Power equipment diagnosis method and system fusing knowledge graph and deep learning

Abstract

The invention discloses an electrical equipment diagnosis method and system fusing a knowledge graph and deep learning, and the method comprises the steps: carrying out the time sequence coding of a dissolved gas concentration data sequence in oil and an oil temperature data sequence in a preset time period through employing a neural network model based on deep learning; the method comprises the following steps of: extracting a time sequence association relationship between the concentration data of the dissolved gas in the oil and the oil temperature data, and further fusing the time sequence characteristics of the concentration data of the dissolved gas in the oil and the time sequence characteristics of the oil temperature data so as to accurately capture an interaction rule of multiple physical quantities under a dynamic working condition of equipment; and then a deep learning diagnosis result is mapped to a knowledge graph topological space, and bidirectional traceability of a fault node and a cause link is realized. Through the mode, the feature representation robustness of the complex coupling fault is improved, the causal closed-loop verification of the diagnosis result and the equipment operation mechanism is realized, and the formation of an intelligent decision-making system with high-precision diagnosis capability and strong interpretation is promoted.

Description

Technical Field

[0001] This application relates to the field of intelligent diagnosis, and more specifically, to a power equipment diagnosis method and system that integrates knowledge graphs and deep learning. Background Art

[0002] Ensuring the safe and stable operation of the power system is crucial for accurately diagnosing the status of key power equipment, especially core hub equipment such as 110 kV oil-immersed power transformers. Traditional diagnostic methods have laid the foundation for this, and among them, dissolved gas analysis (DGA) in oil plays a core role. By regularly detecting the concentrations of characteristic gases such as H 2 , CH 4 , C 2 H 4 , C 2 H 2 in transformer oil and referring to standards such as IEC 60599 and combining map methods such as the three-ratio method and David's triangle for analysis, operation and maintenance personnel can judge whether there are latent faults such as overheating or discharge inside the equipment. In addition to DGA, the traditional diagnostic system also includes temperature monitoring. However, when facing the increasingly complex power grid operation environment and higher reliability requirements, the limitations of these long-term used methods are gradually emerging. For example, diagnostic methods based on fixed thresholds and simplified rules are often affected in accuracy when dealing with situations with fuzzy boundaries or multiple concurrent faults, and it is difficult to capture the subtle evolution characteristics of faults. At the same time, these methods often focus on static or isolated data point analysis and fail to fully explore the rich temporal correlation information and dynamic synergy effects contained in multi-source monitoring data such as DGA and temperature, such as the specific correlation pattern between gas generation rate and temperature change. In addition, the diagnostic process largely depends on experienced experts for comprehensive judgment, which is not only highly subjective and difficult to standardize, but also may not be sensitive enough to weak fault signals in the budding stage, limiting the early warning ability of faults.

[0003] In recent years, the development of deep learning technology has provided new solutions and ideas for the diagnosis of power equipment. However, due to the black-box characteristics of deep learning models, their decision-making processes are opaque and it is difficult to provide intuitive physical explanations. This is a significant application obstacle in the field of high-reliability and high-risk power equipment diagnosis. Operation and maintenance personnel not only need to know "what" the fault is, but also need to understand "why" in order to make correct decisions.

[0004] Therefore, an optimized power equipment diagnosis solution is expected. Summary of the Invention

[0005] To solve the above technical problems, this application is proposed. Embodiments of this application provide a power equipment diagnosis method and system that integrates a knowledge graph and deep learning. By using a neural network model based on deep learning to perform temporal encoding on the dissolved gas concentration data sequence and oil temperature data sequence within a predetermined time period, the temporal correlation relationship in the dissolved gas concentration data and oil temperature data is extracted. Further, the temporal features of the dissolved gas concentration data and the temporal features of the oil temperature data are fused to accurately capture the interaction law of multiple physical quantities under the dynamic working conditions of the equipment. Then, the deep learning diagnosis result is mapped to the topological space of the knowledge graph to realize the bidirectional traceability of the fault node and the cause link. In this way, the robustness of the feature representation of complex coupling faults is improved, the causal closed-loop verification between the diagnosis result and the equipment operation mechanism is realized, and the formation of an intelligent decision-making system with both high-precision diagnosis ability and strong interpretability is promoted.

[0006] According to one aspect of this application, a power equipment diagnosis method that integrates a knowledge graph and deep learning is provided, which includes:

[0007] Obtain the equipment information of the power equipment to be diagnosed, where the power equipment to be diagnosed is a 110 kV oil-immersed power transformer;

[0008] Based on the equipment information, extract the dissolved gas concentration data sequence and oil temperature data sequence within a predetermined time period from the background database;

[0009] Perform temporal encoding on the dissolved gas concentration data sequence and the oil temperature data sequence respectively to obtain the dissolved gas concentration temporal correlation hidden feature vector and the oil temperature temporal correlation hidden feature vector;

[0010] Fuse the dissolved gas concentration temporal correlation hidden feature vector and the oil temperature temporal correlation hidden feature vector to obtain a dissolved gas concentration - oil temperature temporal fine-grained collaborative encoding matrix;

[0011] Based on the dissolved gas concentration - oil temperature temporal fine-grained collaborative encoding matrix, obtain the fault diagnosis result;

[0012] Based on the node position corresponding to the fault diagnosis result in the power equipment knowledge graph, obtain the fault cause explanation text.

[0013] According to another aspect of this application, a power equipment diagnosis system that integrates a knowledge graph and deep learning is provided, which includes:

[0014] An equipment information acquisition module, used to acquire equipment information;

[0015] A data extraction module, used to extract the dissolved gas concentration data sequence and the oil temperature data sequence within a predetermined time period from the background database based on the equipment information;

[0016] A time series encoding module, which is used to perform time series encoding on the dissolved gas concentration data series and the oil temperature data series respectively to obtain a dissolved gas concentration time series correlation hidden feature vector and an oil temperature time series correlation hidden feature vector;

[0017] A time series fusion module, which is used to fuse the dissolved gas concentration time series correlation hidden feature vector and the oil temperature time series correlation hidden feature vector to obtain a dissolved gas concentration - oil temperature time series fine - grained collaborative encoding matrix;

[0018] A fault diagnosis module, which is used to obtain a fault diagnosis result based on the dissolved gas concentration - oil temperature time series fine - grained collaborative encoding matrix;

[0019] A fault text explanation module, which is used to obtain a fault cause explanation text based on the node position corresponding to the fault diagnosis result in the power equipment knowledge graph.

[0020] Compared with the prior art, a power equipment diagnosis method and system integrating a knowledge graph and deep learning provided by the present application performs time series encoding on the dissolved gas concentration data series and the oil temperature data series within a predetermined time period by using a neural network model based on deep learning to extract the time series correlation relationship in the dissolved gas concentration data and the oil temperature data. Further, it fuses the time series features of the dissolved gas concentration data and the time series features of the oil temperature data to accurately capture the interaction law of multiple physical quantities under the dynamic working conditions of the equipment. Then, it maps the deep learning diagnosis result to the topological space of the knowledge graph to realize the two - way traceability of the fault node and the cause link. In this way, the robustness of the feature representation of complex coupled faults is improved, the causal closed - loop verification between the diagnosis result and the equipment operation mechanism is realized, and the formation of an intelligent decision - making system with both high - precision diagnosis ability and strong interpretability is promoted. Brief Description of the Drawings

[0021] By describing the embodiments of the present application in more detail in conjunction with the drawings, the above - mentioned and other objects, features, and advantages of the present application will become more obvious. The drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. They are used to explain the present application together with the embodiments of the present application and do not constitute a limitation to the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0022] Figure 1 It is a flowchart of a power equipment diagnosis method integrating a knowledge graph and deep learning according to an embodiment of the present application;

[0023] Figure 2 It is a schematic diagram of data flow of a power equipment diagnosis method integrating a knowledge graph and deep learning according to an embodiment of the present application;

[0024] Figure 3 It is a block diagram of a power equipment diagnosis system that integrates a knowledge graph and deep learning according to an embodiment of the present application. Detailed implementation manners

[0025] Next, exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all embodiments of the present application. It should be understood that the present application is not limited by the exemplary embodiments described herein.

[0026] As shown in the present application and the claims, unless the context clearly indicates otherwise, words such as "a", "an", "one" and / or "the" are not specifically singular and may also include plural. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of the clearly identified steps and elements, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0027] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, however, any number of different modules can be used and run on the user terminal and / or the server. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0028] In the technical solution of the present application, a power equipment diagnosis method that integrates a knowledge graph and deep learning is proposed.

[0029] Next, exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all embodiments of the present application. It should be understood that the present application is not limited by the exemplary embodiments described herein.

[0030] In the technical solution of the present application, a power equipment diagnosis method that integrates a knowledge graph and deep learning is proposed. Figure 1 It is a flowchart of a power equipment diagnosis method that integrates a knowledge graph and deep learning according to an embodiment of the present application. Figure 2 It is a schematic diagram of data flow of a power equipment diagnosis method that integrates a knowledge graph and deep learning according to an embodiment of the present application. As Figure 1 and Figure 2As shown, a power equipment diagnosis method integrating a knowledge graph and deep learning according to an embodiment of the present application includes the steps of: S1, obtaining device information of the power equipment to be diagnosed, where the power equipment to be diagnosed is a 110 kV oil-immersed power transformer; S2, based on the device information, extracting a sequence of oil-dissolved gas concentration data and a sequence of oil temperature data within a predetermined time period from the background database; S3, respectively performing time series encoding on the sequence of oil-dissolved gas concentration data and the sequence of oil temperature data to obtain an oil-dissolved gas concentration time series correlation hidden feature vector and an oil temperature time series correlation hidden feature vector; S4, fusing the oil-dissolved gas concentration time series correlation hidden feature vector and the oil temperature time series correlation hidden feature vector to obtain an oil-dissolved gas concentration - oil temperature time series fine-grained collaborative encoding matrix; S5, based on the oil-dissolved gas concentration - oil temperature time series fine-grained collaborative encoding matrix, obtaining a fault diagnosis result; S6, based on the node position corresponding to the fault diagnosis result in the power equipment knowledge graph, obtaining a fault cause explanation text.

[0031] Specifically, in S1, obtaining device information of the power equipment to be diagnosed, where the power equipment to be diagnosed is a 110 kV oil-immersed power transformer. Among them, device information refers to key data that can uniquely identify the identity of the target power equipment and its operating background, usually including but not limited to static and semi-static attributes such as device number, device model, device type (such as transformer, circuit breaker, etc.), the substation or line to which it belongs, and rated voltage level. In the technical solution of the present application, obtaining device information can enable the system to accurately lock the target device entity that needs to be diagnosed and analyzed, providing an accurate index and range limit for extracting relevant information from a large amount of monitoring data in the subsequent process. In this way, the system can obtain the unique identifier and key attributes of the target device, ensuring that the sequence of oil-dissolved gas concentration data and the sequence of oil temperature data of this specific device within a predetermined time period can be accurately and error-free extracted from the background database.

[0032] Specifically, in S2, based on the device information, the concentration data sequence of dissolved gases in oil and the oil temperature data sequence within a predetermined time period are extracted from the background database. It should be understood that the concentration data of dissolved gases in oil (such as hydrogen, methane, ethylene, acetylene, carbon monoxide, etc.) characterize the dynamic changes of gases (such as hydrogen, methane, acetylene, etc.) decomposed due to faults such as partial discharge, overheating, or arc in the insulating oil. The concentration combination pattern can clearly distinguish the discharge type and the degree of insulation deterioration; while the oil temperature data reflects the thermodynamic state of the device, and abnormal temperature rise is often directly related to faults such as winding overheating, cooling system failure, or load imbalance. The core reason for performing the above data extraction is that it is difficult for traditional single-parameter analysis to capture the endogenous relationship between the two key physical quantities. For example, a high-temperature environment may accelerate the precipitation of gases in oil, and abnormal gas accumulation may in turn exacerbate local temperature rise. By extracting the concentration data sequence of dissolved gases in oil and the oil temperature data sequence within a predetermined time period from the background database, the system can construct a multi-scale feature coupling relationship under dynamic working conditions, using both the gas concentration data to reveal the specificity of fault types and combining the oil temperature data to lock the thermodynamic abnormal boundary, thus breaking through the limitations of traditional static threshold criteria. By extracting data sequences covering a certain period, rich inputs are provided for subsequent time-series neural networks such as LSTM, enabling the model to learn the non-linear conduction and dynamic evolution laws between various features.

[0033] Specifically, in step S3, time series encoding is respectively performed on the dissolved gas concentration data series in oil and the oil temperature data series to obtain a dissolved gas concentration time series correlation hidden feature vector in oil and an oil temperature time series correlation hidden feature vector in oil. That is, in the embodiments of the present application, first, the dissolved gas concentration data series in oil is passed through a gas concentration time series encoder based on an LSTM model to obtain a dissolved gas concentration time series correlation hidden feature vector in oil. It should be understood that the dissolved gas concentration data in oil, as an important indicator reflecting the internal insulation condition and potential faults of power equipment, has obvious non-linear and non-stationary characteristics in its change process, and is often affected by various physical mechanisms and shows a complex evolution trend. Traditional deep learning models mostly adopt an end-to-end pattern matching, and it is difficult to accurately capture these hidden time dependencies and their connections with fault mechanisms. LSTM (Long Short-Term Memory network) is a special type of recurrent neural network (RNN), which solves the problems of gradient disappearance and gradient explosion existing in traditional RNNs when processing long time series. By introducing a gating mechanism, LSTM can effectively capture long-distance dependencies in time series, retain key historical information, and filter out irrelevant noise at the same time, so as to achieve long-term memory and dynamic pattern learning of complex time series data. Using the LSTM model to encode the time series of dissolved gas concentration in oil can systematically learn these complex change rules and extract a dissolved gas concentration time series correlation hidden feature vector representing the operating state of the equipment and the development trend of faults. The generated dissolved gas concentration time series correlation hidden feature vector not only contains the information of the gas concentration evolving over time, but also contains the potential coupling relationship between different gases and its dynamic manifestation of the fault indication significance. By constructing a gas concentration time series encoder based on the LSTM model, not only the accurate capture of the dynamic change rules of the dissolved gas concentration data in oil is realized, but also a high-quality data carrier rich in physical mechanism information is provided for fusing knowledge graphs, thus greatly improving the ability of the intelligent diagnosis system of power equipment to accurately identify faults and explain the reasons under complex working conditions.

[0034] Furthermore, the oil temperature data sequence is encoded in time series through the oil temperature data time series encoder based on the LSTM model to obtain the oil temperature time series correlation hidden feature vector. It should be understood that as an important physical quantity reflecting the operating state of power equipment, the oil temperature data not only reflects the equipment load, heat dissipation condition and environmental impact over time, but also directly affects the generation rate and distribution characteristics of dissolved gases in insulating oil. Due to the complex and variable operating environment of power equipment, the oil temperature time series data usually shows the characteristics of non-linearity, multiple fluctuations and being affected by multiple factor couplings. It is difficult to reveal its internal law simply relying on static analysis or simple statistics. In the technical solution of this application, by adopting the time series encoder based on LSTM, the time dependence and potential patterns in the oil temperature data can be deeply mined, and its long-term trend and sudden abnormal signals can be effectively captured, providing rich and accurate dynamic feature expressions for subsequent diagnosis. By encoding the oil temperature data in time series based on the LSTM model, it not only significantly enhances the system's perception ability of the change law of the equipment thermal state under dynamic conditions, but also effectively supports the subsequent fine-grained collaborative analysis based on the global interaction network, thus improving the accuracy and stability of the fault diagnosis results. In addition, the generated oil temperature time series correlation hidden feature vector provides high-quality data input for combining with the knowledge graph, enabling the diagnosis system to better combine physical mechanisms and historical experience to realize intelligent and interpretable power equipment fault identification and cause inference.

[0035] Specifically, in step S4, the dissolved gas concentration in oil time series correlation hidden feature vector and the oil temperature time series correlation hidden feature vector are fused to obtain the dissolved gas concentration in oil - oil temperature time series fine-grained collaborative coding matrix. It should be understood that since the fault mechanism of power equipment often involves the coupling effect of multiple physical fields, for example, there is a complex non-linear conduction relationship between the change of dissolved gas concentration in oil and the oil temperature fluctuation. Although the hidden feature vector extracted by the traditional time series encoder through LSTM can capture the time series dynamic characteristics of single variables, it does not explicitly model the endogenous association between features. Specifically, although the dissolved gas concentration data in oil and the oil temperature data are collected synchronously in the time dimension, there are time delay and space diffusion effects in their physical conduction paths. The simple splicing or weighted fusion of single feature vectors is difficult to reveal the co-evolution pattern of cross-domain parameters in the non-linear space, and it is easy to cause the semantic information of key fault features to drift or be confused in the vector space. Therefore, in the technical solution of this application, the dissolved gas concentration in oil time series correlation hidden feature vector and the oil temperature time series correlation hidden feature vector are fused to obtain the dissolved gas concentration in oil - oil temperature time series fine-grained collaborative coding matrix.

[0036] In this process, first, the feature principal component analysis eliminates the collinear interference between the gas concentration in oil and the oil temperature feature through orthogonal decoupling, maps the original high-dimensional features to the low-dimensional principal component space, and provides physically interpretable basis vectors for subsequent fine-grained correlation modeling. Then, the principal component kernel correlation coding network further maps the decoupled principal component pairs to a high-dimensional non-linear space to learn the potential causal temporal patterns between gas evolution and temperature rise (such as the cross-cycle correlation between the slow increase in gas concentration and the lagging temperature rise at the initial stage of discharge). Further, by constructing a performance operator correlation topology matrix, the collaborative contribution degree of the principal component pairs in fault prediction is quantified. For example, the high correlation weight of the hydrogen concentration principal component and the oil temperature principal component in the arc fault. Finally, the GCN model aggregates multi-hop neighborhood information to generate a fine-grained collaborative coding matrix. In particular, in the technical solution of this application, the dimension collapse problem of traditional static feature fusion is overcome. The disordered distribution of the feature principal components in the topological space is dynamically corrected through the cross-section function matrix, and the interference of the random potential field on the correlation topology is eliminated, ensuring that the cross-domain interaction law between the gas concentration and the oil temperature (such as high temperature accelerating gas diffusion and gas aggregation causing local overheating) is structurally characterized in the coding matrix. The generated fine-grained collaborative coding matrix of the dissolved gas concentration-oil temperature time series in oil can capture the temporal lag effect of the gas concentration mutation and the slow change of the oil temperature, and through the knowledge graph interpretable node inversion mechanism, maps the high-weight interaction patterns in the matrix to the actual physical fault link. In this way, the thermodynamic conduction characteristics of the oil temperature parameter and the chemical equilibrium law of gas solubility can form a dynamic interaction path in the topological space of the graph convolutional network, so as to capture the critical state characteristics of the multi-parameter coupling effect in the equipment fault evolution process.

[0037] Specifically, first, perform feature principal component analysis on the time-series correlation implicit feature vectors of the dissolved gas concentration in oil and the time-series correlation implicit feature vectors of the oil temperature to obtain a set of time-series feature principal component coding vectors of the dissolved gas concentration in oil and a set of time-series feature principal component coding vectors of the oil temperature. Since the time-series data of the dissolved gas concentration in oil and the oil temperature have strong coupling in the physical conduction mechanism. For example, an increase in the oil temperature will accelerate the pyrolysis reaction of the insulating oil, resulting in a non-linear increase in the concentrations of gases such as methane and ethane. Although the implicit feature vectors extracted by the traditional LSTM encoder can characterize the dynamic characteristics of single-variable time series, the original features of the two are prone to redundancy or false correlation in the vector space. For example, the periodic fluctuation of the oil temperature may mask the true fault signal of the sudden change in the gas concentration, or the fluctuations of the two in a specific frequency band may generate pseudo-synchronization due to changes in the equipment operating conditions, resulting in a deviation between the linear correlation of the feature vectors and the actual physical correlation, thereby interfering with the robustness of the cross-domain feature interaction modeling. Therefore, in the technical solution of this application, perform feature principal component analysis on the time-series correlation implicit feature vectors of the dissolved gas concentration in oil and the time-series correlation implicit feature vectors of the oil temperature to obtain a set of time-series feature principal component coding vectors of the dissolved gas concentration in oil and a set of time-series feature principal component coding vectors of the oil temperature.

[0038] Specifically, the time-series implicit feature vectors of the dissolved gas concentration in oil may contain independent fault modes that are irrelevant to the oil temperature (such as abnormal hydrogen caused by partial discharge), and the oil temperature feature vectors may also contain thermodynamic noise that is irrelevant to the gas concentration (such as ambient temperature disturbance). Through principal component analysis, the two types of feature vectors can be respectively mapped to the low-dimensional space spanned by the orthogonal basis to eliminate the interference of multicollinearity on cross-domain correlation modeling. In this process, the principal component screening strategy needs to be optimized in combination with the fault mechanism of power equipment: for example, for the Arrhenius equation relationship between the oil temperature and the gas solubility, retain the principal components that can reflect the change in activation energy; for the correlation between the gas diffusion rate and the oil temperature gradient, screen the principal components that characterize the heat convection effect, so as to retain the physical coupling modes strongly related to the fault evolution during the dimensionality reduction process.

[0039] In a specific example of this application, perform feature principal component analysis on the time-series correlation implicit feature vectors of the dissolved gas concentration in oil and the time-series correlation implicit feature vectors of the oil temperature with the following principal component analysis formula to obtain a set of time-series feature principal component coding vectors of the dissolved gas concentration in oil and a set of time-series feature principal component coding vectors of the oil temperature; where the principal component analysis formula is:

[0040]

[0041]

[0042] Where, is the time-series correlation implicit feature vector of the dissolved gas concentration in oil, is the implicit feature vector related to the oil temperature time series, is the principal component analysis of features, is the set of principal component encoding vectors of the time series features of the dissolved gas concentration in oil, is the set of principal component encoding vectors of the time series features of the oil temperature, are the 1st, 2nd, th, and th principal component encoding vectors of the time series features of the dissolved gas concentration in oil in the set of principal component encoding vectors of the time series features of the dissolved gas concentration in oil respectively, are the 1st, 2nd, th, and th principal component encoding vectors of the time series features of the oil temperature in the set of principal component encoding vectors of the time series features of the oil temperature respectively, is the dissolved gas concentration in oil - oil temperature time series feature diagonal matrix, and are respectively and corresponding eigenvalues, is the dissolved gas concentration in oil - oil temperature time series feature diagonal matrix, and are respectively and corresponding eigenvalues.

[0043] Next, the set of principal component encoding vectors of the time series features of the dissolved gas concentration in oil and the set of principal component encoding vectors of the time series features of the oil temperature, and for each group of corresponding principal component encoding vectors of the time series features of the dissolved gas concentration in oil and the principal component encoding vectors of the time series features of the oil temperature in the two sets, they are respectively input into the dissolved gas concentration - oil temperature principal component kernel correlation encoding network to obtain the set of kernel correlation encoding vectors between the principal components of the time series features of the dissolved gas concentration in oil - oil temperature. It should be understood that although the principal component analysis of features has eliminated the multicollinearity in the original time series implicit feature vectors, the interaction patterns between the oil temperature principal components (such as the low - variance components reflecting the thermodynamic equilibrium state) and the gas concentration principal components (such as the orthogonal basis vectors characterizing the gas diffusion rate) in the low - dimensional orthogonal space often exhibit non - linear separability. Therefore, in the technical solution of this application, the set of principal component encoding vectors of the time series features of the dissolved gas concentration in oil and the set of principal component encoding vectors of the time series features of the oil temperature, and for each group of corresponding principal component encoding vectors of the time series features of the dissolved gas concentration in oil and the principal component encoding vectors of the time series features of the oil temperature in the two sets, they are respectively input into the dissolved gas concentration - oil temperature principal component kernel correlation encoding network to obtain the set of kernel correlation encoding vectors between the principal components of the time series features of the dissolved gas concentration in oil - oil temperature.

[0044] Specifically, the heat conduction time constant contained in the oil temperature principal component encoding vector and the gas solubility coefficient in the gas concentration principal component encoding vector are reparameterized into the tensor product form in the high-dimensional kernel space during the forward propagation process of the deep neural network. For example, for the slow-varying characteristics of the oil temperature principal component that characterizes the heat dissipation efficiency and the sudden change components reflecting the gas accumulation rate in the gas concentration principal component, the network establishes the time-varying correlation strength between the two through an adaptive weight mechanism to simulate the hysteresis effect of thermal inertia on the gas diffusion rate in the actual fault evolution. The generated kernel correlation encoding vector between the principal components of the dissolved gas concentration-oil temperature time series characteristics provides a physical and clear input basis for the construction of the subsequent performance operator correlation topology matrix, so that the graph convolutional network can mine fault-sensitive paths based on the geometric structure in the high-dimensional kernel space (such as the tangent space overlap area of ​​the oil temperature principal component manifold and the gas concentration principal component manifold), significantly improving the recognition accuracy of multi-parameter coupled faults (such as overheating and partial discharge concurrent faults).

[0045] In a specific example of the present application, each corresponding set of principal component coding vectors of time series characteristics of dissolved gas concentration in oil and the set of principal component coding vectors of time series characteristics of oil temperature are respectively input into the dissolved gas concentration-oil temperature principal component kernel correlation coding network to obtain a set of kernel correlation coding vectors between principal components of dissolved gas concentration in oil and oil temperature time series characteristics; wherein, the correlation coding formula is:

[0046]

[0047] in, represents the one-norm of a vector, and They represent trainable weighted hyperparameters, is the first kernel correlation coding vector in the set of principal components of the time series characteristics of dissolved gas concentration and oil temperature in oil The kernel correlation encoding vector between the principal components of the time series characteristics of dissolved gas concentration and oil temperature in oil.

[0048] Subsequently, calculate the oil dissolved gas concentration - oil temperature feature performance operator between any two kernel - associated encoding vectors of the oil dissolved gas concentration - oil temperature time - series feature principal components in the set of kernel - associated encoding vectors of the oil dissolved gas concentration - oil temperature time - series features to obtain the oil dissolved gas concentration - oil temperature feature performance operator - associated topological matrix. It should be understood that although the principal component kernel - associated encoding network has established an interactive representation of the oil temperature and the principal components of the gas concentration in the high - dimensional kernel space, the multi - physical - field coupling faults of equipment such as transformers (such as the co - evolution of oil temperature gradient changes and sudden changes in acetylene gas concentration caused by winding overheating) often involve dynamically changing association strengths and action directions. For example, in the case of partial discharge faults in oil - immersed bushings, there is a time - varying delay effect between abnormal oil temperature fluctuations and transient peaks in hydrogen concentration, and static metrics such as conventional cosine similarity or Euclidean distance cannot capture such dynamic association patterns, resulting in the difficulty for the graph convolutional network to construct a physically interpretable topological structure. Therefore, in the technical solution of this application, calculate the oil dissolved gas concentration - oil temperature feature performance operator between any two kernel - associated encoding vectors of the oil dissolved gas concentration - oil temperature time - series feature principal components in the set of kernel - associated encoding vectors of the oil dissolved gas concentration - oil temperature time - series features to obtain the oil dissolved gas concentration - oil temperature feature performance operator - associated topological matrix.

[0049] By calculating the oil dissolved gas concentration - oil temperature feature performance operator, for example, constructing an association strength metric based on the joint probability density function of the gas diffusion kinetic equation and the heat conduction equation, it is possible to quantify the causal contribution weights of the two types of principal components in the fault evolution, so as to transform the physically - mechanism - driven association rules (such as the strong positive correlation between the acetylene concentration principal component and the oil temperature gradient principal component in the arc fault) into computable topological matrix element values, while suppressing the interference of non - fault - related principal component pairs (such as the random fluctuation association between the methane concentration principal component and the steady - state oil temperature principal component) on the model. The generated oil dissolved gas concentration - oil temperature feature performance operator - associated topological matrix can not only characterize the explicit statistical correlation between the principal components, but also strengthen the modeling of implicit causal relationships through physical constraints (such as the theoretical relationship between the gas precipitation rate and the temperature rise rate), enabling the graph convolutional neural network to preferentially propagate high - weight association paths that match the fault mechanism when aggregating neighborhood information.

[0050] In a specific example of this application, calculate the oil dissolved gas concentration - oil temperature feature performance operator between any two kernel - associated encoding vectors of the oil dissolved gas concentration - oil temperature time - series feature principal components in the set of kernel - associated encoding vectors of the oil dissolved gas concentration - oil temperature time - series features according to the following calculation formula to obtain the oil dissolved gas concentration - oil temperature feature performance operator - associated topological matrix; where the calculation formula is:

[0051]

[0052]

[0053] Among them, and are respectively the th and the th eigenvalues at the th position in the kernel correlation coding vectors between the principal components of the dissolved gas concentration - oil temperature time - series characteristics, is the number of eigenvalues in the kernel correlation coding vector between the principal components of the dissolved gas concentration - oil temperature time - series characteristics, is and the dissolved gas concentration - oil temperature characteristic performance operator between them, is the associated topological matrix of the dissolved gas concentration - oil temperature characteristic performance operator.

[0054] Then, perform disorder optimization of the kernel space distribution of the dissolved gas concentration - oil temperature time - series characteristic components for each kernel correlation coding vector between the principal components of the dissolved gas concentration - oil temperature time - series in the set of kernel correlation coding vectors between the principal components of the dissolved gas concentration - oil temperature time - series to obtain a set of optimized kernel correlation coding vectors between the principal components of the dissolved gas concentration - oil temperature time - series. It should be understood that the kernel correlation coding vectors between the principal components of the dissolved gas concentration - oil temperature time - series may show disorder in their distribution in the topological space due to random potential field interference (such as sensor noise or model initialization deviation). For example, the true physical association between the principal component of acetylene concentration caused by partial discharge and the principal component of oil temperature gradient may be masked by random noise, resulting in an abnormal increase in the redundant connection weights unrelated to faults in the associated topological matrix. In a preferred example of the present application, perform disorder optimization of the kernel space distribution of the dissolved gas concentration - oil temperature time - series characteristic components for each kernel correlation coding vector between the principal components of the dissolved gas concentration - oil temperature time - series in the set of kernel correlation coding vectors between the principal components of the dissolved gas concentration - oil temperature time - series to obtain a set of optimized kernel correlation coding vectors between the principal components of the dissolved gas concentration - oil temperature time - series. Here, the random initialization of the dissolved gas concentration - oil temperature time - series cross - section function matrix will introduce unreasonable spatial mapping relationships, causing the kernel correlation coding vectors between the principal components of the dissolved gas concentration - oil temperature time - series to deviate from the actual physical mechanism (such as the coupling law of gas diffusion and heat conduction), and further making the neighborhood information aggregated by the graph convolutional neural network contain a large number of noise paths.

[0055] In the technical solution of this application, by constructing a compactification matrix of the dissolved gas concentration - oil temperature time - series feature cross - section and performing Gaussian correlation iteration with the dissolved gas concentration - oil temperature feature performance operator correlation topology matrix, the initial dissolved gas concentration - oil temperature time - series feature cross - section function matrix is forced to gradually eliminate the random correlation patterns unrelated to the equipment operation mechanism during the optimization process. By using the mechanism that the Gaussian correlation coefficient under variance constraint approaches zero, the disordered distribution among the principal components is reconstructed into a topological space structure that conforms to the fault evolution law of power equipment. For example, it suppresses the abnormal principal component shift caused by the transient drift of the oil temperature sensor, and at the same time strengthens the high - weight correlation between the acetylene principal component and the oil temperature mutation component in the arc fault. The set of the inner - core correlation coding vectors between the optimized dissolved gas concentration - oil temperature time - series feature principal components can accurately represent the endogenous physical correlation between the gas concentration and the oil temperature principal components (such as the low correlation between the hydrogen slow - release principal component and the oil temperature steady - state principal component, and the high synergy between the acetylene mutation component and the oil temperature gradient principal component), enabling the graph convolutional neural network to preferentially aggregate feature information along the real fault link during message passing, and finally realizing the interpretable modeling and stability improvement of the fault diagnosis model for the cross - parameter time - series coupling law.

[0056] As described above, the dissolved gas concentration - oil temperature feature performance operator correlation topology matrix As a topological form space, each inner - core correlation coding vector between the dissolved gas concentration - oil temperature time - series feature principal components will serve as the master - slave in the space and obey the space distribution form, that is , and considering the dimension break between the inner - core correlation coding vector between the dissolved gas concentration - oil temperature time - series feature principal components and the dissolved gas concentration - oil temperature feature performance operator correlation topology matrix , it is necessary to construct a standard transition gauge field from the initial dissolved gas concentration - oil temperature time - series feature cross - section function matrix as the space standard transition gauge field.

[0057] On the other hand, it is also necessary to correct the problem of the disordered space distribution caused by the random potential field in the initial dissolved gas concentration - oil temperature time - series feature cross - section function matrix . First, multiply each inner - core correlation coding vector between the dissolved gas concentration - oil temperature time - series feature principal components with the corresponding initial dissolved gas concentration - oil temperature time - series feature cross - section function matrix to obtain the dissolved gas concentration - oil temperature time - series feature cross - section compactification vector , and then the dissolved gas concentration - oil temperature time - series feature cross - section compactification vectors corresponding to each inner - core correlation coding vector between the dissolved gas concentration - oil temperature time - series feature principal components ​ Two-dimensional splicing to obtain the compactification matrix of the time-series feature cross-section of dissolved gas concentration in oil - oil temperature :

[0058]

[0059]

[0060] Among them, is the th kernel correlation coding vector between the principal components of the time-series features of dissolved gas concentration in oil - oil temperature in the set of kernel correlation coding vectors between the principal components of the time-series features of dissolved gas concentration in oil - oil temperature, is the initial time-series feature cross-section function matrix of dissolved gas concentration in oil - oil temperature, is matrix multiplication, is the th compactification vector of the time-series feature cross-section of dissolved gas concentration in oil - oil temperature in the sequence of compactification vectors of the time-series feature cross-section of dissolved gas concentration in oil - oil temperature, is two-dimensional splicing processing, is the compactification matrix of the time-series feature cross-section of dissolved gas concentration in oil - oil temperature.

[0061] In this way, the Gaussian correlation coefficient between the compactification matrix of the time-series feature cross-section of dissolved gas concentration in oil - oil temperature and the associated topological matrix of the performance operator of the dissolved gas concentration in oil - oil temperature feature can be calculated, and the initial time-series feature cross-section function matrix of dissolved gas concentration in oil - oil temperature can be iterated by making the Gaussian correlation coefficient tend to zero to obtain the optimized time-series feature cross-section function matrix of dissolved gas concentration in oil - oil temperature:

[0062]

[0063] Among them, is subtraction by position, is the F-norm of the matrix, is and is the variance of the set composed of all matrix values of is the value of the natural exponential function with the natural constant e as the base.

[0064] Thus, by further optimizing the time-series feature cross-section function matrix of dissolved gas concentration in oil - oil temperature, the kernel correlation coding vector between the principal components of the time-series features of dissolved gas concentration in oil - oil temperature is optimized:

[0065]

[0066] Among them, is the optimized function matrix of the time - series feature cross - section of dissolved gas concentration in oil - oil temperature, is the optimized kernel - associated coding vector between the principal components of the time - series features of dissolved gas concentration in oil - oil temperature corresponding to the kernel - associated coding vector between the principal components of the time - series features of the nth dissolved gas concentration in oil - oil temperature.

[0067] During the calculation process of the graph convolutional neural network model, it can solve the problem of the disordered spatial distribution brought by the kernel - associated coding vectors between the principal components of the time - series features of dissolved gas concentration in oil - oil temperature under the representation of the random potential field to the topological form space of the topological matrix of the feature performance operator of dissolved gas concentration in oil - oil temperature, thereby improving the calculation results of the graph convolutional neural network model.

[0068] Subsequently, the set of optimized kernel - associated coding vectors between the principal components of the time - series features of dissolved gas concentration - oil temperature and the topological matrix of the feature performance operator of dissolved gas concentration in oil - oil temperature are input into the graph convolutional neural network model to obtain the fine - grained collaborative coding matrix of the time - series of dissolved gas concentration in oil - oil temperature. It should be understood that although the set of optimized kernel - associated coding vectors between the principal components of the time - series features of dissolved gas concentration - oil temperature after the optimization of the disordered distribution has eliminated the interference of the random potential field, it is still necessary to capture the global interaction pattern of cross - domain principal components through graph structure learning. Therefore, in the technical solution of this application, the set of optimized kernel - associated coding vectors between the principal components of the time - series features of dissolved gas concentration - oil temperature and the topological matrix of the feature performance operator of dissolved gas concentration in oil - oil temperature are input into the graph convolutional neural network model to obtain the fine - grained collaborative coding matrix of the time - series of dissolved gas concentration in oil - oil temperature. In this process, the optimized kernel - associated coding vectors between the principal components of the time - series features of dissolved gas concentration - oil temperature are used as node features, and the topological matrix of the feature performance operator of dissolved gas concentration in oil - oil temperature is used as the adjacency matrix, jointly constituting the graph - structured data reflecting the cross - domain interaction of oil temperature - gas concentration. It is worth mentioning that through the multi - layer message - passing mechanism of GCN, the thermodynamic conduction delay characteristics (such as the lag effect of the oil temperature gradient on the gas diffusion rate) in the oil temperature principal component nodes and the chemical reaction equilibrium constraints (such as the dynamic change of the methane - ethane concentration ratio) in the gas concentration principal component nodes are multi - hop aggregated.

[0069] In a specific example of this application, the set of kernel correlation coding vectors between the optimized dissolved gas concentration-oil temperature time series feature principal components and the oil dissolved gas concentration-oil temperature feature performance operator correlation topology matrix are input into a graph convolutional neural network model according to the following formula to obtain an oil dissolved gas concentration-oil temperature time series fine-grained collaborative coding matrix; where the formula is:

[0070]

[0071] Among them, represents the graph convolutional neural network model, is the oil dissolved gas concentration-oil temperature time series fine-grained collaborative coding matrix.

[0072] Specifically, based on the oil dissolved gas concentration-oil temperature time series fine-grained collaborative coding matrix, a fault diagnosis result is obtained. Specifically, in the technical solution of this application, the oil dissolved gas concentration-oil temperature time series fine-grained collaborative coding matrix is passed through a decoder-based fault diagnosis model to obtain a fault diagnosis result, where the fault diagnosis result is used to represent the fault diagnosis score of the device. It should be understood that although the collaborative coding matrix extracted by the graph convolutional network already contains high-order interaction features of cross-domain parameters, its high-dimensional abstract representation is difficult to directly map to specific fault categories. Due to the lack of a structured understanding of the physical mechanism, traditional classifiers are prone to misjudging such complex patterns as single overheating or discharge faults. Therefore, in the technical solution of this application, the oil dissolved gas concentration-oil temperature time series fine-grained collaborative coding matrix is passed through a decoder-based fault diagnosis model to obtain a fault diagnosis result. By introducing a decoder, a mapping between known features and potential fault manifestation forms can be realized, enhancing the model's understanding ability of complex, multi-parameter coupling fault scenarios. In addition, the decoder can also provide more interpretable and traceable results for subsequent diagnosis by gradually generating or reconstructing the target output. In this way, the system can better capture the complex interaction relationship between the oil dissolved gas concentration and the oil temperature and other multi-parameters, thereby improving the understanding ability of the key features after the fusion of multi-source heterogeneous data. Secondly, by gradually restoring the clear fault category or state through the decoder, not only the sensitivity of the model to abnormal situations is enhanced, but also a solid foundation is provided for the cause tracing in the subsequent knowledge graph. This method ultimately realizes efficient, reliable and highly interpretable intelligent diagnosis of power equipment, provides a scientific basis for maintenance decisions, helps with early warning, reduces downtime, and improves the overall operation safety level.

[0073] Specifically, in step S6, based on the node position corresponding to the fault diagnosis result in the power equipment knowledge graph, an explanation text of the fault cause is obtained. Specifically, in the technical solution of the present application, the node corresponding to the fault diagnosis result is located in the power equipment knowledge graph, and the cause is searched backward from the node in the power equipment knowledge graph to obtain the explanation text of the fault cause. It should be understood that the traditional model can only provide a fault score value, but cannot reveal the multi-level conduction link of the fault cause (for example, partial discharge may be jointly caused by multiple factors such as insulation oil aging, humidity penetration, or voltage fluctuation). By mapping the diagnosis result to the fault node in the knowledge graph (such as the "arc fault" node), and performing backward causal reasoning along the predefined causal relationship edges in the graph (for example, using a bidirectional attention mechanism to traverse the parent nodes and associated paths), the physical mechanism chain of the fault occurrence can be analyzed layer by layer. The data-driven diagnosis result is deeply integrated with domain knowledge. For example, when the diagnosis result is "abnormal increase in oil temperature", the backward search may successively locate upstream nodes such as "cooling system blockage" and "dust accumulation on the radiator fin", and calculate the contribution weights of each path through a graph attention network to screen out the most probable cause combination. The generated explanation text of the fault cause not only lists the direct cause (such as "oil pump failure"), but also generates a causal narrative that conforms to the equipment operation and maintenance experience based on the spatio-temporal association rules in the graph (such as "oil pump failure → decrease in oil flow rate → increase in oil temperature gradient"), and at the same time eliminates low-correlation paths (such as the instantaneous impact of ambient temperature fluctuation on oil temperature) through dynamic pruning, enabling the operation and maintenance personnel to quickly locate the root cause of the fault and formulate a targeted maintenance strategy.

[0074] In summary, the power equipment diagnosis method integrating a knowledge graph and deep learning according to the embodiments of the present application is clarified. By using a neural network model based on deep learning to perform temporal encoding on the dissolved gas concentration data sequence and oil temperature data sequence within a predetermined time period, the temporal correlation relationship in the dissolved gas concentration data and oil temperature data is extracted. Further, the temporal characteristics of the dissolved gas concentration data and the temporal characteristics of the oil temperature data are fused to accurately capture the interaction law of multiple physical quantities under the dynamic working conditions of the equipment. Then, the deep learning diagnosis result is mapped to the topological space of the knowledge graph to realize the bidirectional traceability of the fault node and the cause link. In this way, the robustness of the feature representation of complex coupling faults is improved, the causal closed-loop verification of the diagnosis result and the equipment operation mechanism is realized, and the formation of an intelligent decision-making system with both high-precision diagnosis ability and strong interpretability is promoted.

[0075] Furthermore, a power equipment diagnosis system integrating a knowledge graph and deep learning is also provided.

[0076] Figure 3 The block diagram of the power equipment diagnosis system integrating a knowledge graph and deep learning according to the embodiments of the present application is shown as Figure 3As shown in the figure, a power equipment diagnosis system 300 integrating a knowledge graph and deep learning according to an embodiment of the present application includes: a device information acquisition module 310 for acquiring device information; a data extraction module 320 for extracting a dissolved gas concentration data sequence and an oil temperature data sequence in oil within a predetermined time period from a background database based on the device information; a time series encoding module 330 for respectively performing time series encoding on the dissolved gas concentration data sequence and the oil temperature data sequence in oil to obtain a dissolved gas concentration in oil time series correlation implicit feature vector and an oil temperature time series correlation implicit feature vector; a time series fusion module 340 for fusing the dissolved gas concentration in oil time series correlation implicit feature vector and the oil temperature time series correlation implicit feature vector to obtain a dissolved gas concentration - oil temperature time series fine - grained collaborative encoding matrix; a fault diagnosis module 350 for obtaining a fault diagnosis result based on the dissolved gas concentration - oil temperature time series fine - grained collaborative encoding matrix; and a fault text explanation module 360 for obtaining a fault cause explanation text based on the node position corresponding to the fault diagnosis result in the power equipment knowledge graph.

[0077] As described above, the power equipment diagnosis system 300 integrating a knowledge graph and deep learning according to an embodiment of the present application can be implemented in various wireless terminals, such as a server with a power equipment diagnosis algorithm integrating a knowledge graph and deep learning. In a possible implementation manner, the power equipment diagnosis system 300 integrating a knowledge graph and deep learning according to an embodiment of the present application can be integrated into a wireless terminal as a software module and / or a hardware module. For example, the power equipment diagnosis system 300 integrating a knowledge graph and deep learning can be a software module in the operating system of the wireless terminal, or can be an application program developed for the wireless terminal; of course, the power equipment diagnosis system 300 integrating a knowledge graph and deep learning can also be one of the numerous hardware modules of the wireless terminal.

[0078] Alternatively, in another example, the power equipment diagnosis system 300 integrating a knowledge graph and deep learning and the wireless terminal can also be separate devices, and the power equipment diagnosis system 300 integrating a knowledge graph and deep learning can be connected to the wireless terminal through a wired and / or wireless network and transmit and interact information according to a predefined data format.

[0079] The embodiments of the present disclosure have been described above. The above description is exemplary and not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations are obvious to those of ordinary skill in the art in the technical field without departing from the scope and spirit of the described embodiments. The selection of the terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other ordinary skill in the art in the technical field to understand the disclosed embodiments.

Claims

1. A power equipment diagnosis method integrating knowledge graph and deep learning, characterized in that: include: Acquire device information of the power equipment to be diagnosed, wherein the power equipment to be diagnosed is a 110kV oil-immersed power transformer; Based on the equipment information, extract the data sequence of dissolved gas concentration in oil and the data sequence of oil temperature within a predetermined time period from the background database; Performing time series coding on the dissolved gas concentration data sequence in oil and the oil temperature data sequence respectively to obtain the time series associated implicit feature vector of the dissolved gas concentration in oil and the time series associated implicit feature vector of the oil temperature; Fusion of the implicit feature vector associated with the time series of dissolved gas concentration in oil and the implicit feature vector associated with the time series of oil temperature to obtain a fine-grained collaborative coding matrix of the time series of dissolved gas concentration in oil and oil temperature, including: performing a feature-level-based global interactive analysis of the time series of dissolved gas concentration in oil and oil temperature to obtain a fine-grained collaborative coding matrix of the time series of dissolved gas concentration in oil and oil temperature; Based on the fine-grained collaborative coding matrix of dissolved gas concentration in oil and oil temperature time series, the fault diagnosis results are obtained; Based on the node position corresponding to the fault diagnosis result in the power equipment knowledge graph, the fault cause explanation text is obtained.

2. The power equipment diagnosis method integrating knowledge graph and deep learning according to claim 1 is characterized in that: The dissolved gas concentration data sequence in oil and the oil temperature data sequence are respectively time-series encoded to obtain the dissolved gas concentration time-series associated implicit feature vector in oil and the oil temperature time-series associated implicit feature vector, including: The dissolved gas concentration data sequence in oil is passed through the gas concentration time series encoder based on the LSTM model to obtain the time series associated implicit feature vector of dissolved gas concentration in oil; The oil temperature data sequence is time-series encoded through an oil temperature data time-series encoder based on the LSTM model to obtain the oil temperature time-series associated implicit feature vector.

3. The power equipment diagnosis method integrating knowledge graph and deep learning according to claim 1 is characterized in that: The time series associated implicit feature vector of dissolved gas concentration in oil and the time series associated implicit feature vector of oil temperature are fused to obtain the fine-grained collaborative coding matrix of dissolved gas concentration in oil and oil temperature time series, including: Performing kernel correlation coding of the oil internal state based on principal component analysis on the implicit feature vector of the time series correlation of the dissolved gas concentration in oil and the implicit feature vector of the time series correlation of the oil temperature to obtain a set of kernel correlation coding vectors between the principal components of the time series characteristics of the dissolved gas concentration in oil and the oil temperature; A dissolved gas concentration-oil temperature characteristic performance operator is constructed for every two kernel correlation coding vectors between principal components of dissolved gas concentration-oil temperature time series characteristics in the set of kernel correlation coding vectors between principal components of dissolved gas concentration-oil temperature time series characteristics, and based on the dissolved gas concentration-oil temperature characteristic performance operator, the set of kernel correlation coding vectors between principal components of dissolved gas concentration-oil temperature time series characteristics is subjected to graph-structured cross-modal correlation coding to obtain a fine-grained collaborative coding matrix of dissolved gas concentration-oil temperature time series.

4. The power equipment diagnosis method integrating knowledge graph and deep learning according to claim 3 is characterized in that: The implicit feature vectors of dissolved gas concentration time series correlation and oil temperature time series correlation are subjected to kernel correlation coding of oil internal state based on principal component analysis to obtain a set of kernel correlation coding vectors between principal components of dissolved gas concentration in oil and oil temperature time series features, including: Performing principal component analysis on the implicit characteristic vectors associated with the time series of dissolved gas concentration in oil and the implicit characteristic vectors associated with the time series of oil temperature to obtain a set of principal component coding vectors of the time series characteristics of dissolved gas concentration in oil and a set of principal component coding vectors of the time series characteristics of oil temperature; Each corresponding group of principal component coding vectors of time series characteristics of dissolved gas concentration in oil and the principal component coding vectors of time series characteristics of oil temperature are input into the dissolved gas concentration-oil temperature principal component kernel association coding network respectively to obtain a set of kernel association coding vectors between principal components of dissolved gas concentration in oil and oil temperature time series characteristics.

5. The power equipment diagnosis method integrating knowledge graph and deep learning according to claim 4 is characterized in that: Construct a dissolved gas concentration-oil temperature characteristic performance operator for each two kernel correlation coding vectors between principal components of dissolved gas concentration-oil temperature time series characteristics in the set of kernel correlation coding vectors between principal components of dissolved gas concentration-oil temperature time series characteristics, and perform graph-structured cross-modal correlation coding on the set of kernel correlation coding vectors between principal components of dissolved gas concentration-oil temperature time series characteristics based on the dissolved gas concentration-oil temperature characteristic performance operator to obtain a fine-grained collaborative coding matrix of dissolved gas concentration-oil temperature time series, including: Calculate the dissolved gas concentration in oil-oil temperature characteristic performance operator between any two kernel correlation coding vectors between principal components of the dissolved gas concentration in oil-oil temperature time series characteristic in the set of kernel correlation coding vectors between principal components of the dissolved gas concentration in oil-oil temperature time series characteristic to obtain the dissolved gas concentration in oil-oil temperature characteristic performance operator correlation topological matrix; For each of the kernel correlation coding vectors between principal components of dissolved gas concentration-oil temperature time series characteristics in the set of kernel correlation coding vectors between principal components of dissolved gas concentration-oil temperature time series characteristics, the kernel spatial distribution of dissolved gas concentration-oil temperature time series characteristics components is randomly optimized to obtain a set of optimized kernel correlation coding vectors between principal components of dissolved gas concentration-oil temperature time series characteristics; The set of kernel correlation coding vectors between the principal components of the optimized dissolved gas concentration-oil temperature time series characteristics and the dissolved gas concentration-oil temperature characteristic performance operator correlation topological matrix are input into the graph convolutional neural network model to obtain the fine-grained collaborative coding matrix of the dissolved gas concentration-oil temperature time series.

6. The power equipment diagnosis method integrating knowledge graph and deep learning according to claim 5 is characterized in that: Based on the fine-grained collaborative coding matrix of dissolved gas concentration in oil and oil temperature time series, the fault diagnosis results are obtained, including: The fine-grained collaborative coding matrix of dissolved gas concentration in oil and oil temperature time series is passed through a decoder-based fault diagnosis model to obtain a fault diagnosis result, wherein the fault diagnosis result is used to represent the fault diagnosis score of the equipment.

7. The power equipment diagnosis method integrating knowledge graph and deep learning according to claim 6 is characterized in that: Based on the node position corresponding to the fault diagnosis result in the power equipment knowledge graph, the fault cause explanation text is obtained, including: The node corresponding to the fault diagnosis result is located in the power equipment knowledge graph, and the cause is reversely searched in the power equipment knowledge graph from the node to obtain the fault cause explanation text.

8. A power equipment diagnosis system integrating knowledge graph and deep learning, characterized in that: include: Device information acquisition module, used to acquire device information; A data extraction module is used to extract a data sequence of dissolved gas concentration in oil and a data sequence of oil temperature within a predetermined time period from a background database based on equipment information; A time series encoding module, used for performing time series encoding on the dissolved gas concentration data sequence in oil and the oil temperature data sequence respectively to obtain a time series associated implicit feature vector of the dissolved gas concentration in oil and a time series associated implicit feature vector of the oil temperature; A time series fusion module is used to fuse the time series associated implicit feature vector of dissolved gas concentration in oil and the time series associated implicit feature vector of oil temperature to obtain a fine-grained collaborative coding matrix of dissolved gas concentration in oil and oil temperature time series; A fault diagnosis module is used to obtain fault diagnosis results based on a fine-grained collaborative coding matrix of dissolved gas concentration in oil and oil temperature time series; The fault text explanation module is used to obtain the fault cause explanation text based on the node position corresponding to the fault diagnosis result in the power equipment knowledge graph.

Images