Transformer current rising test data analysis method based on deep learning

Abstract

The invention discloses a transformer current rising test data analysis method based on deep learning, and relates to the technical field of power transformation equipment, and the method comprises the following steps: S1, data collection and preprocessing; s2, performing feature engineering; s3, dividing a data set; s4, model selection and architecture design; s5, model training; s6, performing model verification and hyper-parameter tuning; s7, testing and evaluating the model; s8, performing anomaly detection and adaptive learning; s9, result interpretation and model interpretability; and S10, continuously monitoring and updating the model. According to the transformer current rising test data analysis method based on deep learning, based on deep learning, in combination with technologies of feature engineering, model selection and architecture design, anomaly detection and adaptive learning and the like, the accuracy and robustness of a transformer current rising test are improved through comprehensive data analysis; the generalization ability of the model for unseen data can be improved, and the model can adapt to environmental changes in real time.

Description

Technical Field

[0001] The present invention relates to the technical field of power transformation equipment, and in particular to a transformer current boost test data analysis method based on deep learning. Background Art

[0002] Transformer current boost test is an important part of transformer performance evaluation, which aims to detect its stability and safety under high current load. Traditional data analysis methods mostly rely on manual experience and often cannot effectively capture complex nonlinear relationships and time series dynamic changes.

[0003] Existing technologies have certain deficiencies in data collection, processing, and feature extraction, and are unable to fully utilize the potential of deep learning.

[0004] In addition, the lack of anomaly detection mechanism leads to poor sensitivity of the model to abnormal data, affecting the accuracy of prediction. Although some methods try to introduce online learning mechanisms, they lack systematic adaptive capabilities and are difficult to adapt to changing environments and real-time data. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention provides a transformer current boost test data analysis method based on deep learning to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: a transformer current boost test data analysis method based on deep learning, comprising the following steps: S1, data collection and preprocessing; Collect data from transformer current boost tests through different data sources and pre-process the data to improve the accuracy and robustness of subsequent deep learning models; S2, feature engineering; Through the method of feature selection and feature extraction, the feature generation model based on feature engineering combines multiple features to capture the complex relationship of transformer operating status and selects the features that have a significant impact on transformer performance; S3, data set division; The preprocessed data set is divided into training set, validation set and test set, with a ratio of 70% for training, 15% for validation and 15% for testing; the validation set is used for parameter adjustment and the test set is used for evaluating the final model performance; S4, model selection and architecture design; Based on the transformer current boost test data, select the appropriate deep learning model and determine the number of network layers, the number of neurons in each layer, and the hyperparameters of the activation function in the architecture design; S5, model training; Use the training set data to train the selected model and optimize the network parameters through the back propagation algorithm; S6. Model validation and hyperparameter tuning; After the model training is completed, the model is evaluated using the validation set, and the hyperparameters are adjusted based on the evaluation results; S7, model testing and evaluation; After the model hyperparameters are tuned, the model is evaluated on the test set, and the confusion matrix is ​​used to analyze the classification performance of the model, identify possible deviations and weaknesses of the model, and generate ROC curves and AUC value indicators based on the test results; S8, anomaly detection and adaptive learning; Based on the anomaly detection mechanism, unsupervised learning methods are used to identify and process abnormal samples; an adaptive learning mechanism is established to allow the model to learn online when new data arrives and update model parameters to adapt to the changing environment; S9, result interpretation and model interpretability; Explain the results in deep learning models, use SHAP values ​​or LIME methods to analyze the model's decision-making process and understand the contribution of features to the final prediction results; S10, continuous monitoring and model updating; Based on the effectiveness of deep learning models in practical applications, a continuous monitoring mechanism is established to track the performance of the model in real time.

[0007] To further optimize the technical solution, in step S1: Data sources include sensors, monitoring devices, and historical records; The data includes multiple dimensions including current, voltage, temperature, and vibration; Preprocessing includes cleaning, denoising and normalizing the data; At the same time, a denoising autoencoder is used to eliminate measurement noise and standardize the data to ensure that data of different dimensions does not affect the learning effect of the model during training.

[0008] To further optimize the technical solution, in step S2, the feature generation model is as follows: ; in, To generate new features; is the current; is the voltage; is temperature; is the frequency of the signal; is the rate of change of current; is an adjustable weight coefficient; Sine function Used to simulate the periodic changes of current, based on the frequency response characteristics existing in the operation of the transformer, to capture the transient characteristics and oscillation modes; Logarithmic rate of change It is used to capture the nonlinear characteristics of current changes and reflect the dynamic behavior of the system. The sharp changes in current are used to indicate potential fault risks. The product of temperature and current Used to reflect the impact of temperature on current performance and reveal how thermal effects affect transformer efficiency and safety; The influence of each feature is determined by the weight coefficient Adjustments are made, determined through data-driven methods such as linear regression or training of deep learning models, so that the new features better reflect the actual performance of the transformer.

[0009] To further optimize the technical solution, in step S3: When partitioning a dataset, make sure each subset represents the characteristics of the overall data; Use stratified sampling to ensure that different types of data are reasonably distributed in each subset; During stratified sampling, the randomness of the data after division is ensured to prevent overfitting.

[0010] To further optimize the technical solution, in step S4, the deep learning model combines a convolutional neural network (CNN) and a long short-term memory network (LSTM) to capture the spatial and temporal characteristics of time series data. The deep learning model architecture formula is as follows: ; in, is the model output, such as the state prediction of the transformer; The timing characteristics of input data, such as current, voltage, and temperature; is the weight parameter of the convolution layer; is the weight parameter of the LSTM layer; is the weight parameter of the output layer; is the activation function, such as ReLU or Sigmoid; In deep learning models, Input layer: input data Contains time series features such as current, voltage, and temperature. Data is preprocessed to ensure that it is suitable for the input format of CNN and LSTM. Convolutional Neural Network (CNN): The convolution layer is used to extract local features in the data, set up multi-layer convolution operations, and use convolution kernels Convolve the input data to generate feature maps. Indicates that the input data is processed through the convolution layer , get the feature representation; Long Short-Term Memory (LSTM): Apply LSTM layers to the feature maps output by the convolutional layers, using memory to capture long-term dependencies in time series. It means that based on the convolutional features, the data is further processed to capture the temporal dynamics; Output layer: The output of LSTM passes through a fully connected layer Processing to obtain the final prediction result , set the activation function according to the task requirements, such as Sigmoid for binary classification and Softmax for multi-classification.

[0011] To further optimize the technical solution, in step S5: During training, set appropriate hyperparameters for learning rate, batch size, and number of training rounds; Use the learning rate scheduling strategy to dynamically adjust the learning rate based on the performance of the validation set to improve training efficiency; Prevent overfitting based on early stopping method, so that the model can perform well on the training set but degrade on the validation set. During the training process, the loss value and accuracy of each epoch are recorded for subsequent analysis.

[0012] To further optimize the technical solution, in step S6, a grid search or Bayesian optimization method is used to find the best hyperparameter combination, and the verification process focuses on multiple evaluation indicators, such as accuracy, recall, and F1 score, to ensure that the model performs evenly in different aspects; At the same time, the stability and generalization ability of the model are verified based on the cross-validation method.

[0013] To further optimize the technical solution, in step S7, the confusion matrix is ​​as follows: ; The structure of the confusion matrix is: : True examples, the number of samples correctly predicted by the model as positive; : False positive examples, the number of samples that the model incorrectly predicts as positive; : True negative examples, the number of samples correctly predicted by the model as negative; : False negative examples, the number of samples that the model incorrectly predicts as negative.

[0014] To further optimize the technical solution, in step S8, the abnormality detection mechanism is as follows: ; in, Indicates abnormal mark, 1 is abnormal, 0 is normal; is the input sample; is the mean vector of the sample, calculated by clustering or historical data; is the distance between the sample and the mean vector, using Euclidean distance or Mahalanobis distance; is the set threshold used to determine the limit of abnormality; This mechanism determines whether a sample is abnormal by calculating the distance between each input sample and the mean vector; When the distance of the sample Exceeding the set threshold When , the sample is marked as abnormal , otherwise marked as normal .

[0015] To further optimize the technical solution, in step S8, the adaptive learning mechanism is as follows: ; in, and are the updated model parameters and the current model parameters respectively; is the learning rate; is the true label; is the model prediction value; For input sample The gradient of This mechanism enables the model to update model parameters in real time when new data arrives, adapting to environmental changes; By calculating the error between the true label and the predicted value and combining it with the gradient of the input sample, the model parameters are dynamically adjusted. .

[0016] Compared with the prior art, the present invention provides a transformer current boost test data analysis method based on deep learning, which has the following beneficial effects: This deep learning-based transformer current boost test data analysis method aims to improve the accuracy and robustness of the transformer current boost test through comprehensive data analysis, by combining deep learning with feature engineering, model selection and architecture design, anomaly detection and adaptive learning. It can not only improve the model's generalization ability for unseen data, but also adapt to environmental changes in real time to ensure its effectiveness in practical applications, thereby providing more solid technical support for the safe operation of transformers. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A schematic diagram of a flow chart of a transformer current boost test data analysis method based on deep learning proposed by the present invention; Figure 2 A schematic diagram of the flow of data collection and preprocessing in a transformer current boost test data analysis method based on deep learning proposed by the present invention; Figure 3 A schematic flow chart of a feature generation model in a transformer current boost test data analysis method based on deep learning proposed by the present invention; Figure 4 A schematic flow chart of a deep learning model in a transformer current boosting test data analysis method based on deep learning proposed by the present invention; Figure 5 A flow chart of a confusion matrix in a transformer current boosting test data analysis method based on deep learning proposed by the present invention; Figure 6 This is a flow chart of step S8 in a transformer current boost test data analysis method based on deep learning proposed in the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] Example: See also Figure 1 , a transformer current boost test data analysis method based on deep learning, comprising the following steps: S1. Data collection and preprocessing Data from transformer current boost tests are collected through different data sources and preprocessed to improve the accuracy and robustness of subsequent deep learning models.

[0020] like Figure 2As shown, in this embodiment, in step S1: Data sources include sensors, monitoring devices, and historical records; The data includes multiple dimensions including current, voltage, temperature, and vibration; Preprocessing includes cleaning, denoising and normalizing the data; At the same time, a denoising autoencoder is used to eliminate measurement noise and standardize the data to ensure that data of different dimensions does not affect the learning effect of the model during training.

[0021] S2. Feature Engineering Through the methods of feature selection and feature extraction, a feature generation model based on feature engineering is proposed, which combines multiple features to capture the complex relationship of the transformer operating status and selects the features that have an important impact on the transformer performance.

[0022] In this embodiment, in step S2, the feature generation model is as follows: ; in, To generate new features; is the current; is the voltage; is temperature; is the frequency of the signal; is the rate of change of current; is an adjustable weight coefficient; Sine function Used to simulate the periodic changes of current, based on the frequency response characteristics existing in the operation of the transformer, to capture the transient characteristics and oscillation modes; Logarithmic rate of change It is used to capture the nonlinear characteristics of current changes and reflect the dynamic behavior of the system. The sharp changes in current are used to indicate potential fault risks. The product of temperature and current Used to reflect the impact of temperature on current performance and reveal how thermal effects affect transformer efficiency and safety; The influence of each feature is determined by the weight coefficient Adjustments are made, determined through data-driven methods such as linear regression or training of deep learning models, so that the new features better reflect the actual performance of the transformer.

[0023] like Figure 3 As shown, when using the model: Feature calculation: In the data preprocessing stage, the required basic features are first extracted, such as current, voltage, temperature and their rate of change. Then, the set formula is used to calculate .

[0024] Model training: The new features calculated As one of the input features, it is input into the deep learning model together with other original features for training. Through model training, the weight coefficient is adjusted to optimize the influence of the feature.

[0025] Model validation and evaluation: When using the validation set for model evaluation, you can focus on observing new features Impact on model performance. If the feature significantly improves the accuracy and robustness of the model, further hyperparameter tuning can be performed to ensure the best model configuration.

[0026] Feature analysis: After model training and validation, use methods such as SHAP value to analyze The importance of predictive modeling in the model decision-making process is to verify whether it can indeed improve the interpretability and effectiveness of the model.

[0027] It provides a new perspective for data analysis in transformer current-increasing tests and helps to more comprehensively understand the various factors that affect transformer performance.

[0028] Principal component analysis (PCA) can also be used to reduce feature dimensions while retaining most of the information. Time domain and frequency domain feature extraction techniques can also be used to extract the instantaneous and frequency characteristics of the signal. In addition, based on domain knowledge, some derived features can be constructed, such as the relationship between temperature and current, to further improve the performance of the model.

[0029] S3, Dataset Division The preprocessed dataset is divided into training set, validation set and test set, with the ratio of 70% for training, 15% for validation and 15% for testing. The validation set is used for parameter adjustment and the test set is used for evaluating the final model performance.

[0030] In this embodiment, in step S3: When partitioning a dataset, make sure each subset represents the characteristics of the overall data; Use stratified sampling to ensure that different types of data are reasonably distributed in each subset; When stratified sampling is performed, the randomness of the data after division is ensured to prevent overfitting. S4. Model selection and architecture design Based on the transformer current boost test data, select a suitable deep learning model, and in the architecture design, determine the number of network layers, the number of neurons in each layer, and the hyperparameters of the activation function.

[0031] In this embodiment, in step S4, the deep learning model combines a convolutional neural network (CNN) and a long short-term memory network (LSTM) to capture the spatial and temporal characteristics of time series data. The deep learning model architecture formula is as follows: ; in, is the model output, such as the state prediction of the transformer; The timing characteristics of input data, such as current, voltage, and temperature; is the weight parameter of the convolution layer; is the weight parameter of the LSTM layer; is the weight parameter of the output layer; is the activation function, such as ReLU or Sigmoid; In deep learning models, Input layer: input data Contains time series features such as current, voltage, and temperature. The data is preprocessed to ensure that it is in an input format suitable for CNN and LSTM.

[0032] Convolutional Neural Network (CNN): The convolution layer is used to extract local features in the data, set up multi-layer convolution operations, and use convolution kernels Convolve the input data to generate feature maps. Indicates that the input data is processed through the convolution layer , and obtain feature representation; the advantage of using the convolutional layer is that it can effectively extract local features and reduce the data dimension, reducing the complexity of subsequent calculations.

[0033] Long Short-Term Memory (LSTM): Apply LSTM layers to the feature maps output by the convolutional layers, using memory to capture long-term dependencies in time series. It represents that based on the convolutional features, the data is further processed to capture the time series dynamics; the introduction of LSTM enables the model to understand the dynamic changes in time series data, especially the mutations and patterns in the transformer ramp-up process.

[0034] Output layer: The output of LSTM passes through a fully connected layer Processing to obtain the final prediction result , set the activation function according to the task requirements, such as Sigmoid for binary classification and Softmax for multi-classification, to ensure that the model can output results that meet the actual application requirements.

[0035] You can also consider using techniques such as skip connections to enhance the model's learning ability. In addition, the complexity of the model needs to be reasonably controlled according to the amount of data to prevent the model from overfitting.

[0036] like Figure 4 As shown, the model, when used: Model construction: When designing a model, determine hyperparameters such as the number of convolutional layers and LSTM layers, the number of neurons in each layer, and the size of the convolution kernel according to specific task requirements (such as classification or regression). These parameters can be tuned through experiments to achieve optimal performance.

[0037] Model training: Use the preprocessed data set to train the constructed model. During the training process, the weight parameters are optimized through the back propagation algorithm. , and . An adaptive learning rate optimization algorithm such as Adam can be used to speed up the training process.

[0038] Model validation and evaluation: Use the validation set to evaluate the model, and evaluate the generalization ability of the model by monitoring indicators such as loss function and accuracy. At the same time, monitor the loss of training and validation to avoid overfitting.

[0039] Model improvement: After evaluating the model, make necessary hyperparameter adjustments based on the validation results to optimize the model structure. For example, you can try increasing or decreasing the number of convolutional layers, adjusting the number of LSTM units, or introducing skip connections to enhance the model’s learning ability.

[0040] Results Analysis: By analyzing the model’s prediction results, use interpretability tools such as SHAP or LIME to gain insight into the model’s decision-making process and verify its practical applicability to transformer performance.

[0041] By combining the advantages of CNN and LSTM, powerful feature extraction and time series modeling capabilities are provided for transformer current boost test data analysis, which can more effectively capture the complex characteristics of the data.

[0042] S5. Model training The selected model is trained using the training set data, and the network parameters are optimized through the back-propagation algorithm.

[0043] In this embodiment, in step S5: During training, set appropriate hyperparameters for learning rate, batch size, and number of training rounds; Use the learning rate scheduling strategy to dynamically adjust the learning rate based on the performance of the validation set to improve training efficiency; Prevent overfitting based on early stopping method, so that the model can perform well on the training set but degrade on the validation set. During the training process, the loss value and accuracy of each epoch are recorded for subsequent analysis.

[0044] S6. Model validation and hyperparameter tuning After the model training is completed, the validation set is used to evaluate the model and the hyperparameters are adjusted based on the evaluation results.

[0045] In this embodiment, in step S6, a grid search or Bayesian optimization method is used to find the best hyperparameter combination, and the verification process focuses on multiple evaluation indicators, such as accuracy, recall, and F1 score, to ensure that the model performs evenly in different aspects; At the same time, the stability and generalization ability of the model are verified based on the cross-validation method.

[0046] S7. Model testing and evaluation After the model hyperparameters are tuned, they are evaluated on the test set, and the confusion matrix is ​​used to analyze the classification performance of the model, identify possible deviations and weaknesses of the model, and generate ROC curves and AUC value indicators based on the test results.

[0047] In this embodiment, in step S7, the confusion matrix is ​​as follows: ; The structure of the confusion matrix is: : True examples, the number of samples correctly predicted by the model as positive; : False positive examples, the number of samples that the model incorrectly predicts as positive; : True negative examples, the number of samples correctly predicted by the model as negative; : False negative examples, the number of samples that the model incorrectly predicts as negative.

[0048] In this matrix, assuming a specific three-category classification problem (such as normal, overload, and failure), by organizing the prediction results of each category into a confusion matrix, you can intuitively understand the performance of the model on different categories.

[0049] This matrix not only records the TP, FP, and FN of each class, but also provides a clearer visualization for multi-class classification through the interaction of rows and columns.

[0050] For example, the first row could represent the predicted situation for the “normal” class, the second row for “overload”, and the third row for “failure”.

[0051] Several evaluation metrics can be derived from the confusion matrix: Accuracy:

[0052] Precision:

[0053] Recall:

[0054] F1-score:

[0055] These metrics can provide a more comprehensive perspective on the overall performance of the model, especially in the case of class imbalance.

[0056] like Figure 5 As shown, this matrix is ​​used: Model evaluation: In the model testing phase, the test set is used for prediction and a confusion matrix is ​​generated. The confusion matrix can be calculated using the confusion_matrix function in the sklearn library in Python.

[0057] Visualize the confusion matrix: Visualize the generated confusion matrix for intuitive understanding. You can use heatmaps and other methods to display the prediction effects of various categories to help identify confused categories.

[0058] Calculate evaluation indicators: Calculate various evaluation indicators (accuracy, precision, recall, F1-score) based on the confusion matrix and compare them. This process can reveal the strengths and weaknesses of the model in different categories, especially in critical fault detection tasks.

[0059] Performance analysis: By analyzing the confusion matrix and its derived metrics, we can identify possible biases in the model. For example, if the model has a high false negative rate (FN) on the “fault” class, we may need to adjust the model to improve the accuracy of fault detection.

[0060] Results summary and improvement: Based on the analysis results, summarize the model and identify areas for improvement. If the performance of some categories is clearly insufficient, consider adding relevant data or experimenting with different model architectures.

[0061] The confusion matrix can provide a more detailed and comprehensive evaluation perspective in the classification task of transformer current boost test data, ensuring the effectiveness of the model in practical applications.

[0062] S8. Anomaly Detection and Adaptive Learning Based on the anomaly detection mechanism, unsupervised learning methods are used to identify and process abnormal samples; an adaptive learning mechanism is established to allow the model to perform online learning when new data arrives and update model parameters to adapt to the ever-changing environment.

[0063] In this embodiment, in step S8, the abnormality detection mechanism is as follows: ; in, Indicates abnormal mark, 1 is abnormal, 0 is normal; is the input sample; is the mean vector of the sample, calculated by clustering or historical data; is the distance between the sample and the mean vector, using Euclidean distance or Mahalanobis distance; is the set threshold used to determine the limit of abnormality; This mechanism determines whether a sample is abnormal by calculating the distance between each input sample and the mean vector; When the distance of the sample Exceeding the set threshold When , the sample is marked as abnormal , otherwise marked as normal .

[0064] This mechanism can help timely identify abnormal conditions that may occur during current ramp-up tests, such as sudden changes in current or temperature, which may indicate potential equipment failures.

[0065] In step S8, the adaptive learning mechanism is as follows: ; in, and are the updated model parameters and the current model parameters respectively; is the learning rate; is the true label; is the model prediction value; For input sample The gradient of This mechanism enables the model to update model parameters in real time when new data arrives, adapting to environmental changes; By calculating the error between the true label and the predicted value and combining it with the gradient of the input sample, the model parameters are dynamically adjusted. .

[0066] This mechanism ensures that the model can adjust itself according to actual conditions, thereby improving its adaptability in new environments.

[0067] like Figure 6 As shown, the two mechanisms are in use: Data preprocessing and clustering: Before applying the model, perform cluster analysis on historical data and calculate the mean vector and distance threshold The clustering method can use unsupervised learning methods such as K-means or DBSCAN to ensure that the mean can represent the characteristics of normal samples.

[0068] Anomaly detection: When new data arrives, the anomaly detection mechanism described above is used to calculate the anomaly marker for each sample. If abnormal samples are found, they can be processed separately to avoid negative impact on model training.

[0069] Online learning: For normal samples, the model parameters are updated in real time according to the adaptive learning mechanism. As new data continues to arrive, the model can be gradually adjusted to adapt to the new operating state. This process can be achieved through small batch online learning methods.

[0070] Model evaluation and monitoring: Regularly evaluate the performance of the model on new data sets, and ensure that the model can continue to adapt to environmental changes by monitoring the proportion of abnormal samples and the prediction accuracy of the model. If the model performance is found to be degraded, consider retraining or adjusting parameters.

[0071] Feedback mechanism: The detected abnormal data and its corresponding processing results are fed back to the model, and this information is used to further optimize the model's anomaly detection and adaptive learning capabilities.

[0072] It can provide continuous monitoring and real-time adjustment capabilities for data analysis in transformer current-raising tests, thereby ensuring the effectiveness and reliability of the model in practical applications.

[0073] S9. Results interpretation and model interpretability Explain the results in deep learning models, use SHAP values ​​or LIME methods to analyze the model's decision-making process and understand the contribution of features to the final prediction results.

[0074] In this example, in deep learning models, the interpretability of results is crucial, especially in industrial applications. This not only helps verify the reliability of the model, but also helps technicians make decisions in subsequent work. At the same time, comparing the model's prediction results with actual monitoring data can help identify differences in the model's performance under specific conditions and provide directions for improvement. S10. Continuous monitoring and model updating Based on the effectiveness of deep learning models in practical applications, a continuous monitoring mechanism is established to track the performance of the model in real time.

[0075] In this embodiment, it includes collecting new data, regularly evaluating model performance, and updating the model based on actual conditions. During the data collection phase, the data flow can be automated to ensure that the model always receives the latest input. In addition, thresholds should be set to trigger automatic retraining or model update mechanisms when the performance of the model deteriorates within a specific time period. This process will ensure that the deep learning model can adapt to various changes that may occur during the transformer current increase test and maintain efficient prediction capabilities.

[0076] The beneficial effects of the present invention are: This deep learning-based transformer current boost test data analysis method aims to improve the accuracy and robustness of the transformer current boost test through comprehensive data analysis, by combining deep learning with feature engineering, model selection and architecture design, anomaly detection and adaptive learning. It can not only improve the model's generalization ability for unseen data, but also adapt to environmental changes in real time to ensure its effectiveness in practical applications, thereby providing more solid technical support for the safe operation of transformers.

[0077] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0078] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A transformer current boost test data analysis method based on deep learning, characterized in that: The following steps are involved: S1, data collection and preprocessing; Collect data from transformer current boost tests through different data sources and pre-process the data to improve the accuracy and robustness of subsequent deep learning models; S2, feature engineering; Through the method of feature selection and feature extraction, the feature generation model based on feature engineering combines multiple features to capture the complex relationship of transformer operating status and selects the features that have a significant impact on transformer performance; S3, data set division; The preprocessed data set is divided into training set, validation set and test set, with a ratio of 70% for training, 15% for validation and 15% for testing; the validation set is used for parameter adjustment and the test set is used for evaluating the final model performance; S4, model selection and architecture design; Based on the transformer current boost test data, select the appropriate deep learning model and determine the number of network layers, the number of neurons in each layer, and the hyperparameters of the activation function in the architecture design; S5, model training; Use the training set data to train the selected model and optimize the network parameters through the back propagation algorithm; S6. Model validation and hyperparameter tuning; After the model training is completed, the model is evaluated using the validation set, and the hyperparameters are adjusted based on the evaluation results; S7, model testing and evaluation; After the model hyperparameters are tuned, the model is evaluated on the test set, and the confusion matrix is ​​used to analyze the classification performance of the model, identify possible deviations and weaknesses of the model, and generate ROC curves and AUC value indicators based on the test results; S8, anomaly detection and adaptive learning; Based on the anomaly detection mechanism, unsupervised learning methods are used to identify and process abnormal samples; an adaptive learning mechanism is established to allow the model to learn online when new data arrives and update model parameters to adapt to the changing environment; S9, result interpretation and model interpretability; Explain the results in deep learning models, use SHAP values ​​or LIME methods to analyze the model's decision-making process and understand the contribution of features to the final prediction results; S10, continuous monitoring and model updating; Based on the effectiveness of deep learning models in practical applications, a continuous monitoring mechanism is established to track the performance of the model in real time.

2. A transformer current boost test data analysis method based on deep learning according to claim 1, characterized in that: In step S1: Data sources include sensors, monitoring devices, and historical records; The data includes multiple dimensions including current, voltage, temperature, and vibration; Preprocessing includes cleaning, denoising and normalizing the data; At the same time, a denoising autoencoder is used to eliminate measurement noise and standardize the data to ensure that data of different dimensions does not affect the learning effect of the model during training.

3. A transformer current boost test data analysis method based on deep learning according to claim 1, characterized in that: In step S2, the feature generation model is as follows: ； in, To generate new features; is the current; is the voltage; is temperature; is the frequency of the signal; is the rate of change of current; is an adjustable weight coefficient; Sine function Used to simulate the periodic changes of current, based on the frequency response characteristics existing in the operation of the transformer, to capture the transient characteristics and oscillation modes; Logarithmic rate of change It is used to capture the nonlinear characteristics of current changes and reflect the dynamic behavior of the system. The sharp changes in current are used to indicate potential fault risks. The product of temperature and current Used to reflect the impact of temperature on current performance and reveal how thermal effects affect transformer efficiency and safety; The influence of each feature is determined by the weight coefficient Adjustments are made, determined through data-driven methods such as linear regression or training of deep learning models, so that the new features better reflect the actual performance of the transformer.

4. The transformer current boost test data analysis method based on deep learning according to claim 1 is characterized in that: In step S3: When partitioning a dataset, make sure each subset represents the characteristics of the overall data; Use stratified sampling to ensure that different types of data are reasonably distributed in each subset; During stratified sampling, the randomness of the data after division is ensured to prevent overfitting.

5. The transformer current boost test data analysis method based on deep learning according to claim 1 is characterized in that: In step S4, the deep learning model combines a convolutional neural network (CNN) and a long short-term memory network (LSTM) to capture the spatial and temporal characteristics of time series data. The deep learning model architecture formula is as follows: ； in, is the model output, such as the state prediction of the transformer; The timing characteristics of input data, such as current, voltage, and temperature; is the weight parameter of the convolution layer; is the weight parameter of the LSTM layer; is the weight parameter of the output layer; is the activation function, such as ReLU or Sigmoid; In deep learning models, Input layer: input data Contains time series features such as current, voltage, and temperature. Data is preprocessed to ensure that it is suitable for the input format of CNN and LSTM. Convolutional Neural Network (CNN): The convolution layer is used to extract local features in the data, set up multi-layer convolution operations, and use convolution kernels Convolve the input data to generate feature maps. Indicates that the input data is processed through the convolution layer , get the feature representation; Long Short-Term Memory (LSTM): Apply LSTM layers to the feature maps output by the convolutional layers, using memory to capture long-term dependencies in time series. It means that based on the convolutional features, the data is further processed to capture the temporal dynamics; Output layer: The output of LSTM passes through a fully connected layer Processing to obtain the final prediction result , set the activation function according to the task requirements, such as Sigmoid for binary classification and Softmax for multi-classification.

6. The transformer current boost test data analysis method based on deep learning according to claim 1 is characterized in that: In step S5: During training, set appropriate hyperparameters for learning rate, batch size, and number of training rounds; Use the learning rate scheduling strategy to dynamically adjust the learning rate based on the performance of the validation set to improve training efficiency; Prevent overfitting based on early stopping method, so that the model can perform well on the training set but degrade on the validation set. During the training process, the loss value and accuracy of each epoch are recorded for subsequent analysis.

7. The transformer current boost test data analysis method based on deep learning according to claim 1 is characterized in that: In step S6, a grid search or Bayesian optimization method is used to find the best hyperparameter combination, and the verification process focuses on multiple evaluation indicators, such as accuracy, recall, and F1 score, to ensure that the model performs evenly in different aspects; At the same time, the stability and generalization ability of the model are verified based on the cross-validation method.

8. The transformer current boost test data analysis method based on deep learning according to claim 1 is characterized in that: In step S7, the confusion matrix is ​​as follows: ； The structure of the confusion matrix is: : True examples, the number of samples correctly predicted by the model as positive; : False positive examples, the number of samples that the model incorrectly predicts as positive; : True negative examples, the number of samples correctly predicted by the model as negative; : False negative examples, the number of samples that the model incorrectly predicts as negative.

9. The transformer current boost test data analysis method based on deep learning according to claim 1 is characterized in that: In step S8, the abnormality detection mechanism is as follows: ； in, Indicates abnormal mark, 1 is abnormal, 0 is normal; is the input sample; is the mean vector of the sample, calculated by clustering or historical data; is the distance between the sample and the mean vector, using Euclidean distance or Mahalanobis distance; is the set threshold used to determine the limit of abnormality; This mechanism determines whether a sample is abnormal by calculating the distance between each input sample and the mean vector; When the distance of the sample Exceeding the set threshold When , the sample is marked as abnormal , otherwise marked as normal .

10. The transformer current boost test data analysis method based on deep learning according to claim 1, characterized in that: In step S8, the adaptive learning mechanism is as follows: ； in, and are the updated model parameters and the current model parameters respectively; is the learning rate; is the true label; is the model prediction value; For input sample The gradient of This mechanism enables the model to update model parameters in real time when new data arrives, adapting to environmental changes; By calculating the error between the true label and the predicted value and combining it with the gradient of the input sample, the model parameters are dynamically adjusted. .

Images