Transformer multi-modal data diagnosis method
Through the transformer multimodal data diagnosis method, multiple modal data are collected and fused, and fault diagnosis is performed using multi-core SVM model. The problem of poor robustness of traditional methods when data is missing is solved, achieving higher diagnostic accuracy and robustness.
Patent Information
- Application Number
- CN202411399083.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-09
- Publication Date
- 2025-05-30
AI Technical Summary
Traditional transformer fault diagnosis methods show poor robustness in the face of data loss, especially when the fault phenomenon is fuzzy, the diagnostic accuracy will be reduced.
A multimodal data diagnosis method for transformer is proposed, which uses multiple mode data to collect and fuse data from multiple modes (such as vibration, temperature, oil state, etc.) and uses a multi-core support vector machine (SVM) model to perform fault diagnosis. This method includes steps such as data collection and preprocessing, feature extraction, building multi-core SVM models, training models, model evaluation and fault diagnosis.
Provide more comprehensive fault information through multimodal data fusion, improve diagnosis accuracy and robustness, dynamically adjust the hyperparameters of Gaussian and Sigmoid cores, and improve the adaptability and robustness of the model.
Smart Images
Figure CN120067825A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of transformer fault diagnosis, and particularly relates to a method for diagnosing multi-modal data of a transformer. Background Art
[0002] A transformer is an electrical device used to change the magnitude of alternating voltage in a power system. It transfers electrical energy from one circuit to another through the principle of electromagnetic induction, and can increase or decrease the voltage to meet different power requirements. Transformers are widely used in power transmission and distribution networks. A transformer is a closed whole integrating multi-disciplinary technologies such as mechanics, electricity, chemistry, and thermodynamics, and the reasons affecting its faults are complex.
[0003] Traditional transformer fault diagnosis often uses the IEC three-ratio method. However, the IEC three-ratio method diagnoses faults by analyzing transformer oil-gas data, which has the advantages of being able to detect live and being immune to electromagnetic interference. However, the IEC three-ratio method will lead to a decrease in the diagnostic accuracy when the fault phenomenon is fuzzy, and has certain limitations. These limitations will cause traditional transformer fault diagnosis methods to show poor robustness in the face of data loss. In view of this, a method for diagnosing multi-modal data of a transformer is explored and studied. Summary of the Invention
[0004] The present invention proposes a method for diagnosing multi-modal data of a transformer to solve the technical problems existing in the background art.
[0005] To solve the above problems, the technical solution of the present invention is as follows:
[0006] A method for diagnosing multi-modal data of a transformer includes the following steps:
[0007] (S1) Data collection and preprocessing: Collect multi-modal data of the transformer, preprocess the multi-modal data, and fuse data of different modalities into a unified data format;
[0008] (S2) Feature extraction: Extract key features from the multi-modal data;
[0009] (S3) Construct a multi-core SVM model: First, design a composite kernel function, combine a Gaussian kernel and a Sigmoid kernel to capture the non-linear relationship of the data, use the composite kernel function to train the SVM model, and determine the σ of the Gaussian kernel and the coef0 parameter of the Sigmoid kernel;
[0010] (S4) Train the model: Divide the data set into a training set and a test set, and use the stochastic gradient descent optimization algorithm to adjust the weights of the Gaussian kernel and the Sigmoid kernel;
[0011] (S5)Integrated model evaluation: Evaluate the classification accuracy and performance of the model on the test set, and adjust the proportionality coefficients λ1 and λ2 of the kernel function according to the evaluation results;
[0012] (S6)Fault diagnosis: Apply the trained multi-kernel SVM model to new transformer data for fault diagnosis and generate data diagnosis results.
[0013] Preferably, in the S3, the Gaussian kernel function formula is:
[0014] K_Gaussian(xi,x) = exp(-γ1*||xi - x||^2)
[0015] In the formula, γ1 is the hyperparameter of the Gaussian kernel.
[0016] Preferably, in the S3, the Sigmoid kernel function formula is:
[0017] K_Sigmoid(xi,x) = tanh(γ2*xi^T x + coef0)
[0018] In the formula, γ2 is the hyperparameter of the Sigmoid kernel.
[0019] Preferably, in the S3, the formula of the composite kernel function is:
[0020] Kl*(xi,x) = λ1*K_Gaussian(xi,x) + λ2*K_Sigmoid(xi,x)
[0021] In the formula, λ1 is the proportionality coefficient of the Gaussian kernel, and λ2 is the proportionality coefficient of the Sigmoid kernel.
[0022] Preferably, in the S3, the steps for determining the σ of the Gaussian kernel and the coef0 parameter of the Sigmoid kernel are specifically as follows:
[0023] Step 1: First, define the value ranges for σ and coef0. Among them, set σ to [0.1, 10] and set coef0 to [-5, 5];
[0024] Step 2: Randomly initialize a set of hyperparameter values (σ, coef0);
[0025] Step 3: For the given hyperparameter values, train the model and calculate its accuracy on the cross-validation set;
[0026] Step 4: Use a surrogate model such as a Gaussian process to fit the relationship between the hyperparameter values and the accuracy;
[0027] Step 5: According to the surrogate model, select the parameter combination that is most likely to optimize the objective function value;
[0028] Step 6: Add the new hyperparameter values to the search space and delete the worst hyperparameter values;
[0029] Step 7: Repeat Steps 3 to 6 until the maximum number of iterations is reached or the accuracy does not improve significantly;
[0030] Step 8: Based on the final optimization results, select the hyperparameter values (σ, coef0) that produce the highest accuracy on the cross-validation set.
[0031] Preferably, in S5, when the model is overfitting, increase the values of λ1 and λ2 to reduce the influence of the kernel function, and when the model is underfitting, reduce the values of λ1 and λ2 to increase the influence of the kernel function.
[0032] The above technical solution of the present invention has the following beneficial technical effects: By collecting and fusing multi-modal data, the present invention provides more comprehensive fault information. This multi-modal data fusion can make up for the deficiencies of a single data source and improve the accuracy of diagnosis. Through cross-validation and hyperparameter optimization steps, the hyperparameters of the Gaussian kernel and the Sigmoid kernel can be dynamically adjusted, which can improve the adaptability and robustness of the model. Description of the Drawings
[0033] Figure 1 It is a schematic flow chart of a multi-modal data diagnosis method for a transformer according to the present invention.
[0034] Figure 2 It is a schematic flow chart of constructing a multi-kernel SVM model in a multi-modal data diagnosis method for a transformer according to the present invention. Detailed Embodiments
[0035] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with specific embodiments. It should be understood that these descriptions are exemplary and are not intended to limit the scope of the present invention. In addition, in the following description, the descriptions of well-known structures and technologies are omitted to avoid unnecessarily confusing the concepts of the present invention.
[0036] The following is a further detailed description of the present application in conjunction with the attached Figure 1-2 A multi-modal data diagnosis method for a transformer includes the following steps:
[0037] (S1) Data collection and preprocessing: Collect multi-modal data of the transformer, preprocess the multi-modal data, and fuse data of different modalities into a unified data format;
[0038] (S2) Feature extraction: Extract key features from the multi-modal data;
[0039] (S3) Construct a multi-core SVM model: First, design a composite kernel function by combining the Gaussian kernel and the Sigmoid kernel to capture the non-linear relationship of the data. Then, use the composite kernel function to train the SVM model and determine the σ of the Gaussian kernel and the coef0 parameter of the Sigmoid kernel.
[0040] (S4) Train the model: Divide the data set into a training set and a test set, and use the stochastic gradient descent optimization algorithm to adjust the weights of the Gaussian kernel and the Sigmoid kernel.
[0041] Among them, the stochastic gradient descent is an efficient optimization algorithm that can quickly adjust the weights of the kernel function and improve the training speed of the model.
[0042] (S5) Evaluate the integrated model: Evaluate the classification accuracy and performance of the model on the test set, and adjust the proportionality coefficients λ1 and λ2 of the kernel function according to the evaluation results.
[0043] (S6) Fault diagnosis: Apply the trained multi-core SVM model to new transformer data for fault diagnosis and generate data diagnosis results.
[0044] Among them, the present invention collects and fuses various modalities of data (such as vibration, temperature, oil condition, etc.), thereby providing more comprehensive fault information. This multi-modal data fusion can make up for the deficiencies of a single data source and improve the accuracy and robustness of diagnosis.
[0045] Further, in the S3, the formula of the Gaussian kernel function is:
[0046] K_Gaussian(xi,x) = exp(-γ1 * ||xi - x||^2)
[0047] In the formula, γ1 is the hyperparameter of the Gaussian kernel.
[0048] Further, in the S3, the formula of the Sigmoid kernel function is:
[0049] K_Sigmoid(xi,x) = tanh(γ2 * xi^T x + coef0)
[0050] In the formula, γ2 is the hyperparameter of the Sigmoid kernel.
[0051] Further, in the S3, the formula of the composite kernel function is:
[0052] Kl*(xi,x) = λ1 * K_Gaussian(xi,x) + λ2 * K_Sigmoid(xi,x)
[0053] In the formula, λ1 is the proportionality coefficient of the Gaussian kernel, and λ2 is the proportionality coefficient of the Sigmoid kernel.
[0054] Among them, the composite kernel function combines the Gaussian kernel and the Sigmoid kernel, which can capture the non-linear relationship in the data and improve the diagnostic accuracy of the model. By optimizing the σ of the Gaussian kernel and the coef0 parameter of the Sigmoid kernel, the best combination of hyperparameters can be found to optimize the model performance and enhance the adaptability of the model to different types of data.
[0055] Furthermore, in step S3, the steps for determining the σ of the Gaussian kernel and the coef0 parameter of the Sigmoid kernel are specifically as follows:
[0056] Step 1: First, define the value ranges for σ and coef0. Among them, set σ to [0.1, 10] and set coef0 to [-5, 5].
[0057] Step 2: Randomly initialize a set of hyperparameter values (σ, coef0).
[0058] Step 3: For the given hyperparameter values, train the model and calculate its accuracy on the cross-validation set.
[0059] Step 4: Use a surrogate model such as a Gaussian process to fit the relationship between the hyperparameter values and the accuracy.
[0060] Step 5: According to the surrogate model, select the parameter combination that is most likely to optimize the objective function value.
[0061] Step 6: Add the new hyperparameter values to the search space and delete the worst hyperparameter values.
[0062] Step 7: Repeat steps 3 to 6 until the maximum number of iterations is reached or the accuracy does not increase significantly.
[0063] Step 8: Based on the final optimization result, select the hyperparameter values (σ, coef0) that produce the highest accuracy on the cross-validation set.
[0064] Among them, the σ of the Gaussian kernel and the coef0 parameter of the Sigmoid kernel can be determined through the following example. The specific steps are as follows:
[0065] Step 1: First, define the value ranges, where:
[0066] σ: 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0;
[0067] coef0: -5, -4, -3, -2, -1, 0, 1, 2, 3, 4, 5.
[0068] Step 2: Randomly select a set of hyperparameter values from the defined value ranges. For example: σ = 0.5, coef0 = 1.
[0069] Step 3: Train the model using the given hyperparameter values and calculate its accuracy on the cross-validation set. For example, assume the accuracy is 80%.
[0070] Step 4: Use a surrogate model such as a Gaussian process to fit the relationship between the hyperparameter values and the accuracy.
[0071] Step 5: Based on the surrogate model, select the parameter combination that is most likely to optimize the objective function value. For example, the surrogate model predicts that the accuracy is highest when σ = 0.6 and coef0 = 2.
[0072] Step 6: Add the new hyperparameter values (σ = 0.6, coef0 = 2) to the search space and remove the worst hyperparameter values (e.g., σ = 0.5, coef0 = 1).
[0073] Step 7: Repeat Steps 3 - 6 until the maximum number of iterations is reached or the accuracy does not improve significantly.
[0074] Step 8: Based on the final optimization result, select the hyperparameter values (σ, coef0) that produce the highest accuracy on the cross-validation set. For example, assume that σ = 0.7 and coef0 = 3 are finally selected.
[0075] Through the above Steps 1 - 8, the optimal values of the σ parameter of the Gaussian kernel and the coef0 parameter of the Sigmoid kernel can be determined.
[0076] Furthermore, in S5, when the model is overfitting, increase the values of λ1 and λ2 to reduce the influence of the kernel function. When the model is underfitting, reduce the values of λ1 and λ2 to increase the influence of the kernel function.
[0077] Among them, through cross-validation and hyperparameter optimization steps (such as Gaussian processes), the hyperparameters of the Gaussian kernel and the Sigmoid kernel (such as σ and coef0) are dynamically adjusted, which can improve the adaptability and robustness of the model. For the problem of model overfitting or underfitting, the performance of the model is optimized by adjusting the scaling coefficients (λ1 and λ2) of the kernel function, thereby enhancing the diagnostic ability for transformer faults.
[0078] Specifically, assume that the classification accuracy of the model on the test set is 80%. However, the F1 score is low, indicating that the model performs poorly in terms of recall or precision. After analyzing the evaluation results, it is found that the model is overfitting the training set. To solve this problem, the values of λ1 and λ2 can be increased to reduce the influence of the Gaussian kernel and the Sigmoid kernel. Through cross-validation, it is found that when λ1 = 0.7 and λ2 = 0.3, the F1 score is improved while the classification accuracy still remains at a relatively high level.
[0079] By integrating data of multiple modalities, the data of different modalities can complement each other, reducing the impact of the absence of a single data source on the overall diagnostic performance, thereby enhancing the robustness to data absence.
[0080] The above are only the preferred embodiments of the present invention, which are only used to help understand the method and its core idea of the present application. The protection scope of the present invention is not limited to the above embodiments. All technical solutions falling within the idea of the present invention belong to the protection scope of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements should also be regarded as the protection scope of the present invention. The above is only the preferred embodiment of the present invention and is not used to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A transformer multimodal data diagnosis method, characterized in that: The following steps are involved: (S1) Data collection and preprocessing: Collect multimodal data of the transformer, preprocess the multimodal data, and fuse data of different modes into a unified data format; (S2) Feature extraction: extracting key features from multimodal data; (S3) Construct a multi-kernel SVM model: First, design a composite kernel function that combines the Gaussian kernel and the Sigmoid kernel to capture the nonlinear relationship of the data, use the composite kernel function to train the SVM model, and determine the σ of the Gaussian kernel and the coef0 parameters of the Sigmoid kernel; (S4) Training model: Divide the data set into training set and test set, and use the stochastic gradient descent optimization algorithm to adjust the weights of Gaussian kernel and Sigmoid kernel; (S5) Integrated model evaluation: Evaluate the classification accuracy and performance of the model on the test set, and adjust the proportional coefficients λ1 and λ2 of the kernel function according to the evaluation results; (S6) Fault diagnosis: Apply the trained multi-core SVM model to new transformer data to perform fault diagnosis and generate data diagnosis results.
2. A transformer multimodal data diagnosis method according to claim 1, characterized in that: In S3, the Gaussian kernel function formula is: K_Gaussian(xi,x)=exp(-γ1*||xi-x||^2) Where γ1 is the hyperparameter of the Gaussian kernel.
3. A transformer multimodal data diagnosis method according to claim 1, characterized in that: In S3, the Sigmoid kernel function formula is: K_Sigmoid(xi,x)=tanh(γ2*xi^T x+coef0) Where γ2 is the hyperparameter of the Sigmoid kernel.
4. A transformer multimodal data diagnosis method according to claim 2, characterized in that: In S3, the formula of the composite kernel function is: Kl*(xi,x)=λ1*K_Gaussian(xi,x)+λ2*K_Sigmoid(xi,x) Where λ1 is the scaling factor of the Gaussian kernel, and λ2 is the scaling factor of the Sigmoid kernel.
5. A transformer multimodal data diagnosis method according to claim 1, characterized in that: In S3, the steps of determining the Gaussian kernel σ and the coef0 parameter of the Sigmoid kernel are specifically as follows: Step 1: First define the value range for σ and coef0, where σ is set to [0.1, 10] and coef0 is set to [-5, 5]; Step 2: Randomly initialize a set of hyperparameter values (σ, coef0); Step 3: For a given hyperparameter value, train the model and calculate its accuracy on the cross-validation set; Step 4: Use a surrogate model such as Gaussian process to fit the relationship between hyperparameter values and accuracy; Step 5: According to the surrogate model, select the parameter combination that makes the objective function value most likely to be optimal; Step 6: Add new hyperparameter values to the search space and remove the worst hyperparameter values; Step 7: Repeat steps 3 to 6 until the maximum number of iterations is reached or the accuracy is not significantly improved; Step 8: Based on the final optimization results, select the hyperparameter value (σ, coef0) that produces the highest accuracy on the cross-validation set.
6. A transformer multimodal data diagnosis method according to claim 1, characterized in that: In S5, when the model is overfitting, the values of λ1 and λ2 are increased to reduce the influence of the kernel function, and when the model is underfitting, the values of λ1 and λ2 are reduced to increase the influence of the kernel function.