidate failure paths The first in The nth node, i.e., the nth node Each node is the direct predecessor node on the fault path; Indicates in Under the condition that the node's state occurs, The conditional probability of a node's state occurring is stored in the causal suspected network, from... point to The weight value associated with that directed edge; The data fitting cost is used to quantify the cost of observational evidence that the candidate fault evidence chain fails to explain, and is calculated as the sum of the information content of all digital evidence that is not explained by the chain.

6. The power equipment fault data tracing and analysis system according to claim 1, characterized in that, The fault evidence chain inference engine employs a heuristic best-first search algorithm to find the candidate fault evidence chain with the minimum total explanation cost in the causal suspect network. The algorithm determines the priority of node expansion through an evaluation function whose value is the sum of the actual cumulative path cost from the starting node to the current node and the calculated cost of fitting the current data.

7. The power equipment fault data tracing and analysis system according to claim 1, characterized in that, The iterative control function of the adaptive iterative control module includes proactive evidence inquiry; this function is executed when it is determined that there are two or more candidate fault evidence chains, the total interpretation cost of which is lower than a preset acceptable threshold, and the cost difference between them is less than a preset ambiguity distinction threshold.

8. The power equipment fault data tracing and analysis system according to claim 7, characterized in that, The execution process of the proactive evidence inquiry function includes: First, calculate the information entropy of the current candidate fault evidence chain set to quantify the uncertainty of the reasoning result; Then, for each piece of evidence to be queried in a list of evidence to be queried, calculate the expected information gain that can be obtained after obtaining the evidence; Finally, the evidence with the greatest information gain is selected as the key evidence to be queried, and a query instruction is generated.

9. The power equipment fault data tracing and analysis system according to claim 1, characterized in that, The iterative control function of the adaptive iterative control module also includes causal network counterfactual correction; This function is executed when the total explanatory cost of all candidate fault evidence chains is higher than a preset contradiction determination threshold.

10. A power equipment fault data tracing and analysis system according to claim 9, characterized in that, The execution process of the causal network counterfactual correction function includes: First, locate one or more stubborn digital pieces of evidence that cause excessively high data fitting costs; Secondly, a new directed edge is generated as a candidate causal hypothesis. The source node of this new directed edge exists in the current lowest-cost candidate fault evidence chain, and the observable phenomenon associated with its target node can explain the stubborn digital evidence. Finally, the candidate causal hypothesis is sent to the diagnostic report and interaction module for user evaluation, and an update operation is performed on the causal knowledge base module after receiving confirmation from the user.

Images