Fault discrimination method and system for mechanism data hybrid driven transformer, and medium
By constructing the training data set and combining the mixed driving method of KAN network and oil chromatography three-ratio method, the problems of insufficient accuracy and poor interpretability in transformer fault diagnosis are solved, and efficient and accurate fault judgment is achieved.
Patent Information
- Application Number
- CN202510613139.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-13
- Publication Date
- 2025-08-15
AI Technical Summary
Existing transformer fault diagnosis methods have problems such as insufficient accuracy, low efficiency and poor interpretability, especially the traditional oil chromatography analysis and expert experience methods lack objectivity and consistency, and the data-driven methods have strong data dependence and poor interpretability.
Using the method of hybrid driving of mechanism data, the distance between unknown data samples and data center points is calculated, combined with the Kolmogolov-Arnold network (KAN) and the oil chromatography tri-ratio method, a combination of data-driven model and mechanism-driven model are used for fault judgment.
It realizes efficient, accurate and interpretable transformer fault judgment, improves adaptability to different fault types and data distributions, and improves the accuracy and reliability of fault diagnosis.
Smart Images
Figure CN120492849A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of transformer fault diagnosis, and in particular relates to a transformer fault identification method, system and medium driven by a hybrid mechanism and data. Background Art
[0002] Transformers are one of the most critical electrical devices in power systems. Their primary function is to change voltage levels, enabling the efficient transmission and distribution of electrical energy. The proper functioning of transformers is crucial for the safe, stable, and reliable operation of power systems. However, due to various factors, including long-term operation, environmental factors, aging, and sudden failures, transformers may experience various faults, such as insulation degradation, partial discharge, overheating, and short circuits. These faults can not only damage the transformer but also cause power outages and even endanger the safe operation of the entire power system. Therefore, timely and accurate diagnosis of transformer faults and implementation of effective maintenance measures are extremely important for ensuring the stable operation of power systems.
[0003] Traditional transformer fault diagnosis methods primarily rely on oil chromatography analysis and expert experience. Oil chromatography analysis measures the content of dissolved gases (such as hydrogen, methane, ethane, ethylene, and acetylene) in transformer oil and uses methods such as the three-ratio method to determine the fault type. While this method can reflect the internal condition of the transformer to a certain extent, the three-ratio method's encoding rules and fault type determination are primarily based on experience, which can lead to inaccurate judgments for complex or mixed fault types. Relying solely on gas content ratios makes it difficult to fully understand the physical process of the fault and lacks in-depth analysis of the fault mechanism. Expert experience-based diagnosis relies on long-term expert experience in transformer fault characteristics and develops a series of rules to determine the fault type. This reliance on expert experience and judgment can lead to different conclusions among different experts, lacking objectivity and consistency. Empirical rules are difficult to formalize and quantify, making them difficult to generalize and apply to different equipment and scenarios.
[0004] In recent years, with the development of machine learning and deep learning technologies, data-driven fault diagnosis methods have gradually attracted attention. These methods build fault diagnosis models by learning fault characteristics from large amounts of historical data. However, these methods also suffer from strong data dependence and poor interpretability. They require a large amount of historical data for training, placing high demands on both data quality and quantity. Most data-driven models (such as neural networks) are black box models, making it difficult to explain their decision-making process and rationale. For new, unseen fault types or data distributions, the models may not generalize well enough, making misjudgments more likely. Summary of the Invention
[0005] The purpose of the present invention is to address the problems in the above-mentioned prior art and provide a transformer fault identification method, system and medium driven by a hybrid mechanism and data, so as to achieve efficient, accurate and explainable transformer fault identification.
[0006] In order to achieve the above object, the present invention has the following technical solutions:
[0007] In a first aspect, a transformer fault identification method driven by a hybrid mechanism and data is provided, comprising:
[0008] Based on the historical fault data of the transformer, a training data set is constructed, and the data center point and maximum radius of the training data set are determined;
[0009] Collecting transformer unknown data samples, calculating the distance between the transformer unknown data samples and the center point of the training data set, and comparing the calculated distance with the maximum radius of the training data set; if the calculated distance between the transformer unknown data sample and the center point of the training data set is not greater than the maximum radius of the training data set, then determining that the transformer unknown data sample is within the range of the training data set; otherwise, determining that the transformer unknown data sample is outside the range of the training data set;
[0010] If the unknown data sample of the transformer is within the range of the training data set, the pre-trained data-driven model is used to perform fault discrimination and obtain the fault discrimination result; if the unknown data sample of the transformer is outside the range of the training data set, the pre-trained mechanism-driven model is used to perform fault discrimination and obtain the fault discrimination result.
[0011] As a preferred solution, the maximum radius of the training data set is determined as follows:
[0012] Calculate the distance d between the center point and each data sample point:
[0013] d=‖x0-x m ‖m=1,…,m
[0014] In the formula, x0 is the center point of the training data set, x m is the data sample point in the training data set other than the center point, and M is the total number of data sample points;
[0015] Calculate the maximum value of the distance d between the center point and each data sample point to obtain the maximum radius of the training data set:
[0016] d max =maxd.
[0017] As a preferred solution, the data-driven model is built based on the Kolmogorov-Arnold network KAN. For any multivariable continuous function f(x1, ..., x n ), are all expressed as a combination of finite unary continuous functions:
[0018]
[0019] Where x n is the input data; q,p is an internal function; Φ q For external functions;
[0020] The KAN network consists of multiple layers, including input layer, hidden layer and output layer; the input layer is used to receive the characteristic information of the transformer, the hidden layer extracts and maps the features of the input data through nonlinear transformation, and the output layer performs fault diagnosis based on the output results of the hidden layer; each hidden layer is composed of multiple neurons, each neuron corresponds to an internal function, and the internal functions are combined in a specific way to form an external function, thereby achieving function approximation.
[0021] As a preferred solution, during the training process of the KAN network, the parameters of the KAN network are optimized. The goal of parameter optimization is to minimize the loss function, which includes prediction error and regularization term. The prediction error is used to measure the difference between the network output and the true label, and the regularization term is used to prevent the KAN network from overfitting. During the training process, the gradient of the loss function with respect to the KAN network parameters is calculated through the backpropagation algorithm, and the KAN network parameters are updated according to the gradient.
[0022] As a preferred solution, the mechanism-driven model uses oil chromatography as transformer status data and a three-ratio method as a mechanism-level identification tool; based on the mutual dependence of the relative concentrations of gas components generated by the decomposition of oil and insulation in oil-filled electrical equipment under fault conditions and temperature, two gases are selected from the characteristic gases according to their solubility and diffusion coefficient to form three ratios, which are represented by different codes; and the coding rules and fault type judgment method are used as the basis for diagnosing the nature of the fault.
[0023] As a preferred solution, the encoding rules are as follows:
[0024] When the gas range is <0.1, the C2H2 / C2H4 ratio range is coded as 0, the CH4 / H2 ratio range is coded as 1, and the C2H2 / C2H6 ratio range is coded as 0;
[0025] When the gas range belongs to [0.1,1), the ratio range of C2H2 / C2H4 is coded as 1, the ratio range of CH4 / H2 is coded as 0, and the ratio range of C2H2 / C2H6 is coded as 0;
[0026] When the gas range belongs to [1,3), the ratio range of C2H2 / C2H4 is coded as 1, the ratio range of CH4 / H2 is coded as 2, and the ratio range of C2H2 / C2H6 is coded as 1;
[0027] When the gas range is ≥ 3, the C2H2 / C2H4 ratio range is coded as 2, the CH4 / H2 ratio range is coded as 2, and the C2H2 / C2H6 ratio range is coded as 2.
[0028] As a preferred solution, the fault type determination method is as follows:
[0029] When the coding combination is C2H2 / C2H4 is 0, CH4 / H2 is 0, and C2H2 / C2H6 is 0, the fault type is low-temperature overheating below 150°C. Typical faults caused include overheating of paper-insulated wires;
[0030] When the coding combination is C2H2 / C2H4=0, CH4 / H2=2, and C2H2 / C2H6=0, the fault type is low-temperature overheating between 150°C and 300°C. When the coding combination is C2H2 / C2H4=0, CH4 / H2=2, and C2H2 / C2H6=1, the fault type is medium-temperature overheating between 300°C and 700°C. When the coding combination is C2H2 / C2H4=0, CH4 / H2=0, 1, or 2, and C2H2 / C2H6=2, the fault type is high-temperature overheating above 700°C. Typical faults caused include poor contact of the tap changer, poor lead connection, poor welding of wire joints, overheating caused by short circuit between strands, multiple grounding of the core, and local short circuit between silicon steel sheets.
[0031] When the coding combination is C2H2 / C2H4 is 0, CH4 / H2 is 1, and C2H2 / C2H6 is 0, the fault type is partial discharge, and the typical faults caused include discharge with low energy density;
[0032] When the coding combination is C2H2 / C2H4=2, CH4 / H2=0 or 1, C2H2 / C2H6=0 or 1 or 2, the fault type is low-energy discharge. When the coding combination is C2H2 / C2H4=2, CH4 / H2=2, C2H2 / C2H6=0 or 1 or 2, the fault type is low-energy discharge and overheating. Typical faults caused include spark discharge between different potentials and circulating current between the lead and the cable sleeve or lead shielding tube.
[0033] When the coding combination is C2H2 / C2H4 is 1, CH4 / H2 is 0 or 1, and C2H2 / C2H6 is 0 or 1 or 2, the fault type is arc discharge. When the coding combination is C2H2 / C2H4 is 1, CH4 / H2 is 2, and C2H2 / C2H6 is 0 or 1 or 2, the fault type is arc discharge and overheating. Typical faults caused include discharge between coil turns and layers, phase flashover, oil gap flashover between tap leads, arcing of selector switch, and discharge of leads to box shell or other grounding bodies.
[0034] As a preferred solution, in the step of collecting unknown data samples of the transformer, calculating the distance between the unknown data samples of the transformer and the center point of the training data set, and comparing the calculated distance with the maximum radius of the training data set, an unknown oil chromatogram data sample is collected for state discrimination, and the unknown oil chromatogram data sample is subjected to the same preprocessing operations as the training data set, including data cleaning, feature selection and extraction, and data normalization, and then the distance d between the unknown oil chromatogram data sample and the center point of the training data set is calculated. u ; Let the unknown sample point be x u , the center point of the training data set is x0, then the distance d u The calculation expression is as follows:
[0035] d u =‖x u -x0‖
[0036] Among them, ‖x u -x0‖ represents x u and the L2 norm of x0.
[0037] In the second aspect, a transformer fault identification system driven by a hybrid mechanism and data is provided, including:
[0038] A training data set construction module is used to construct a training data set based on the historical fault data of the transformer and determine the data center point and maximum radius of the training data set;
[0039] An unknown data sample range determination module is used to collect unknown data samples of the transformer, calculate the distance between the unknown data samples of the transformer and the data center point of the training data set, and compare the calculated distance with the maximum radius of the training data set. If the calculated distance between the unknown data samples of the transformer and the data center point of the training data set is not greater than the maximum radius of the training data set, then the unknown data samples of the transformer are determined to be within the range of the training data set; otherwise, the unknown data samples of the transformer are determined to be outside the range of the training data set.
[0040] The hybrid-driven fault discrimination module is used to use a pre-trained data-driven model to perform fault discrimination if the transformer unknown data sample is within the range of the training data set to obtain a fault discrimination result; if the transformer unknown data sample is outside the range of the training data set, the pre-trained mechanism-driven model is used to perform fault discrimination to obtain a fault discrimination result.
[0041] As a preferred solution, the training data set construction module determines the maximum radius of the training data set in the following manner:
[0042] Calculate the distance d between the center point and each data sample point:
[0043] d=‖x0-x m ‖m=1,…,M
[0044] In the formula, x0 is the center point of the training data set, x m is the data sample point in the training data set other than the center point, and M is the total number of data sample points;
[0045] Calculate the maximum value of the distance d between the center point and each data sample point to obtain the maximum radius of the training data set:
[0046] d max =maxd.
[0047] As a preferred solution, the hybrid drive fault discrimination module builds a data-driven model based on the Kolmogorov-Arnold network KAN. For any multivariable continuous function f(x1, ..., x n ), are all expressed as a combination of finite unary continuous functions:
[0048]
[0049] Where x n is the input data; q,p is an internal function; Φ q For external functions;
[0050] The KAN network consists of multiple layers, including input layer, hidden layer and output layer; the input layer is used to receive the characteristic information of the transformer, the hidden layer extracts and maps the features of the input data through nonlinear transformation, and the output layer performs fault diagnosis based on the output results of the hidden layer; each hidden layer is composed of multiple neurons, each neuron corresponds to an internal function, and the internal functions are combined in a specific way to form an external function, thereby achieving function approximation.
[0051] As a preferred solution, the hybrid drive fault discrimination module optimizes the parameters of the KAN network during the training process of the KAN network. The goal of parameter optimization is to minimize the loss function, which includes prediction error and regularization term. The prediction error is used to measure the difference between the network output and the true label, and the regularization term is used to prevent the KAN network from overfitting. During the training process, the gradient of the loss function with respect to the KAN network parameters is calculated through the back propagation algorithm, and the KAN network parameters are updated according to the gradient.
[0052] As a preferred solution, when the hybrid-driven fault discrimination module uses a pre-trained mechanism-driven model for fault discrimination, the mechanism-driven model uses oil chromatography as the data for discriminating the transformer status, and uses the three-ratio method as a discrimination tool at the mechanism level; based on the mutual dependence of the relative concentration of the gas component content generated by the decomposition of oil and insulation in oil-filled electrical equipment under fault conditions and temperature, two gases are selected from the characteristic gases according to the solubility and diffusion coefficient to form three ratio values, which are represented by different codes; the coding rules and the fault type judgment method are used as the basis for diagnosing the nature of the fault.
[0053] As a preferred solution, the unknown data sample range determination module collects an unknown oil chromatogram data sample for state determination, performs the same preprocessing operations on the unknown oil chromatogram data sample as the training data set, including data cleaning, feature selection and extraction, and data normalization, and then calculates the distance d between the unknown oil chromatogram data sample and the center point of the training data set. u ;
[0054] Let the unknown sample point be x u , the center point of the training data set is x0, then the distance d u The calculation expression is as follows:
[0055] d u =‖x u -x0‖
[0056] Among them, ‖x u -x0‖ represents x u and the L2 norm of x0.
[0057] In a third aspect, an electronic device is provided, comprising a processor and a memory, wherein the processor is configured to execute a computer program stored in the memory to implement the transformer fault identification method driven by hybrid mechanism data.
[0058] In a fourth aspect, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores at least one instruction, and when the at least one instruction is executed by a processor, the transformer fault identification method driven by hybrid mechanism data is implemented.
[0059] Compared with the prior art, the first aspect of the present invention has at least the following beneficial effects:
[0060] A training data set is constructed based on the historical fault data of the transformer. The distance between the unknown data sample of the transformer and the center point of the training data set is calculated and compared with the maximum radius of the training data set to determine whether the unknown data sample of the transformer is within the range of the training data set or outside the range. If the unknown data sample of the transformer is within the range of the training data set, the data-driven model is used for fault discrimination. If the unknown data sample of the transformer is outside the range of the training data set, the mechanism-driven model is used for fault discrimination to obtain a fault discrimination result. The present invention can achieve efficient, accurate and explainable fault diagnosis by integrating deep learning technology and physical constraint models. By integrating data-driven and mechanism-driven methods, it can also better adapt to different fault types and data distributions. The data-driven model can learn fault characteristics from a large amount of historical data, while the mechanism-driven model provides an in-depth understanding of the physical process of the fault. This fusion method enables the model to have stronger generalization ability when facing new and unseen fault types or data distributions, and can perform fault diagnosis more accurately.
[0061] Furthermore, the data-driven model of the present invention is built based on the Kolmogorov-Arnold network KAN. The mechanism-driven model uses oil chromatography as the data for distinguishing transformer status, and uses the three-ratio method as a discrimination tool at the mechanism level. According to the mutual dependence of the relative concentration of the gas component content generated by the decomposition of oil and insulation in oil-filled electrical equipment under fault conditions and temperature, two gases are selected from the characteristic gases according to the solubility and diffusion coefficient to form three ratios, which are represented by different codes; according to the coding rules and the fault type judgment method as the basis for diagnosing the nature of the fault, this method eliminates the influence of the volume effect of the oil and can obtain a more reliable diagnosis of the fault state. By combining the data-driven KAN network and the mechanism-driven three-ratio method, the present invention fully utilizes the statistical laws of the data and the prior knowledge of the physical model, and can more comprehensively capture the fault characteristics, thereby significantly improving the accuracy of fault diagnosis and enabling the decision-making process of the model to be explained based on physical laws.
[0062] It can be understood that the beneficial effects of the second to fourth aspects mentioned above can be found in the relevant description of the first aspect mentioned above, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS
[0063] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments or descriptions of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.
[0064] Figure 1 Flowchart of the transformer fault identification method driven by hybrid mechanism data according to an embodiment of the present invention;
[0065] Figure 2 Schematic diagram of the KAN network architecture according to an embodiment of the present invention;
[0066] Figure 3 A structural block diagram of a transformer fault identification system driven by a hybrid mechanism and data according to an embodiment of the present invention. DETAILED DESCRIPTION
[0067] In the following description, specific details such as specific system structures and techniques are provided for purposes of illustration rather than limitation to facilitate a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid obscuring the description of the present application with unnecessary detail.
[0068] See also Figure 1 The embodiment of the present invention proposes a transformer fault identification method driven by a hybrid mechanism and data, including:
[0069] Based on the historical fault data of the transformer, a training data set is constructed, and the data center point and maximum radius of the training data set are determined;
[0070] Collecting transformer unknown data samples, calculating the distance between the transformer unknown data samples and the center point of the training data set, and comparing the calculated distance with the maximum radius of the training data set; if the calculated distance between the transformer unknown data sample and the center point of the training data set is not greater than the maximum radius of the training data set, then determining that the transformer unknown data sample is within the range of the training data set; otherwise, determining that the transformer unknown data sample is outside the range of the training data set;
[0071] If the unknown data sample of the transformer is within the range of the training data set, the pre-trained data-driven model is used to perform fault discrimination and obtain the fault discrimination result; if the unknown data sample of the transformer is outside the range of the training data set, the pre-trained mechanism-driven model is used to perform fault discrimination and obtain the fault discrimination result.
[0072] In one possible implementation, the maximum radius of the training data set is determined as follows:
[0073] Calculate the distance d between the center point and each data sample point:
[0074] d=‖x0-x m ‖m=1,…,M
[0075] In the formula, x0 is the center point of the training data set, x m is the data sample point in the training data set other than the center point, and M is the total number of data sample points;
[0076] Calculate the maximum value of the distance d between the center point and each data sample point to obtain the maximum radius of the training data set:
[0077] d max =maxd
[0078] In one possible implementation, a data-driven transformer fault discrimination model is built based on KAN (Kolmogorov-Arnold Networks). The KAN network adopts the Kolmogorov-Arnold representation theorem, aiming to improve the flexibility and expressiveness of the model while maintaining the interpretability of the model by introducing a learnable activation function on the network edge (i.e., weight). The KAN network of the embodiment of the present invention is used for any multivariable continuous function f(x1, ..., x1) defined in a closed interval. n ), are all expressed as a combination of finite unary continuous functions:
[0079]
[0080] Where x n is the input data; q,p is an internal function; Φ q For external functions;
[0081] The KAN network consists of multiple layers, including input layer, hidden layer and output layer; the input layer is used to receive the characteristic information of the transformer, the hidden layer extracts and maps the input data through nonlinear transformation, and the output layer performs fault identification based on the output results of the hidden layer; each hidden layer is composed of multiple neurons, each neuron corresponds to an internal function, and the internal functions are combined in a specific way to form an external function, thereby achieving approximation of complex functions. Figure 1 shown.
[0082] During the training process of the KAN network, the parameters of the KAN network are optimized. The goal of parameter optimization is to minimize the loss function, which includes prediction error and regularization term. The prediction error is used to measure the difference between the network output and the true label, and the regularization term is used to prevent the KAN network from overfitting. During the training process, the gradient of the loss function with respect to the KAN network parameters is calculated through the backpropagation algorithm, and the KAN network parameters are updated according to the gradient to gradually optimize the network performance.
[0083] In one possible implementation, the mechanism-driven model uses oil chromatography as data for identifying transformer status. The embodiment of the present invention uses the three-ratio method as a mechanism-level identification tool. Based on the interdependence between the relative concentrations of gas components generated by the decomposition of oil and insulation in oil-filled electrical equipment under fault conditions and temperature, two gases with similar solubility and diffusion coefficients are selected from five characteristic gases to form three ratios, which are represented by different codes. The coding rules in Table 1 and the fault type determination method in Table 2 are used as the basis for diagnosing the nature of the fault. This method eliminates the influence of the volume effect of the oil, is the primary method for determining the fault type of oil-filled electrical equipment, and can provide a more reliable diagnosis of the fault condition.
[0084] Table 1
[0085]
[0086]
[0087] As shown in Table 1, the encoding rules are as follows:
[0088] When the gas range is <0.1, the C2H2 / C2H4 ratio range is coded as 0, the CH4 / H2 ratio range is coded as 1, and the C2H2 / C2H6 ratio range is coded as 0;
[0089] When the gas range belongs to [0.1,1), the ratio range of C2H2 / C2H4 is coded as 1, the ratio range of CH4 / H2 is coded as 0, and the ratio range of C2H2 / C2H6 is coded as 0;
[0090] When the gas range belongs to [1,3), the ratio range of C2H2 / C2H4 is coded as 1, the ratio range of CH4 / H2 is coded as 2, and the ratio range of C2H2 / C2H6 is coded as 1;
[0091] When the gas range is ≥ 3, the C2H2 / C2H4 ratio range is coded as 2, the CH4 / H2 ratio range is coded as 2, and the C2H2 / C2H6 ratio range is coded as 2.
[0092] Table 2
[0093]
[0094]
[0095] As shown in Table 2, the fault type judgment method is as follows:
[0096] When the coding combination is C2H2 / C2H4 is 0, CH4 / H2 is 0, and C2H2 / C2H6 is 0, the fault type is low-temperature overheating below 150°C. Typical faults caused include overheating of paper-insulated wires;
[0097] When the coding combination is C2H2 / C2H4=0, CH4 / H2=2, and C2H2 / C2H6=0, the fault type is low-temperature overheating between 150°C and 300°C. When the coding combination is C2H2 / C2H4=0, CH4 / H2=2, and C2H2 / C2H6=1, the fault type is medium-temperature overheating between 300°C and 700°C. When the coding combination is C2H2 / C2H4=0, CH4 / H2=0, 1, or 2, and C2H2 / C2H6=2, the fault type is high-temperature overheating above 700°C. Typical faults caused include poor contact of the tap changer, poor lead connection, poor welding of wire joints, overheating caused by short circuit between strands, multiple grounding of the core, and local short circuit between silicon steel sheets.
[0098] When the coding combination is C2H2 / C2H4 is 0, CH4 / H2 is 1, and C2H2 / C2H6 is 0, the fault type is partial discharge, and the typical faults caused include discharge with low energy density;
[0099] When the coding combination is C2H2 / C2H4=2, CH4 / H2=0 or 1, C2H2 / C2H6=0 or 1 or 2, the fault type is low-energy discharge. When the coding combination is C2H2 / C2H4=2, CH4 / H2=2, C2H2 / C2H6=0 or 1 or 2, the fault type is low-energy discharge and overheating. Typical faults caused include spark discharge between different potentials and circulating current between the lead and the cable sleeve or lead shielding tube.
[0100] When the coding combination is C2H2 / C2H4 is 1, CH4 / H2 is 0 or 1, and C2H2 / C2H6 is 0 or 1 or 2, the fault type is arc discharge. When the coding combination is C2H2 / C2H4 is 1, CH4 / H2 is 2, and C2H2 / C2H6 is 0 or 1 or 2, the fault type is arc discharge and overheating. Typical faults caused include discharge between coil turns and layers, phase flashover, oil gap flashover between tap leads, arcing of selector switch, and discharge of leads to box shell or other grounding bodies.
[0101] In one possible implementation, an unknown oil chromatogram data sample is collected for state identification. The unknown oil chromatogram data sample is subjected to the same preprocessing operations as the training data set, including data cleaning, feature selection and extraction, and data normalization, to ensure data consistency and comparability. The distance d between the unknown oil chromatogram data sample and the center point of the training data set is then calculated. u ; Let the unknown sample point be x u , the center point of the training data set is x0, then the distance d u The calculation expression is as follows:
[0102] d u =‖xu -x0‖
[0103] Among them, ‖x u -x0‖ represents x u and the L2 norm of x0.
[0104] The calculated distance d u The maximum radius d of the training dataset max Compare. If d u ≤d max , the unknown sample point is considered to be within the range of the data set; otherwise, the unknown sample point is considered to be outside the range of the data set.
[0105] If the unknown sample point is within the dataset, the trained KAN network model can be used for fault identification. The feature value of the unknown sample point is input into the KAN network to obtain the fault identification result. If the unknown sample point is outside the dataset, the mechanism-driven transformer fault identification model is used for fault identification.
[0106] The present invention proposes a hybrid driving method that combines a data-driven model (based on a KAN network) with a mechanism-driven model (based on a three-ratio method) for transformer fault identification. This method fully utilizes the statistical learning ability of the data-driven model and the physical interpretability of the mechanism-driven model, overcomes the limitations of traditional single methods, and significantly improves the accuracy and reliability of fault diagnosis. By integrating the advantages of the two models, the accuracy, efficiency, and interpretability of fault diagnosis are significantly improved. By calculating the distance between the unknown sample point and the center point of the training data set, it is determined whether the sample is within the range of the data set. This method provides a scientific basis for model selection, ensuring that the most appropriate fault identification model is selected in different scenarios.
[0107] See also Figure 3 The embodiment of the present invention further proposes a transformer fault identification system driven by a hybrid mechanism and data, including:
[0108] A training data set construction module is used to construct a training data set based on the historical fault data of the transformer and determine the data center point and maximum radius of the training data set;
[0109] An unknown data sample range determination module is used to collect unknown data samples of the transformer, calculate the distance between the unknown data samples of the transformer and the data center point of the training data set, and compare the calculated distance with the maximum radius of the training data set. If the calculated distance between the unknown data samples of the transformer and the data center point of the training data set is not greater than the maximum radius of the training data set, then the unknown data samples of the transformer are determined to be within the range of the training data set; otherwise, the unknown data samples of the transformer are determined to be outside the range of the training data set.
[0110] The hybrid-driven fault discrimination module is used to use a pre-trained data-driven model to perform fault discrimination if the transformer unknown data sample is within the range of the training data set to obtain a fault discrimination result; if the transformer unknown data sample is outside the range of the training data set, the pre-trained mechanism-driven model is used to perform fault discrimination to obtain a fault discrimination result.
[0111] Furthermore, the training dataset construction module determines the maximum radius of the training dataset as follows:
[0112] Calculate the distance d between the center point and each data sample point:
[0113] d=‖x0-x m ‖m=1,…,M
[0114] In the formula, x0 is the center point of the training data set, x m is the data sample point in the training data set other than the center point, and M is the total number of data sample points;
[0115] Calculate the maximum value of the distance d between the center point and each data sample point to obtain the maximum radius of the training data set:
[0116] d max =maxd.
[0117] In a possible implementation, the hybrid drive fault discrimination module builds a data-driven model based on the Kolmogorov-Arnold network KAN. For any multivariable continuous function f(x1, ..., x n ), are all expressed as a combination of finite unary continuous functions:
[0118]
[0119] Where x n is the input data; q,p is an internal function; Φ q For external functions;
[0120] The KAN network consists of multiple layers, including input layer, hidden layer and output layer; the input layer is used to receive the characteristic information of the transformer, the hidden layer extracts and maps the features of the input data through nonlinear transformation, and the output layer performs fault diagnosis based on the output results of the hidden layer; each hidden layer is composed of multiple neurons, each neuron corresponds to an internal function, and the internal functions are combined in a specific way to form an external function, thereby achieving function approximation.
[0121] In one possible implementation, the hybrid drive fault discrimination module optimizes the parameters of the KAN network during the training process of the KAN network. The goal of parameter optimization is to minimize the loss function, which includes a prediction error and a regularization term. The prediction error is used to measure the difference between the network output and the true label, and the regularization term is used to prevent the KAN network from overfitting. During the training process, the gradient of the loss function with respect to the KAN network parameters is calculated through a backpropagation algorithm, and the KAN network parameters are updated based on the gradient.
[0122] In one possible embodiment, when the hybrid-driven fault discrimination module uses a pre-trained mechanism-driven model for fault discrimination, the mechanism-driven model uses oil chromatography as transformer status data for discrimination, and employs a three-ratio method as a mechanism-level discrimination tool; based on the mutual dependence between the relative concentrations of gas components generated by the decomposition of oil and insulation in oil-filled electrical equipment under fault conditions and temperature, two gases are selected from the characteristic gases according to their solubility and diffusion coefficient to form three ratio values, which are represented by different codes; and the coding rules and the fault type judgment method are used as the basis for diagnosing the nature of the fault.
[0123] In one possible implementation, the unknown data sample range determination module collects an unknown oil chromatogram data sample for state determination, performs the same preprocessing operations on the unknown oil chromatogram data sample as the training data set, including data cleaning, feature selection and extraction, and data normalization, and then calculates the distance d between the unknown oil chromatogram data sample and the center point of the training data set. u ; Let the unknown sample point be x u , the center point of the training data set is x0, then the distance d u The calculation expression is as follows:
[0124] d u =‖x u -x0‖
[0125] Among them, ‖x u -x0‖ represents x u and the L2 norm of x0.
[0126] Another embodiment of the present invention further provides an electronic device, including a processor and a memory, wherein the processor is configured to execute a computer program stored in the memory to implement the transformer fault identification method driven by hybrid mechanism data.
[0127] Another embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores at least one instruction, and when the at least one instruction is executed by a processor, the transformer fault identification method driven by hybrid mechanism data is implemented.
[0128] By combining deep learning technology and physical constraint models, the present invention achieves efficient, accurate and interpretable transformer fault identification, thereby solving the problems of insufficient accuracy, low efficiency and poor interpretability existing in traditional fault diagnosis methods.
[0129] The computer program includes computer program code, which may be in source code form, object code form, executable file or some intermediate form. The computer-readable storage medium may include: any entity or device, medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory, random access memory, electric carrier signal, telecommunication signal and software distribution medium that can carry the computer program code. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals. For ease of explanation, the above content only shows the part related to the embodiment of the present invention. For specific technical details not disclosed, please refer to the method part of the embodiment of the present invention. The computer-readable storage medium is non-transitory and can be stored in a storage device formed by various electronic devices, and can implement the execution process recorded in the method of the embodiment of the present invention.
[0130] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, 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 magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0131] 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 flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, 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 flowcharts and / or block diagrams. 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.
[0132] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work 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 The function specified in one or more boxes.
[0133] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The steps for the function specified in one or more boxes.
[0134] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.
Claims
1. A transformer fault identification method driven by hybrid mechanism data, characterized in that: include: Based on the historical fault data of the transformer, a training data set is constructed, and the data center point and maximum radius of the training data set are determined; Collecting transformer unknown data samples, calculating the distance between the transformer unknown data samples and the center point of the training data set, and comparing the calculated distance with the maximum radius of the training data set; if the calculated distance between the transformer unknown data sample and the center point of the training data set is not greater than the maximum radius of the training data set, then determining that the transformer unknown data sample is within the range of the training data set; otherwise, determining that the transformer unknown data sample is outside the range of the training data set; If the unknown data sample of the transformer is within the range of the training data set, the pre-trained data-driven model is used to perform fault identification and obtain the fault identification result; If the unknown data sample of the transformer is outside the range of the training data set, the pre-trained mechanism-driven model is used to perform fault discrimination and obtain the fault discrimination result.
2. The transformer fault identification method driven by hybrid mechanism data according to claim 1 is characterized in that: The maximum radius of the training data set is determined as follows: Calculate the distance d between the center point and each data sample point: d=‖x0-x m ‖m=1,…,M In the formula, x0 is the center point of the training data set, x m is the data sample point in the training data set other than the center point, and M is the total number of data sample points; Calculate the maximum value of the distance d between the center point and each data sample point to obtain the maximum radius of the training data set: <h2 style=";text-align:left;direction:ltr">d<h2 style=";text-align:left;direction:ltr"> max <h2 style=";text-align:left;direction:ltr"> (maxd) 3. The transformer fault identification method driven by hybrid mechanism data according to claim 1 is characterized in that: The data-driven model is built based on the Kolmogorov-Arnold network KAN. For any multivariable continuous function f(x1, ..., x n ), are all expressed as a combination of finite unary continuous functions: Where x n is the input data; q,p is an internal function; Φ q For external functions; The KAN network consists of multiple layers, including input layer, hidden layer and output layer; the input layer is used to receive the characteristic information of the transformer, the hidden layer extracts and maps the features of the input data through nonlinear transformation, and the output layer performs fault diagnosis based on the output results of the hidden layer; each hidden layer is composed of multiple neurons, each neuron corresponds to an internal function, and the internal functions are combined in a specific way to form an external function, thereby achieving function approximation.
4. The transformer fault identification method driven by mechanism data hybrid according to claim 3 is characterized in that: During the training process of the KAN network, the parameters of the KAN network are optimized. The goal of parameter optimization is to minimize the loss function, which includes prediction error and regularization term. The prediction error is used to measure the difference between the network output and the true label, and the regularization term is used to prevent the KAN network from overfitting. During the training process, the gradient of the loss function with respect to the KAN network parameters is calculated through the backpropagation algorithm, and the KAN network parameters are updated according to the gradient.
5. The transformer fault identification method driven by hybrid mechanism data according to claim 1 is characterized in that: The mechanism-driven model uses oil chromatography as data for identifying transformer status and a three-ratio method as a mechanism-level identification tool. Based on the interdependence between the relative concentrations of gas components generated by the decomposition of oil and insulation in oil-filled electrical equipment under fault conditions and temperature, two gases are selected from characteristic gases according to their solubility and diffusion coefficient to form three ratios, which are represented by different codes. The coding rules and the fault type judgment method are used as the basis for diagnosing the nature of the fault.
6. The transformer fault identification method driven by mechanism data hybrid according to claim 5 is characterized in that: The encoding rules are as follows: When the gas range is <0.1, the C2H2 / C2H4 ratio range is coded as 0, the CH4 / H2 ratio range is coded as 1, and the C2H2 / C2H6 ratio range is coded as 0; When the gas range belongs to [0.1,1), the ratio range of C2H2 / C2H4 is coded as 1, the ratio range of CH4 / H2 is coded as 0, and the ratio range of C2H2 / C2H6 is coded as 0; When the gas range belongs to [1,3), the ratio range of C2H2 / C2H4 is coded as 1, the ratio range of CH4 / H2 is coded as 2, and the ratio range of C2H2 / C2H6 is coded as 1; When the gas range is ≥ 3, the C2H2 / C2H4 ratio range is coded as 2, the CH4 / H2 ratio range is coded as 2, and the C2H2 / C2H6 ratio range is coded as 2.
7. The transformer fault identification method driven by mechanism data hybrid according to claim 6 is characterized in that: The method for determining the fault type is as follows: When the coding combination is C2H2 / C2H4 is 0, CH4 / H2 is 0, and C2H2 / C2H6 is 0, the fault type is low-temperature overheating below 150°C. Typical faults caused include overheating of paper-insulated wires; When the coding combination is C2H2 / C2H4=0, CH4 / H2=2, and C2H2 / C2H6=0, the fault type is low-temperature overheating between 150°C and 300°C. When the coding combination is C2H2 / C2H4=0, CH4 / H2=2, and C2H2 / C2H6=1, the fault type is medium-temperature overheating between 300°C and 700°C. When the coding combination is C2H2 / C2H4=0, CH4 / H2=0, 1, or 2, and C2H2 / C2H6=2, the fault type is high-temperature overheating above 700°C. Typical faults caused include poor contact of the tap changer, poor lead connection, poor welding of wire joints, overheating caused by short circuit between strands, multiple grounding of the core, and local short circuit between silicon steel sheets. When the coding combination is C2H2 / C2H4 is 0, CH4 / H2 is 1, and C2H2 / C2H6 is 0, the fault type is partial discharge, and the typical faults caused include discharge with low energy density; When the coding combination is C2H2 / C2H4=2, CH4 / H2=0 or 1, C2H2 / C2H6=0 or 1 or 2, the fault type is low-energy discharge. When the coding combination is C2H2 / C2H4=2, CH4 / H2=2, C2H2 / C2H6=0 or 1 or 2, the fault type is low-energy discharge and overheating. Typical faults caused include spark discharge between different potentials and circulating current between the lead and the cable sleeve or lead shielding tube. When the coding combination is C2H2 / C2H4 is 1, CH4 / H2 is 0 or 1, and C2H2 / C2H6 is 0 or 1 or 2, the fault type is arc discharge. When the coding combination is C2H2 / C2H4 is 1, CH4 / H2 is 2, and C2H2 / C2H6 is 0 or 1 or 2, the fault type is arc discharge and overheating. Typical faults caused include discharge between coil turns and layers, phase flashover, oil gap flashover between tap leads, arcing of selector switch, and discharge of leads to box shell or other grounding bodies.
8. The transformer fault identification method driven by hybrid mechanism data according to claim 5 is characterized in that: In the steps of collecting unknown data samples of the transformer, calculating the distance between the unknown data samples of the transformer and the center point of the training data set, and comparing the calculated distance with the maximum radius of the training data set, an unknown oil chromatogram data sample is collected for state discrimination, and the unknown oil chromatogram data sample is subjected to the same preprocessing operations as the training data set, including data cleaning, feature selection and extraction, and data normalization, and then the distance d between the unknown oil chromatogram data sample and the center point of the training data set is calculated. u ; Let the unknown sample point be x u , the center point of the training data set is x0, then the distance d u The calculation expression is as follows: d u =‖x u -x0‖ Among them, ‖x u -x0‖ represents x u and the L2 norm of x0.
9. A transformer fault identification system driven by a hybrid mechanism and data, characterized in that: include: A training data set construction module is used to construct a training data set based on the historical fault data of the transformer and determine the data center point and maximum radius of the training data set; An unknown data sample range determination module is used to collect unknown data samples of the transformer, calculate the distance between the unknown data samples of the transformer and the data center point of the training data set, and compare the calculated distance with the maximum radius of the training data set. If the calculated distance between the unknown data samples of the transformer and the data center point of the training data set is not greater than the maximum radius of the training data set, then the unknown data samples of the transformer are determined to be within the range of the training data set; otherwise, the unknown data samples of the transformer are determined to be outside the range of the training data set. A hybrid drive fault discrimination module is used to use a pre-trained data-driven model to perform fault discrimination if the transformer unknown data sample is within the range of the training data set to obtain a fault discrimination result; If the unknown data sample of the transformer is outside the range of the training data set, the pre-trained mechanism-driven model is used to perform fault discrimination and obtain the fault discrimination result.
10. The transformer fault identification system driven by hybrid mechanism and data according to claim 9 is characterized in that: The training data set construction module determines the maximum radius of the training data set in the following manner: Calculate the distance d between the center point and each data sample point: d=‖x0-x m ‖m=1,…,M In the formula, x0 is the center point of the training data set, x m is the data sample point in the training data set other than the center point, and M is the total number of data sample points; Calculate the maximum value of the distance d between the center point and each data sample point to obtain the maximum radius of the training data set: <h2 style=";text-align:left;direction:ltr">d<h2 style=";text-align:left;direction:ltr"> max <h2 style=";text-align:left;direction:ltr"> (maxd) 11. The transformer fault identification system driven by mechanism data hybrid according to claim 9 is characterized in that: The hybrid drive fault discrimination module builds a data-driven model based on the Kolmogorov-Arnold network KAN. For any multivariable continuous function f(x1, ..., x n ), are all expressed as a combination of finite unary continuous functions: Where x n is the input data; q,p is an internal function; φ q For external functions; The KAN network consists of multiple layers, including input layer, hidden layer and output layer; the input layer is used to receive the characteristic information of the transformer, the hidden layer extracts and maps the features of the input data through nonlinear transformation, and the output layer performs fault diagnosis based on the output results of the hidden layer; each hidden layer is composed of multiple neurons, each neuron corresponds to an internal function, and the internal functions are combined in a specific way to form an external function, thereby achieving function approximation.
12. The transformer fault identification system driven by hybrid mechanism and data according to claim 11 is characterized in that: During the training process of the KAN network, the hybrid drive fault discrimination module optimizes the parameters of the KAN network. The goal of parameter optimization is to minimize the loss function, which includes a prediction error and a regularization term. The prediction error is used to measure the difference between the network output and the true label, and the regularization term is used to prevent the KAN network from overfitting. During the training process, the gradient of the loss function with respect to the KAN network parameters is calculated through the backpropagation algorithm, and the KAN network parameters are updated according to the gradient.
13. The transformer fault identification system driven by hybrid mechanism and data according to claim 9 is characterized in that: When the hybrid-driven fault discrimination module uses a pre-trained mechanism-driven model for fault discrimination, the mechanism-driven model uses oil chromatography as transformer status data for discrimination and employs a three-ratio method as a mechanism-level discrimination tool. Based on the interdependence between the relative concentrations of gas components generated by the decomposition of oil and insulation in oil-filled electrical equipment under fault conditions and temperature, two gases are selected from characteristic gases according to solubility and diffusion coefficient to form three ratio values, which are represented by different codes. The coding rules and the fault type judgment method are used as the basis for diagnosing the nature of the fault.
14. The transformer fault identification system driven by hybrid mechanism and data according to claim 13 is characterized in that: The unknown data sample range determination module collects an unknown oil chromatogram data sample for state determination, performs the same preprocessing operations on the unknown oil chromatogram data sample as the training data set, including data cleaning, feature selection and extraction, and data normalization, and then calculates the distance d between the unknown oil chromatogram data sample and the center point of the training data set. u ; Let the unknown sample point be x u , the center point of the training data set is x0, then the distance d u The calculation expression is as follows: d u =‖x u -x0‖ Among them, ‖x u -x0‖ represents x u and the L2 norm of x0.
15. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the processor is used to execute a computer program stored in the memory to implement the transformer fault identification method driven by mechanism data hybrid as claimed in any one of claims 1 to 8.
16. A computer-readable storage medium, characterized in that The computer-readable storage medium stores at least one instruction, and when the at least one instruction is executed by a processor, the transformer fault identification method driven by hybrid mechanism data according to any one of claims 1 to 8 is implemented.