A method and system for fault diagnosis of electric power equipment based on causal knowledge guidance

Abstract

The present invention discloses a method and system for diagnosing faults of electric power equipment guided by causal knowledge, belonging to the technical field of fault prediction. The method comprises: inputting the real-time monitoring data into the corresponding spatial mapping network according to the modality to obtain mapping features; sending the mapping features into the pre-trained fault diagnosis model to obtain the fault diagnosis results; the training process of the pre-trained fault diagnosis model comprises: constructing a knowledge-guided causal reasoning fault diagnosis model, introducing the causal relationship graph into the model as prior knowledge, guiding the training of the causal reasoning fault diagnosis model oriented to the fusion features, completing the training and reasoning tasks of the knowledge-guided causal reasoning model based on historical data, and obtaining the pre-trained fault diagnosis model. The present invention relates to the fault diagnosis of electric power equipment, realizes the fault diagnosis analysis of electric power equipment combined with maintenance knowledge, and provides the fault diagnosis results of electric power equipment with high accuracy and strong interpretability.

Description

Technical Field

[0001] The present invention proposes a method and system for diagnosing electric power equipment faults based on causal knowledge guidance, belonging to the field of fault diagnosis business. Background Art

[0002] Deep learning is a common method of artificial intelligence technology. It imitates the working mode of human brain neural network to build a multi-layer neural network model to automatically learn and extract complex features and patterns from data. This multi-layer structure allows deep learning models to represent data layer by layer, from low-level features (such as edges or colors) to high-level abstract concepts (such as objects or semantic information), enabling them to perform well in various tasks. The key elements of deep learning include neural network architecture, large-scale data sets, back-propagation algorithms, and powerful computing hardware. Neural networks are composed of many neurons, each of which is connected to the neurons in the previous layer and passes information by learning appropriate weights. Through a large amount of data and back-propagation algorithms, the model can continuously adjust these weights to minimize prediction errors and thus improve performance.

[0003] Data-driven AI technology can achieve real-time monitoring and remote diagnosis of the performance and status of power equipment. By collecting a large amount of sensor data and operation records, the system can detect potential problems in a timely manner, predict equipment failures, and provide timely alerts, thereby reducing the risk of power outages and equipment damage. Secondly, this technology can effectively process large-scale data and identify complex patterns and trends. This helps power companies better understand the operation of equipment, discover hidden failure modes, and take preventive maintenance measures to reduce repair costs and equipment downtime.

[0004] Although data-driven artificial intelligence technology has significant advantages in the field of power equipment fault diagnosis, it also has some potential disadvantages. One of them is that for certain specific fault types, the model may face the problem of inaccurate diagnosis, especially when the equipment has problems in rare or abnormal situations. This is because relying solely on data-driven models may not fully understand the complex working principles and failure mechanisms of power equipment, resulting in the model being unable to effectively identify rare or novel failure modes; in addition, there is less sample data on power equipment failures, which makes it difficult to meet the parameter fitting requirements of the guarantee model.

[0005] Most of the existing power equipment fault diagnosis methods are purely data-driven methods. However, in actual production environments, there is often a shortage of fault samples, which makes model training difficult. In addition, due to the black box characteristics of data-driven models, their output results lack interpretability, making it difficult to effectively assist on-site staff in equipment operation and maintenance. Summary of the invention

[0006] In view of the problem of low accuracy of power equipment fault diagnosis in the prior art, the present invention proposes a power equipment fault diagnosis method and system based on causal knowledge guidance. The present invention performs power equipment fault diagnosis based on causal knowledge guidance, realizes power equipment fault diagnosis analysis combined with maintenance knowledge, and provides power equipment fault diagnosis results with high accuracy and strong interpretability.

[0007] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0008] In a first aspect, the present invention provides a method for diagnosing faults of electric power equipment based on causal knowledge guidance, comprising:

[0009] The real-time monitoring data is input into the corresponding space mapping network according to the modality to obtain mapping features; the mapping features are input into the pre-trained fault diagnosis model to obtain the fault diagnosis results; the training process of the pre-trained fault diagnosis model includes:

[0010] Based on the existing knowledge of power equipment maintenance, the pseudo-causal intervention effects between various physical quantities and fault types are analyzed and mined, and the causal relationship diagram between multimodal physical information and fault phenomena is constructed based on the pseudo-causal intervention effects.

[0011] Construct a multimodal feature fusion module to normalize multiple physical quantities, perform feature concatenation, and perform intervention calculation based on the attention mechanism to obtain intervention features of the encoded feature changes;

[0012] A knowledge-guided causal reasoning fault diagnosis model is constructed, and the causal relationship graph is introduced into the model as prior knowledge to guide the training of the causal reasoning fault diagnosis model for fusion features, and the training and reasoning tasks of the knowledge-guided causal reasoning model based on historical data are completed to obtain the pre-trained fault diagnosis model.

[0013] As a further improvement of the present invention, the method of analyzing and mining the pseudo-causal intervention effects between multiple physical quantities and fault types based on the existing knowledge of power equipment maintenance, and constructing a causal relationship diagram between multimodal physical information and fault phenomena based on the pseudo-causal intervention effects, specifically includes:

[0014] According to the fault type, measured physical quantity, and measurement cause, the equipment mechanism knowledge is summarized and extracted, a dynamic equipment mechanism knowledge database is created, and the equipment mechanism knowledge database is gradually supplemented and completed;

[0015] Characterize the equipment mechanism knowledge inside the power equipment in the database, and convert the equipment mechanism knowledge into a mathematical expression for constructing a cause-effect relationship diagram in the form of an indicator function;

[0016] Arrange the already sorted indicator functions into an indicator function set ,Based on the pseudo causal relationship between multi-modal physical information and fault phenomena, a causal relationship graph between multi-modal physical information and fault phenomena is constructed.

[0017] As a further improvement of the present invention, the feature characterization of the equipment mechanism knowledge inside the power equipment in the database and the conversion of the equipment mechanism knowledge into a mathematical expression for constructing a cause-effect relationship diagram in the form of an indicator function include:

[0018] by Indicates the parameter fault type, Indicates potentially related components, Represent the measured physical quantities related to the fault and construct the indicator function for causal relationship representation:

[0019] ;

[0020] in, Represents an indicator function, which is used to determine whether an event has occurred. The value of an event that has been determined to have occurred is 1, and the value of an event that has been determined not to have occurred is 0; Expresses the expectation of an event. To intervene in the situation means to intervene in some way in the current state of things; Indicates when the components may be related Type of parameter failure When the measured physical quantity related to the fault The changes that would occur, indicating the results of the representation, would serve as pseudo-intervention information.

[0021] As a further improvement of the present invention, the already sorted indicator functions are sorted into an indicator function set. , based on the pseudo causal relationship between multi-modal physical information and fault phenomena, a causal relationship diagram between multi-modal physical information and fault phenomena is constructed. The specific steps are as follows:

[0022] Determine each node in the causal relationship diagram, including component nodes, fault type nodes, parameter fault type nodes, and parameter change nodes; connect component nodes, parameter fault type nodes, parameter change nodes, and fault type nodes one by one to construct the essential causal relationship diagram , and initialize the final output causal relationship graph , a set of indicator functions that record the intervened variables ; Represents a dataset;

[0023] Will The set of all connected blocks on , initialize the connected block set of the ancestor edge to be determined for ,in, For collection A connected block in Indicates that the number of connected block parameters is greater than 1, Indicates the existence of a set A connected block in The number of parameters is greater than 1, Representation diagram The variable The set of all ancestor variables of Represents a connected block There are variables in The ancestor variable of When, collection A connected block in There are variables in ancestor variable;

[0024] Cycle judgment Is it an empty set? If it is an empty set, end the causal relationship diagram The loop updates and jumps out of the loop; if it is not an empty set, the causal relationship diagram is started The cycle update from Randomly select a connected block from , and in Among all the unintervention variables, select the variable with the most undirected edges as the pseudo intervention variable , calculated based on Local causal relationship diagram of is the number of local causal relationship graphs, according to the indicator function set Neutralizing variables Correlation indicator function value to local causal relationship graph Update the edges and transform all local causal graphs The public directed edges in are updated to In the updated causal relationship diagram and connected block set , and proceed to the next cycle;

[0025] After the judgment cycle is completed, a causal relationship diagram between multimodal physical information and fault phenomena is finally obtained.

[0026] As a further improvement of the present invention, the multimodal feature fusion module is constructed to normalize multiple physical quantities, fuse features, and perform intervention calculation processing based on the attention mechanism to obtain intervention features of the encoded feature changes, including:

[0027] Each modal feature set Input corresponding spatial mapping network , get the feature information vector set ,in, , k is the number of modes, Represents batch normalization operation express The number of features in represents the dimension of a single feature vector after spatial mapping, Indicates the size of the set; according to the feature annotation, select In the figure, the feature vector corresponding to the normal operation of the equipment is calculated by the mean to obtain the mean feature vector of the normal operation of the corresponding mode. ;

[0028] For each modal feature information vector that has been obtained Perform feature splicing to obtain a fused feature set [ ] and normal operating characteristics [ ];

[0029] Constructing an intervention network based on attention mechanism Changes in encoding features, intervention network based on attention mechanism The weights are initialized to ,Will Send to the network Get intervention features of changes in coded features ; Complete dynamic fusion of multimodal physical information.

[0030] As a further improvement of the present invention, the construction of a knowledge-guided causal reasoning fault diagnosis model introduces a causal relationship graph as prior knowledge into the model to guide the training of the intervention-feature-oriented causal reasoning fault diagnosis model, including:

[0031] Based on the cause-effect diagram and intervention characteristics , build a fault diagnosis module guided by causal knowledge;

[0032] Cause and effect diagram The nodes in the multimodal feature set are extracted and divided into two parts: fault category and measurement parameter. The number of nodes in the fault category is equal to the number of label categories in the multimodal feature set, and the number of nodes in the measurement parameter part is consistent with the number of modes in the multimodal feature set. When extracting the measurement parameter nodes, the nodes of the same mode are fused into one point, and the corresponding edge weights are connected to the fault category nodes through accumulation as new edges to form a new edge weight matrix. ;

[0033] The edge weight matrix Set as a guidance module based on the attention mechanism Initial weight value of intervention feature It will first be sent to the guidance module and then to the classifier network , and obtain the final fault diagnosis result , ;

[0034] The loss function L is used as the loss function of the model, combined with the fault diagnosis network model guided by causal knowledge, and after initializing the network parameters, the stochastic gradient descent method is used to complete the model training to obtain the pre-trained fault diagnosis model.

[0035] As a further improvement of the present invention, the loss function is as follows:

[0036]

[0037] in, and They represent the number of labeled samples and the number of unlabeled samples, respectively. represents the true standard of the i-th sample in the labeled sample, represents the recognition result of the i-th sample in the labeled samples, represents the recognition result of the jth sample in the unlabeled sample. The first term in the loss function is the standard cross entropy loss, and the second term is the standard self-information entropy loss. Represents a logarithmic function.

[0038] In a second aspect, the present invention provides a method for diagnosing faults of electric power equipment based on causal knowledge guidance, comprising:

[0039] A feature mapping module is used to input the real-time monitoring data into the corresponding spatial mapping network according to the modality to obtain mapping features;

[0040] A fault diagnosis module is used to input the mapping features into a pre-trained fault diagnosis model to obtain a fault diagnosis result;

[0041] A pre-training module, used for the training process of the pre-trained fault diagnosis model, comprises:

[0042] A causal relationship construction unit is used to analyze and mine the pseudo-causal intervention effects between multiple physical quantities and fault types based on the existing knowledge of power equipment maintenance, and to construct a causal relationship diagram between multimodal physical information and fault phenomena based on the pseudo-causal intervention effects;

[0043] The intervention feature acquisition unit is used to construct a multimodal feature fusion module, normalize multiple physical quantities, perform feature concatenation, and perform intervention calculation processing based on the attention mechanism to obtain the intervention features of the encoded feature changes;

[0044] The model training unit is used to construct a knowledge-guided causal reasoning fault diagnosis model, introduce the causal relationship diagram into the model as prior knowledge, guide the training of the causal reasoning fault diagnosis model for fusion features, complete the training and reasoning tasks of the knowledge-guided causal reasoning model based on historical data, and obtain the pre-trained fault diagnosis model.

[0045] In a third aspect, the present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the causal knowledge-guided power equipment fault diagnosis method when executing the computer program.

[0046] In a fourth aspect, the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for diagnosing faults of power equipment guided by causal knowledge is implemented.

[0047] In a fifth aspect, the present invention provides a computer program product, wherein the computer program product includes computer instructions, and the computer instructions instruct a computer to execute the power equipment fault diagnosis method guided by causal knowledge.

[0048] The beneficial effects of the present invention compared with the prior art are as follows:

[0049] The present invention is based on a method for diagnosing faults of electric power equipment guided by causal knowledge, including a strategy for constructing a causal relationship graph based on maintenance knowledge, a strategy for constructing a multimodal information fusion network module, and a maintenance knowledge guidance strategy for characterizing maintenance knowledge of electric power equipment and producing a causal relationship graph based on pseudo-intervention effects. The existing maintenance knowledge is characterized, and the pseudo-intervention effects between various physical quantities and fault phenomena are analyzed and mined. Based on the pseudo-intervention effects, the corresponding causal effects are calculated using an active causal effect discovery algorithm to construct a causal relationship graph between multimodal physical information and fault phenomena. In the training process of the fault diagnosis model, the feature changes and the causal relationship graph constructed based on maintenance knowledge are encoded into the middle layer features of the network through a network structure based on an attention mechanism, thereby realizing model training guided by maintenance knowledge and reducing the model's dependence on data. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the embodiments of the present invention or the drawings of related technical solutions in the prior art are introduced below. It should be understood that the drawings introduced below are only for the convenience of clearly describing some embodiments of the technical solutions of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0051] Figure 1 The flowchart of the power equipment fault diagnosis method based on causal knowledge guidance is shown in FIG.

[0052] Figure 2 Flowchart of the training process for the pre-trained fault diagnosis model;

[0053] Figure 3 This is a flow chart of the algorithm scheme of an embodiment of the present invention;

[0054] Figure 4 It is a schematic diagram of nodes and corresponding edge weight matrices;

[0055] Figure 5 This is a diagram of the fault diagnosis system of power equipment guided by causal knowledge;

[0056] Figure 6 The figure is a schematic diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0057] The embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and are not to be construed as limitations of the present invention. For the step numbers in the following embodiments, they are only provided for the convenience of explanation, and the order between the steps is not limited in any way, and the execution order of each step in the embodiment can be adaptively adjusted according to the understanding of those skilled in the art.

[0058] In the description of the present invention, unless otherwise clearly defined, terms such as setting, installing, connecting, etc. should be understood in a broad sense, and technicians in the relevant technical field can reasonably determine the specific meanings of the above terms in the present invention based on the specific content of the technical solution.

[0059] Definition of terms:

[0060] FISVDD, Fast Incremental Support Vector Data Description, is an extended algorithm based on Support Vector Data Description (SVDD).

[0061] Gated Recurrent Unit, GRU, is a neural network model commonly used in natural language processing and sequence data processing.

[0062] Adam algorithm, Adaptive Moment Estimation, is a first-order gradient optimization algorithm based on low-order moment adaptive stochastic objective function. It combines the advantages of momentum method and adaptive learning rate and is one of the most popular and commonly used optimization algorithms in deep learning.

[0063] In order to overcome the shortcomings of the background technology, the knowledge of power equipment faults can be used as auxiliary input information for deep learning model training. Incorporating physical characteristics into model training can increase the interpretability and explainability of the model. In turn, the model can better understand the working principle of the power equipment and increase the robustness of the power equipment fault diagnosis system. Even in the case of poor data quality or incomplete data, the model can still rely on the information of physical characteristics for diagnosis, reducing the dependence on high-quality data. By combining the knowledge of power equipment faults with deep learning models, the shortcomings of pure data-driven methods can be compensated and the reliability and applicability of power equipment fault diagnosis can be improved.

[0064] like Figure 1 As shown, the present invention proposes a method for diagnosing power equipment faults based on causal knowledge guidance, comprising:

[0065] S1, input the real-time monitoring data into the corresponding spatial mapping network according to the modality to obtain mapping features;

[0066] S2, sending the mapping features into the pre-trained fault diagnosis model to obtain the fault diagnosis results;

[0067] Compared with traditional fault diagnosis methods, the power equipment fault diagnosis method based on causal knowledge guidance of the present invention improves diagnostic accuracy. By inputting real-time monitoring data into the corresponding spatial mapping network according to the modality, the method can more effectively extract feature information in the data.

[0068] Among them, the pre-trained fault diagnosis model utilizes a large amount of historical data and causal knowledge, which can more accurately identify fault modes, reduce misdiagnosis and missed diagnosis, and thus improve the accuracy of fault diagnosis. Since this method is based on the concepts of modal and spatial mapping, it can process various types of power equipment data. This flexibility enables this method to be applied to a wide range of power equipment and enhances its generalization ability. The pre-trained model has learned a variety of fault modes and normal states during the training process, so it can also give more accurate diagnostic results when facing new or unknown faults.

[0069] Furthermore, the instant processing of real-time monitoring data and the rapid extraction of mapping features enable this method to provide fault diagnosis results in a short time. This is particularly important for power equipment faults that require a quick response, which can reduce downtime and economic losses. The automated and intelligent diagnosis process reduces the need for manual intervention, improves diagnostic efficiency, and reduces the possibility of human error.

[0070] Therefore, this method places special emphasis on the guidance of causal knowledge, which means that in the fault diagnosis process, not only the correlation between data is considered, but also the causal relationship of the fault occurrence is deeply analyzed. This in-depth analysis can reveal the essential cause of the fault and provide more valuable guidance for subsequent fault repair and prevention. By combining causal knowledge, this method can predict potential fault risks and take measures to intervene in advance, thereby avoiding the occurrence of faults or reducing the impact of faults.

[0071] like Figure 2 As shown, the training process of the pre-trained fault diagnosis model includes the following steps:

[0072] S101, based on the existing knowledge of power equipment maintenance, analyze and mine the pseudo-causal intervention effects between multiple physical quantities and fault types, and construct a causal relationship diagram between multimodal physical information and fault phenomena based on the pseudo-causal intervention effects;

[0073] S102, constructing a multimodal feature fusion module, normalizing multiple physical quantities, feature concatenation, and intervention calculation processing based on an attention mechanism to obtain intervention features of encoded feature changes;

[0074] S103, construct a knowledge-guided causal reasoning fault diagnosis model, introduce the causal relationship diagram into the model as prior knowledge, guide the training of the causal reasoning fault diagnosis model for fusion features, complete the training and reasoning tasks of the knowledge-guided causal reasoning model based on historical data, and obtain the pre-trained fault diagnosis model.

[0075] First, based on the equipment maintenance knowledge, the pseudo-causal effects between multiple physical quantities are analyzed and mined to form causal prior knowledge. Secondly, multimodal physical information is dynamically fused. Finally, based on the constructed causal relationship graph and multi-source fusion features, a knowledge-guided causal reasoning fault diagnosis model framework is constructed. In the fault diagnosis process, causal prior knowledge is used to guide model training. In the scenario of data scarcity, the hybrid drive of maintenance knowledge and historical data is realized, thereby improving the accuracy and reliability of fault identification.

[0076] The core principle of the power equipment fault diagnosis method based on causal knowledge guidance of the present invention is to combine causal reasoning with deep learning to deal with complex power equipment fault diagnosis problems where data may be insufficient. The principle is as follows:

[0077] First, the present invention analyzes the existing knowledge of power equipment maintenance and mines the pseudo-causal intervention effects between various physical quantities and fault types. The "pseudo-causality" here means that in a complex system, the direct influence relationship between physical quantities may not be a strict causal relationship, but there is a statistical correlation or intervention effect. Based on these pseudo-causal intervention effects, a causal relationship diagram between multimodal physical information and fault phenomena is constructed. This step converts the expert's prior knowledge into a graph structure that can be understood by the model, providing guidance for subsequent model training.

[0078] For example, the introduction of causal knowledge as a priori guidance enables the model to pay more attention to the actual relationship between physical quantities during training, reduce the possibility of misjudgment, and thus improve the accuracy of fault diagnosis. The causal relationship diagram provides a global perspective that helps the model understand the interaction between different physical quantities, thereby enhancing the model's generalization ability in different scenarios.

[0079] For example, in scenarios where data is scarce, traditional data-driven methods often fail to achieve good results. However, this method can reduce the reliance on data to a certain extent by introducing causal knowledge and multimodal feature fusion, and realize the hybrid drive of maintenance knowledge and historical data.

[0080] Secondly, a multimodal feature fusion module is constructed to normalize multiple physical quantities from different sensors to ensure that data of different dimensions can be compared on the same scale. Through feature concatenation and intervention calculation processing based on the attention mechanism, intervention features of the encoded feature changes are obtained. The role of the attention mechanism here is to help the model pay more attention to features that have an important impact on fault diagnosis and improve the accuracy of diagnosis.

[0081] Exemplarily, the multimodal feature fusion module and the intervention computation processing based on the attention mechanism enable the model to extract information useful for fault diagnosis more quickly, thereby improving the diagnostic efficiency.

[0082] Finally, the constructed causal relationship graph is introduced into the model as prior knowledge to guide the training of the causal reasoning fault diagnosis model for fusion features. This step enables the model to utilize the prior knowledge of experts during the training process and reduce the dependence on the amount of data, especially in scenarios with scarce data, so that effective fault diagnosis can still be performed. By training and reasoning the knowledge-guided causal reasoning model based on historical data, a pre-trained model that can accurately identify power equipment faults is finally obtained.

[0083] For example, the introduction of the causal relationship diagram makes the decision-making process of the model more transparent and explainable, helping users understand why the model makes a specific diagnosis result. It can significantly improve the accuracy and reliability of fault diagnosis while reducing the dependence on the amount of data.

[0084] The causal knowledge-guided power equipment fault diagnosis method uses a causal relationship diagram constructed based on maintenance knowledge to guide model training during the training process of the fault diagnosis model, thereby reducing the model's dependence on data and improving the model's reliability and accuracy.

[0085] For the same type of power equipment, a universal power equipment fault diagnosis modeling method based on causal knowledge guidance is proposed. After the model training is completed, it can be flexibly applied to quickly perform fault diagnosis and avoid repeated training.

[0086] As a further optimization of the present invention, based on the existing knowledge of power equipment maintenance, the pseudo-causal intervention effect between multiple physical quantities and fault types is analyzed and mined, and a causal relationship diagram between multimodal physical information and fault phenomena is constructed based on the pseudo-causal intervention effect, specifically including:

[0087] According to the fault type, measured physical quantity, and measurement cause, the equipment mechanism knowledge is summarized and extracted, a dynamic equipment mechanism knowledge database is created, and the equipment mechanism knowledge database is gradually supplemented and completed;

[0088] Characterize the equipment mechanism knowledge inside the power equipment in the database and convert it into a mathematical expression for constructing a cause-effect relationship diagram as an indicator function;

[0089] Arrange the already sorted indicator functions into an indicator function set ,Based on the pseudo causal relationship between multi-modal physical information and fault phenomena, a causal relationship graph between multi-modal physical information and fault phenomena is constructed.

[0090] Exemplarily, the process of constructing a causal relationship diagram between multimodal physical information and fault phenomena based on pseudo-causal intervention effects in the present invention can significantly improve the accuracy and flexibility of fault diagnosis by introducing a dynamic equipment mechanism knowledge database and the conversion of mathematical expressions. By summarizing and extracting equipment mechanism knowledge in the form of fault type, measured physical quantity, measurement cause, etc., and creating a dynamic equipment mechanism knowledge database, it can be ensured that the knowledge base is continuously updated and improved as new data is added and new knowledge is discovered. This dynamism enables the model to cope with changing equipment states and new fault modes.

[0091] Furthermore, the equipment mechanism knowledge in the database is characterized and converted into indicator functions, which can more accurately describe the pseudo-causal intervention effect between physical quantities and fault types. The introduction of indicator functions not only facilitates computer processing, but also reduces the subjectivity of human interpretation and improves the accuracy of diagnosis. The organized set of indicator functions is used to construct a causal relationship diagram, which can systematically display the complex relationship between multimodal physical information and fault phenomena. This graphical representation is not only intuitive and easy to understand, but also helps to discover potential knowledge loopholes and errors, facilitating subsequent maintenance and correction. The causal relationship diagram constructed based on a complete equipment mechanism knowledge database and an accurate set of indicator functions can cover a wider range of fault types and physical quantity combinations. This enables the model to make reasonable inferences based on existing knowledge when facing new or unknown fault conditions, thereby improving the generalization ability of the model.

[0092] Furthermore, the process of building a dynamic equipment mechanism knowledge database and causal relationship diagram is actually to transform the expert's experiential knowledge into a shareable and reusable digital resource. This not only helps the knowledge inheritance and collaboration within the team, but also provides reference and reference for research in other related fields.

[0093] The pseudo-causal intervention effect between physical quantities and fault types is displayed through the causal relationship diagram, making the fault diagnosis process more transparent and explainable. This helps users understand why the model makes a specific diagnostic result and improves users' trust and acceptance of the diagnostic results. The process of constructing a causal relationship diagram between multimodal physical information and fault phenomena based on the pseudo-causal intervention effect can significantly improve the accuracy, flexibility, generalization ability and explainability of fault diagnosis, and provide more solid technical support for intelligent fault diagnosis of power equipment.

[0094] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0095] The purpose of the present invention is to establish a method for diagnosing power equipment faults based on causal knowledge guidance. Specifically, a method for producing a causal relationship diagram for power equipment maintenance knowledge is proposed. According to the maintenance knowledge, the pseudo-causal intervention effect between multiple physical quantities is analyzed and mined, and a causal relationship diagram between fault phenomena and multimodal physical information is constructed based on this pseudo-causal intervention effect to support the construction of a knowledge-guided data-driven model framework.

[0096] Specifically, the power equipment fault diagnosis method based on causal knowledge guidance includes a causal relationship graph construction strategy based on maintenance knowledge, a multimodal information fusion network module construction strategy, and a maintenance knowledge guidance model training strategy. Figure 3 As shown, the technical route proposed by the present invention is as follows:

[0097] Firstly, based on the existing maintenance knowledge, the pseudo-causal intervention effect between various physical quantities and fault types is analyzed and mined. Based on the pseudo-causal intervention effect, a causal relationship diagram between multimodal physical information such as oil chromatography, optical signal, soundprint signal, vibration signal, text data and fault phenomenon is constructed.

[0098] Secondly, a multimodal feature fusion module is constructed to realize the normalization, feature splicing, and intervention calculation based on the attention mechanism of feature quantities such as oil chromatography, optical signals, voiceprint signals, vibration signals, and text data, so as to realize the dynamic fusion of multimodal physical information; thirdly, a knowledge-guided causal reasoning fault diagnosis model is constructed, and the causal relationship graph is introduced into the model as prior knowledge to guide the training of the causal reasoning fault diagnosis model for fusion features;

[0099] Finally, the training and reasoning tasks of the knowledge-guided causal reasoning model based on historical data are completed to realize intelligent diagnosis of power equipment faults.

[0100] The present invention solves the problem of low interpretability of traditional data methods by creating a power equipment fault diagnosis method guided by causal knowledge, and also solves the problem of low accuracy and reliability of the fault diagnosis model when there are few fault cases and insufficient data samples. The specific steps proposed by the present invention are as follows:

[0101] Step 1: Create a cause-effect relationship diagram based on maintenance knowledge.

[0102] By summarizing and generalizing the existing knowledge on power equipment maintenance, the equipment mechanism knowledge inside the power equipment is characterized and converted into a mathematical expression (such as an exponential function) for constructing a causal relationship diagram. Finally, a causal relationship diagram between multimodal physical information such as oil chromatography, optical signals, voiceprint signals, vibration signals, text data and fault phenomena is constructed based on the converted mathematical expression to support subsequent causal knowledge-guided model training.

[0103] Step 1: Construction of equipment mechanism knowledge base. Taking the transformer as an example, Table 1 summarizes and extracts transformer equipment mechanism knowledge in the form of fault type, measured physical quantity, and measurement cause (fault phenomenon) based on the feedback from front-line transformer inspection personnel and important reference materials such as the power industry standards "Guidelines for Analysis and Judgment of Dissolved Gases in Transformer Oil" and "Guidelines for Maintenance of Power Transformers", and creates a dynamic equipment mechanism knowledge database. With the accumulation of experience, the equipment mechanism knowledge database can be gradually supplemented and completed.

[0104] Table 1 Equipment mechanism knowledge database

[0105]

[0106] Step 2: Equipment mechanism knowledge representation. As the equipment mechanism knowledge database is continuously improved, the equipment mechanism knowledge in the database is characterized to support the subsequent cause-effect relationship diagram construction. Indicates the parameter fault type, Indicates potentially related components, Represents the measured physical quantity related to the fault and constructs an indicator function for causal relationship representation In the above characterization, Represents an indicator function, which is used to determine whether an event has occurred. The value of an event that has been determined to have occurred is 1, and the value of an event that has been determined not to have occurred is 0. Expresses the expectation of an event. To intervene in the current state of things means to intervene in some way. Indicates when the component occur In case of failure, the parameter The results of the indication representation will be used as pseudo-intervention information to support the construction of the subsequent causal relationship diagram. Taking the equipment mechanism knowledge in Table 1 as an example, the knowledge representation is performed according to the following rules.

[0107] Table 2 Equipment mechanism knowledge representation

[0108]

[0109] Taking the knowledge characterization process of the core grounding fault equipment mechanism in Table 2 as an example, It means that when the core has a ground fault, the operating core ground current will be greater than 300mA. Indicates that the core grounding current is greater than 300mA. Indicates that a core ground fault has occurred in the core.

[0110] Step 3: Create a cause-effect relationship diagram. Arrange the indicator functions that have been sorted into an indicator function set. , based on the pseudo causal relationship between these multi-physical modal information and the fault phenomenon, the corresponding causal relationship diagram is made. The specific steps are as follows:

[0111] Step 3.1: Identify the nodes in the causal relationship diagram, including components Node, Failure Type Node, parameter fault type Node and parameter changes Node. Connect component nodes, parameter fault type nodes, parameter change nodes and fault type nodes one by one to build an essential causal relationship diagram , and initialize the final output causal relationship graph and a set of indicator functions that record the intervened variables ; Represents a dataset.

[0112] Step 3.2: The set of all connected blocks on , initialize the connected block set of the ancestor edge to be determined for ,in, For collection A connected block in Indicates that the number of connected block parameters is greater than 1, Indicates the existence of a set A connected block in The number of parameters is greater than 1, Representation diagram The variable The set of all ancestor variables of Represents a connected block There are variables in ancestor variable, so when When A connected block in There are variables in The ancestor variable of From From a variable in For a directed path, there are undetermined ancestor edges, and further intervention instructions need to be introduced.

[0113] Step 3.3: Loop judgment Is it an empty set? If it is an empty set, end the causal relationship diagram The loop updates and jumps out of the loop; if it is not an empty set, the causal relationship diagram is started The cyclic update of Randomly select a connected block from , and in Among all the unintervention variables, select the variable with the most undirected edges as the pseudo intervention variable , calculated based on Local causal relationship diagram of is the number of local causal relationship graphs, according to the indicator function set and variables Related indicator functions for local causal graphs Update the edges.

[0114] Taking the knowledge characterization process of the core grounding fault equipment mechanism in Table 2 as an example, , where the corresponding component node , Fault Type Node , parameter fault type node And parameter change nodes , according to the indicator function, retain point to The directed edge of point to The directed edge of point to directed edges, and all local causal graphs The public directed edges in are updated to In the updated causal relationship diagram and connected block set , and proceed to the next cycle.

[0115] After the judgment cycle is completed, a causal relationship diagram between multimodal physical information and fault phenomena is finally obtained.

[0116] Step 2: Construction of multimodal feature fusion module. Through normalization of oil chromatography, light signal, voiceprint signal, vibration signal, text data and other feature quantities, feature splicing, intervention calculation based on attention mechanism and other steps, dynamic fusion of multimodal physical information is realized. The specific process is as follows:

[0117] Step 1: Multimodal feature extraction. Extract each modal feature set , k is the number of modes, input the corresponding spatial mapping network Get feature information vector set ,in, Represents batch normalization operation express The number of features in represents the dimension of a single feature vector after spatial mapping, Indicates the size of the set. At the same time, according to the feature annotation, select In the figure, the feature vector corresponding to the normal operation of the equipment is calculated by the mean to obtain the mean feature vector of the normal operation of the corresponding mode. (Hereafter referred to as normal characteristics).

[0118] Step 2: Multimodal feature fusion. Perform feature splicing to obtain a fused feature set [ ] and normal operating characteristics [ ]. In the subsequent equipment fault diagnosis process based on causal reasoning, the influence of equipment state changes on the changes of each modal input feature is complex and cannot be simply represented by feature values. In order to more clearly represent the influence of input feature changes, the present invention designs an intervention network based on the attention mechanism To encode feature changes, the subsequent reasoning network pays more attention to the modal features with drastic changes, so as to further improve the effectiveness of the reasoning model. The weight of the network is initialized as ,Will Send to the network Get intervention features of changes in coded features .

[0119] Step 3: Construct a fault diagnosis model guided by causal knowledge. Based on the causal relationship diagram obtained in step 1 and the intervention characteristics obtained in step 2 , a fault diagnosis module guided by causal knowledge was designed.

[0120] Compared with the intervention network based on the attention mechanism in step 2 Similarly, in order to encode known knowledge information into the fault diagnosis module, the present invention converts the cause-effect relationship diagram into The nodes in the multimodal feature set are extracted and divided into two parts: fault category and measurement parameter. The number of nodes in the fault category is equal to the number of label categories in the multimodal feature set, and the number of nodes in the measurement parameter part is consistent with the number of modes in the multimodal feature set. When extracting the measurement parameter nodes, the nodes of the same mode are fused into one point, and the corresponding edge weights are connected to the fault category nodes through accumulation as new edges to form a new edge weight matrix. .

[0121] like Figure 4 As shown, the node extraction and the corresponding edge weight matrix are Figure 4 In the upper left part, the node fusion and the corresponding edge weight matrix are Figure 4 In the upper right part, triangles and squares represent measurement parameter nodes of different modes, and circles represent fault category nodes.

[0122] The edge weight matrix Set as a guidance module based on the attention mechanism Initial weight value of intervention feature It will first be sent to the guidance module and then to the classifier network , and obtain the final fault diagnosis result ,Right now , the overall loss function of the network is as follows:

[0123]

[0124] in, and They represent the number of labeled samples and the number of unlabeled samples, respectively. represents the true standard of the i-th sample in the labeled sample, represents the recognition result of the i-th sample in the labeled samples, It represents the recognition result of the jth sample in the unlabeled sample. The first term of the loss function is the standard cross entropy loss, and the second term is the standard self-information entropy loss.

[0125] Step 4: Training and application of power equipment fault diagnosis model guided by causal knowledge.

[0126] The loss function L is used as the loss function of the whole model, combined with the causal knowledge-guided fault diagnosis network model established in steps 1 to 3, and after initializing the network parameters, the model training is completed using the stochastic gradient descent method. Based on the trained fault diagnosis model, dynamic parameter updates and fine-tuning are performed in combination with the actual situation on site, and field application work can be realized, that is, the real-time monitoring data is input into the corresponding spatial mapping network according to its modal input to obtain mapping features, and then sent to the subsequent reasoning model to obtain the fault diagnosis results, so as to realize the intelligent diagnosis of power equipment faults.

[0127] The present invention proposes a method for producing a causal relationship diagram for power equipment maintenance knowledge. Based on the maintenance knowledge, the pseudo causal effects between multiple physical quantities are analyzed and mined, and a causal relationship diagram between multimodal physical information and fault phenomena is constructed to support the construction of a knowledge-guided data-driven model framework.

[0128] During the training process of the fault diagnosis model, the feature changes and the causal relationship diagram constructed based on maintenance knowledge are encoded into the middle-layer features of the network through a network structure based on the attention mechanism, thereby guiding the model training based on maintenance knowledge, reducing the model's dependence on data, and improving the reliability and accuracy of the model.

[0129] For the same type of power equipment, the universal power equipment fault diagnosis modeling method based on causal knowledge guidance proposed in the embodiment of the present invention can be flexibly applied to quickly perform fault diagnosis after the model training is completed, avoiding repeated training.

[0130] As an improved solution of the present invention, the model parameter adjustment optimization method can be replaced by other adjustment methods, such as the Adam algorithm, etc. The present invention is not limited to a specific algorithm.

[0131] like Figure 5 As shown, the second object of the present invention is to provide a power equipment fault diagnosis system guided by causal knowledge, comprising:

[0132] The feature mapping module 100 is used to input the real-time monitoring data into the corresponding spatial mapping network according to the modality to obtain mapping features;

[0133] The fault diagnosis module 200 is used to input the mapping features into a pre-trained fault diagnosis model to obtain a fault diagnosis result;

[0134] The pre-training module 300 is used for the training process of the pre-trained fault diagnosis model. The pre-training module 300 includes:

[0135] A causal relationship construction unit is used to analyze and mine the pseudo-causal intervention effects between multiple physical quantities and fault types based on the existing knowledge of power equipment maintenance, and to construct a causal relationship diagram between multimodal physical information and fault phenomena based on the pseudo-causal intervention effects;

[0136] The intervention feature acquisition unit is used to construct a multimodal feature fusion module, normalize multiple physical quantities, perform feature concatenation, and perform intervention calculation processing based on the attention mechanism to obtain the intervention features of the encoded feature changes;

[0137] The model training unit is used to construct a knowledge-guided causal reasoning fault diagnosis model, introduce the causal relationship diagram into the model as prior knowledge, guide the training of the causal reasoning fault diagnosis model for fusion features, complete the training and reasoning tasks of the knowledge-guided causal reasoning model based on historical data, and obtain the pre-trained fault diagnosis model.

[0138] As an example, the causal relationship building unit summarizes and extracts equipment mechanism knowledge in the form of fault type, measured physical quantity, and measurement cause, creates a dynamic equipment mechanism knowledge database, and gradually supplements and completes the equipment mechanism knowledge database;

[0139] The causal relationship construction unit characterizes the equipment mechanism knowledge inside the power equipment in the database, and converts the equipment mechanism knowledge into a mathematical expression for constructing a causal relationship diagram in the form of an indicator function;

[0140] The causal relationship building unit organizes the already organized indicator functions into an indicator function set ,Based on the pseudo causal relationship between multi-modal physical information and fault phenomena, a causal relationship graph between multi-modal physical information and fault phenomena is constructed.

[0141] As an example, the causal relationship building unit is Indicates the parameter fault type, Indicates potentially related components, Represents the measured physical quantity related to the fault and constructs an indicator function for causal relationship representation ;

[0142] in, Represents an indicator function, which is used to determine whether an event has occurred. The value of an event that has been determined to have occurred is 1, and the value of an event that has been determined not to have occurred is 0; Expresses the expectation of an event. To intervene in the situation means to intervene in some way in the current state of things; Indicates when the components may be related Type of parameter failure When the measured physical quantity related to the fault The changes that will occur, indicating the results of the representation, will serve as pseudo-intervention information.

[0143] As an example, the already sorted indicator functions are sorted into an indicator function set. , based on the pseudo causal relationship between multi-modal physical information and fault phenomena, a causal relationship diagram between multi-modal physical information and fault phenomena is constructed. The specific steps are as follows:

[0144] Determine each node in the causal relationship diagram, including component nodes, fault type nodes, parameter fault type nodes, and parameter change nodes; connect component nodes, parameter fault type nodes, parameter change nodes, and fault type nodes one by one to construct the essential causal relationship diagram , and initialize the final output causal relationship graph , a set of indicator functions that record the intervened variables ; Represents a dataset;

[0145] Will The set of all connected blocks on , initialize the connected block set of the ancestor edge to be determined for ,in, For collection A connected block in Indicates that the number of connected block parameters is greater than 1, Indicates the existence of a set A connected block in The number of parameters is greater than 1, Representation diagram The variable The set of all ancestor variables of Represents a connected block There are variables in The ancestor variable of When, collection A connected block in There are variables in ancestor variable;

[0146] Cycle judgment Is it an empty set? If it is an empty set, end the causal relationship diagram The loop updates and jumps out of the loop; if it is not an empty set, the causal relationship diagram is started The cycle update from Randomly select a connected block from , and in Among all the unintervention variables, select the variable with the most undirected edges as the pseudo intervention variable , calculated based on Local causal relationship diagram of is the number of local causal relationship graphs, according to the indicator function set Neutralizing variables Correlation indicator function value to local causal relationship graph Update the edges and transform all local causal graphs The public directed edges in are updated to In the updated causal relationship diagram and connected block set , and proceed to the next cycle;

[0147] After the judgment cycle is completed, a causal relationship diagram between multimodal physical information and fault phenomena is finally obtained.

[0148] As an example, the multimodal feature fusion module is constructed to normalize multiple physical quantities, fuse features, and perform intervention calculation processing based on the attention mechanism to obtain intervention features of the encoded feature changes, including:

[0149] Each modal feature set Input corresponding spatial mapping network , get the feature information vector set ,in, , k is the number of modes, Represents batch normalization operation express The number of features in represents the dimension of a single feature vector after spatial mapping, Indicates the size of the set; according to the feature annotation, select In the figure, the feature vector corresponding to the normal operation of the equipment is calculated by the mean to obtain the mean feature vector of the normal operation of the corresponding mode. ;

[0150] For each modal feature information vector that has been obtained Perform feature splicing to obtain a fused feature set [ ] and normal operating characteristics [ ];

[0151] Constructing an intervention network based on attention mechanism Changes in encoding features, intervention network based on attention mechanism The weights are initialized to ,Will Send to the network Get intervention features of changes in coded features ; Complete dynamic fusion of multimodal physical information.

[0152] As an example, the construction of a knowledge-guided causal reasoning fault diagnosis model introduces a causal relationship graph as prior knowledge into the model to guide the training of the intervention-feature-oriented causal reasoning fault diagnosis model, including:

[0153] Based on the cause-effect diagram and intervention characteristics , build a fault diagnosis module guided by causal knowledge;

[0154] Cause and effect diagram The nodes in the multimodal feature set are extracted and divided into two parts: fault category and measurement parameter. The number of nodes in the fault category is equal to the number of label categories in the multimodal feature set, and the number of nodes in the measurement parameter part is consistent with the number of modes in the multimodal feature set. When extracting the measurement parameter nodes, the nodes of the same mode are fused into one point, and the corresponding edge weights are connected to the fault category nodes through accumulation as new edges to form a new edge weight matrix. ;

[0155] The edge weight matrix Set as a guidance module based on the attention mechanism Initial weight value of intervention feature It will first be sent to the guidance module and then to the classifier network , and obtain the final fault diagnosis result , ;

[0156] The loss function L is used as the loss function of the model, combined with the fault diagnosis network model guided by causal knowledge, and after initializing the network parameters, the stochastic gradient descent method is used to complete the model training to obtain the pre-trained fault diagnosis model.

[0157] As an example, the loss function is as follows:

[0158]

[0159] in, and They represent the number of labeled samples and the number of unlabeled samples, respectively. represents the true standard of the i-th sample in the labeled sample, represents the recognition result of the i-th sample in the labeled samples, represents the recognition result of the jth sample in the unlabeled sample. The first term in the loss function is the standard cross entropy loss, and the second term is the standard self-information entropy loss. Represents a logarithmic function.

[0160] The causal knowledge-guided power equipment fault diagnosis system of the present invention is based on the causal knowledge-guided power equipment fault diagnosis method mentioned above.

[0161] like Figure 6 As shown, the third object of the embodiment of the present invention is to provide an electronic device, including a memory 701, a processor 702, and a computer program stored in the memory 701 and executable on the processor, wherein the processor implements the power equipment fault diagnosis method based on causal knowledge guidance when executing the computer program. It also includes a communication interface 703 and a bus 704.

[0162] A fourth objective of an embodiment of the present invention is to provide a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned method for diagnosing faults of electric power equipment guided by causal knowledge is implemented.

[0163] A fifth objective of an embodiment of the present invention is to provide a computer program product, wherein the computer program product comprises computer instructions, wherein the computer instructions instruct a computer to execute the above-mentioned method for diagnosing faults of electric power equipment guided by causal knowledge.

[0164] With the continuous development of artificial intelligence and knowledge-data fusion applications, the fault diagnosis technology that integrates equipment maintenance knowledge and deep learning provided by the present invention has broad application prospects, and can effectively assist operation and maintenance personnel to obtain the fault type of power equipment in a timely manner, grasp the development trend of faults, assist in the scheduling of maintenance work, and ensure the safe and reliable operation of substation equipment. When applied, the present invention can be applied to main substation equipment of various voltage levels from 110kV to ultra-high voltage. The developed software modules and algorithm models can be integrated into provincial / municipal substation remote intelligent inspections or related systems in the form of components or micro-applications for promotion and application. At the same time, the project results can be combined with the company's artificial intelligence "two libraries and one platform", combined with business data such as online monitoring, status maintenance, live detection, and robot inspections, to support the forward and reverse inference analysis of defects, hidden dangers, and faults of substation equipment, and improve the automation, digitization, and intelligence level of active operation and maintenance of equipment.

[0165] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0166] The present invention provides computer program instructions that can also be loaded onto a computer or other programmable data processing device, so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0167] The present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, readable storage media, optical storage, etc.) containing computer-usable program codes.

[0168] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0169] Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0170] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for diagnosing power equipment faults based on causal knowledge, characterized in that: include: The real-time monitoring data is input into the corresponding spatial mapping network according to the mode to obtain mapping features; Send the mapping features into the pre-trained fault diagnosis model to obtain the fault diagnosis results; The training process of the pre-trained fault diagnosis model includes: Based on the existing knowledge of power equipment maintenance, the pseudo-causal intervention effects between various physical quantities and fault types are analyzed and mined, and the causal relationship diagram between multimodal physical information and fault phenomena is constructed based on the pseudo-causal intervention effects. Construct a multimodal feature fusion module to normalize multiple physical quantities, perform feature fusion, and perform intervention calculation based on the attention mechanism to obtain intervention features of the encoded feature changes; Constructing a knowledge-guided causal reasoning fault diagnosis model, introducing a causal relationship graph into the model as prior knowledge, guiding the training of the intervention-feature-oriented causal reasoning fault diagnosis model, completing the training and reasoning tasks of the knowledge-guided causal reasoning model based on historical data, and obtaining the pre-trained fault diagnosis model; The construction of the knowledge-guided causal reasoning fault diagnosis model introduces the causal relationship graph as prior knowledge into the model to guide the training of the causal reasoning fault diagnosis model oriented to the intervention feature, including: Based on the cause-effect diagram and intervention characteristics , build a fault diagnosis module guided by causal knowledge; Cause and effect diagram The nodes in the multimodal feature set are extracted and divided into two parts: fault category and measurement parameter. The number of nodes in the fault category is equal to the number of label categories in the multimodal feature set, and the number of nodes in the measurement parameter part is consistent with the number of modes in the multimodal feature set. When extracting the measurement parameter nodes, the nodes of the same mode are fused into one point, and the corresponding edge weights are connected to the fault category nodes through accumulation as new edges to form a new edge weight matrix. ; The edge weight matrix Set as a guidance module based on the attention mechanism Initial weight value of intervention feature It will first be sent to the guidance module and then to the classifier network , and obtain the final fault diagnosis result , ; The loss function L is used as the loss function of the model, combined with the fault diagnosis network model guided by causal knowledge, and after initializing the network parameters, the stochastic gradient descent method is used to complete the model training to obtain the pre-trained fault diagnosis model.

2. The method for diagnosing power equipment faults based on causal knowledge guidance according to claim 1 is characterized in that: Based on the existing knowledge of power equipment maintenance, the pseudo-causal intervention effect between multiple physical quantities and fault types is analyzed and mined, and a causal relationship diagram between multimodal physical information and fault phenomena is constructed based on the pseudo-causal intervention effect, which specifically includes: According to the fault type, measured physical quantity, and measurement cause, the equipment mechanism knowledge is summarized and extracted, a dynamic equipment mechanism knowledge database is created, and the equipment mechanism knowledge database is gradually supplemented and completed; Characterize the equipment mechanism knowledge inside the power equipment in the database, and convert the equipment mechanism knowledge into a mathematical expression for constructing a cause-effect relationship diagram in the form of an indicator function; Arrange the already sorted indicator functions into an indicator function set ,Based on the pseudo causal relationship between multi-modal physical information and fault phenomena, a causal relationship graph between multi-modal physical information and fault phenomena is constructed.

3. The method for diagnosing faults of electric power equipment based on causal knowledge guidance according to claim 2 is characterized in that: The method of characterizing the device mechanism knowledge inside the power equipment in the database and converting the device mechanism knowledge into a mathematical expression for constructing a causal relationship diagram in the form of an indicator function includes: by Indicates the parameter fault type, Indicates related components, Represent the measured physical quantities related to the fault and construct the indicator function of the causal relationship representation: ； in, Represents an indicator function, which is used to determine whether an event has occurred. The value of an event that has been determined to have occurred is 1, and the value of an event that has been determined not to have occurred is 0; Expresses the expectation of an event. To intervene in the situation means to intervene in some way in the current state of things; Indicates when the relevant components Type of parameter failure When the measured physical quantity related to the fault The changes that would occur, indicating the results of the representation, would serve as pseudo-intervention information.

4. The method for diagnosing faults of electric power equipment based on causal knowledge guidance according to claim 2 is characterized in that: The method of arranging the arranged indicator functions into an indicator function set , based on the pseudo causal relationship between multi-modal physical information and fault phenomena, a causal relationship diagram between multi-modal physical information and fault phenomena is constructed. The specific steps are as follows: Determine each node in the causal relationship diagram, including component nodes, fault type nodes, parameter fault type nodes, and parameter change nodes; connect component nodes, parameter fault type nodes, parameter change nodes, and fault type nodes to construct an essential causal relationship diagram , and initialize the final output causal relationship graph , a set of indicator functions that record the intervened variables ; Represents a dataset; Will The set of all connected blocks on , initialize the connected block set of the ancestor edge to be determined for ,in, For collection A connected block in Indicates that the number of connected block parameters is greater than 1, Indicates the existence of a set A connected block in The number of parameters is greater than 1, Representation diagram The variable The set of all ancestor variables of Represents a connected block There are variables in The ancestor variable of When, collection A connected block in There are variables in ancestor variable; Cycle judgment Is it an empty set? If it is an empty set, end the causal relationship diagram The loop updates and jumps out of the loop; if it is not an empty set, the causal relationship diagram is started The cycle update from Randomly select a connected block from , and in Among all the unintervention variables, select the variable with the most undirected edges as the pseudo intervention variable , calculated based on Local causal relationship diagram of ,in, , is the number of local causal graphs, according to the indicator function set Neutralizing variables Correlation indicator function value to local causal relationship graph Update the edges and transform all local causal graphs The public directed edges in are updated to In the updated causal relationship diagram and connected block set , and proceed to the next cycle; After the judgment cycle is completed, a causal relationship diagram between multimodal physical information and fault phenomena is finally obtained.

5. The method for diagnosing faults of electric power equipment based on causal knowledge guidance according to claim 1, characterized in that: The multimodal feature fusion module is constructed to normalize multiple physical quantities, fuse features, and perform intervention calculation processing based on the attention mechanism to obtain intervention features of the encoded feature changes, including: Each modal feature set Input corresponding spatial mapping network , get the feature information vector set ,in, , k is the number of modes, Represents batch normalization operation express The number of features in represents the dimension of a single feature vector after spatial mapping, Indicates the size of the set; according to the feature annotation, select In the figure, the feature vector corresponding to the normal operation of the equipment is calculated by the mean to obtain the mean feature vector of the normal operation of the corresponding mode. ; For each modal feature information vector that has been obtained Perform feature splicing to obtain a fused feature set [ ] and normal operating characteristics [ ]; Constructing an intervention network based on attention mechanism Changes in encoding features, intervention network based on attention mechanism The weights are initialized to ,Will Send to the network Get intervention features of changes in coded features ; Complete dynamic fusion of multimodal physical information.

6. The method for diagnosing faults of electric power equipment based on causal knowledge guidance according to claim 1, characterized in that: The loss function is as follows: in, and They represent the number of labeled samples and the number of unlabeled samples, respectively. represents the true standard of the i-th sample in the labeled sample, represents the recognition result of the i-th sample in the labeled samples, represents the recognition result of the jth sample in the unlabeled sample. The first term in the loss function is the standard cross entropy loss, and the second term is the standard self-information entropy loss. Represents a logarithmic function.

7. A power equipment fault diagnosis system based on causal knowledge guidance, based on the power equipment fault diagnosis method based on causal knowledge guidance according to any one of claims 1 to 6, characterized in that: include: A feature mapping module is used to input the real-time monitoring data into the corresponding spatial mapping network according to the modality to obtain mapping features; A fault diagnosis module is used to input the mapping features into a pre-trained fault diagnosis model to obtain a fault diagnosis result; A pre-training module is used for the training process of the pre-trained fault diagnosis model, and the pre-training module includes: A causal relationship construction unit is used to analyze and mine the pseudo-causal intervention effects between multiple physical quantities and fault types based on the existing knowledge of power equipment maintenance, and to construct a causal relationship diagram between multimodal physical information and fault phenomena based on the pseudo-causal intervention effects; The intervention feature acquisition unit is used to construct a multimodal feature fusion module, normalize multiple physical quantities, perform feature concatenation, and perform intervention calculation processing based on the attention mechanism to obtain the intervention features of the encoded feature changes; The model training unit is used to construct a knowledge-guided causal reasoning fault diagnosis model, introduce the causal relationship diagram into the model as prior knowledge, guide the training of the causal reasoning fault diagnosis model for fusion features, complete the training and reasoning tasks of the knowledge-guided causal reasoning model based on historical data, and obtain the pre-trained fault diagnosis model.

8. An electronic device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the causal knowledge-guided power equipment fault diagnosis method as claimed in any one of claims 1 to 6 when executing the computer program.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for diagnosing faults of electric power equipment based on causal knowledge guidance as claimed in any one of claims 1 to 6 is implemented.

10. A computer program product, comprising computer instructions, characterized in that: The computer instructions instruct the computer to execute the power equipment fault diagnosis method guided by causal knowledge as described in any one of claims 1-6.

Images