A method for analyzing transformer current boost test data based on deep learning

Abstract

This invention discloses a deep learning-based method for analyzing transformer current-boosting test data, relating to the field of power equipment technology. The method includes the following steps: S1, data collection and preprocessing; S2, feature engineering; S3, dataset partitioning; S4, model selection and architecture design; S5, model training; S6, model validation and hyperparameter tuning; S7, model testing and evaluation; S8, anomaly detection and adaptive learning; S9, result interpretation and model interpretability; and S10, continuous monitoring and model updating. This deep learning-based method for analyzing transformer current-boosting test data, by combining deep learning with features engineering, model selection and architecture design, anomaly detection, and adaptive learning techniques, aims to improve the accuracy and robustness of transformer current-boosting tests through comprehensive data analysis. It not only enhances the model's generalization ability to unseen data but also allows it to adapt to environmental changes in real time.

Description

Technical Field

[0001] This invention relates to the field of power equipment technology, specifically to a method for analyzing transformer current boost test data based on deep learning. Background Technology

[0002] Transformer current boosting tests are a crucial step in evaluating transformer performance, aiming to test its stability and safety under high current loads. Traditional data analysis methods rely heavily on human experience and often fail to effectively capture complex nonlinear relationships and time-series dynamic changes.

[0003] Existing technologies have certain shortcomings in data acquisition, processing, and feature extraction, and cannot fully utilize the potential of deep learning.

[0004] Furthermore, the lack of anomaly detection mechanisms results in poor model sensitivity to anomalous data, affecting prediction accuracy. Although some methods attempt to introduce online learning mechanisms, they lack systematic adaptability and struggle to adapt to constantly changing environments and real-time data. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a deep learning-based method for analyzing transformer current-boosting test data, thereby solving the problems mentioned in the background section.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for analyzing transformer current-boosting test data based on deep learning, comprising the following steps:

[0007] S1. Data collection and preprocessing;

[0008] Data from transformer current-boosting tests were collected from different data sources and preprocessed to improve the accuracy and robustness of subsequent deep learning models.

[0009] S2, Feature Engineering;

[0010] By employing feature selection and feature extraction methods, and based on a feature generation model derived from feature engineering, multiple features are combined to capture the complex relationships of transformer operating states, thereby selecting features that have a significant impact on transformer performance.

[0011] S3. Dataset partitioning;

[0012] The preprocessed dataset is divided into a training set, a validation set, and a test set, with a ratio of 70% for training, 15% for validation, and 15% for testing; the validation set is used for parameter tuning, and the test set is used for evaluating the final model performance.

[0013] S4. Model selection and architecture design;

[0014] Based on transformer current boosting test data, a suitable deep learning model is selected, and in terms of architecture design, the number of network layers, the number of neurons in each layer, and the hyperparameters of the activation function are determined.

[0015] S5, Model Training;

[0016] The selected model is trained using the training set data, and the network parameters are optimized using the backpropagation algorithm.

[0017] S6. Model validation and hyperparameter tuning;

[0018] After the model is trained, the model is evaluated using a validation set, and the hyperparameters are adjusted based on the evaluation results.

[0019] S7. Model Testing and Evaluation;

[0020] After the model hyperparameters are tuned, they are evaluated on the test set. The confusion matrix is ​​used to analyze the classification performance of the model, identify possible biases and weaknesses of the model, and generate ROC curves and AUC values ​​based on the test results.

[0021] S8, Anomaly Detection and Adaptive Learning;

[0022] Based on the anomaly detection mechanism, an unsupervised learning method is used to identify and process anomalous samples; an adaptive learning mechanism is established to allow the model to learn online and update model parameters when new data arrives, so as to adapt to the constantly changing environment.

[0023] S9. Results Interpretation and Model Interpretability;

[0024] Interpreting results in deep learning models involves using SHAP values ​​or LIME methods to analyze the model's decision-making process and understand the contribution of features to the final prediction results.

[0025] S10, Continuous monitoring and model updates;

[0026] Based on the effectiveness of deep learning models in practical applications, a continuous monitoring mechanism is established to track the model's performance in real time.

[0027] To further optimize this technical solution, in step S1:

[0028] Data sources include sensors, monitoring devices, and historical records;

[0029] The data includes multiple dimensions such as current, voltage, temperature, and vibration.

[0030] Preprocessing includes cleaning, denoising, and normalizing the data;

[0031] Meanwhile, a denoising autoencoder is used to eliminate measurement noise and the data is standardized to ensure that data of different dimensions do not affect the model's learning performance during training.

[0032] To further optimize this technical solution, in step S2, the feature generation model is as follows:

[0033] ;

[0034] in, For the newly generated features; For current; Voltage; For temperature; The frequency of the signal; The rate of change of current; These are adjustable weighting coefficients;

[0035] sine function It is used to simulate the periodic changes of current, and captures instantaneous characteristics and oscillation modes based on the frequency response characteristics of transformers during operation.

[0036] logarithmic rate of change Nonlinear characteristics used to capture current changes reflect the dynamic behavior of the system; sharp changes in current are used to indicate potential fault risks.

[0037] The product of temperature and current Used to reflect the effect of temperature on current performance, revealing the impact of thermal effects on transformer efficiency and safety;

[0038] The degree of influence of each feature is determined by the weighting coefficient. Adjustments are made through data-driven methods, such as training linear regression or deep learning models, to ensure that the new features better reflect the actual performance of the transformer.

[0039] To further optimize this technical solution, in step S3:

[0040] When partitioning the dataset, ensure that each subset represents the characteristics of the overall data;

[0041] Stratified sampling is used to ensure that different types of data are reasonably distributed in each subset;

[0042] Stratified sampling ensures the randomness of the data after partitioning to prevent overfitting.

[0043] To further optimize this 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 shown below:

[0044] ;

[0045] in,

[0046] For model output, such as transformer state prediction;

[0047] For input data, such as the time-series characteristics of current, voltage, and temperature;

[0048] These are the weight parameters of the convolutional layer;

[0049] These are the weight parameters of the LSTM layer;

[0050] These are the weight parameters for the output layer;

[0051] For activation functions, such as ReLU or Sigmoid;

[0052] In deep learning models,

[0053] Input layer: Input data Including time-series features such as current, voltage, and temperature, the data is preprocessed to ensure an input format suitable for CNN and LSTM;

[0054] Convolutional Neural Networks (CNNs): Convolutional layers are used to extract local features from data. Multiple convolutional operations are defined, using convolutional kernels. Convolution is performed on the input data to generate feature maps. This indicates that the input data is processed through a convolutional layer. The feature representation is obtained;

[0055] Long Short-Term Memory (LSTM) networks: LSTM layers are applied to the feature maps output by convolutional layers to capture long-term dependencies in time series data using memory capabilities. This indicates that, based on convolutional features, the data is further processed to capture temporal dynamics;

[0056] Output layer: The output of the LSTM is passed through a fully connected layer. The process is performed to obtain the final prediction result. Activation functions are set according to task requirements, such as Sigmoid for binary classification and Softmax for multi-class classification.

[0057] To further optimize this technical solution, in step S5:

[0058] During training, set appropriate hyperparameters such as learning rate, batch size, and number of training epochs;

[0059] A learning rate scheduling strategy is used to dynamically adjust the learning rate based on the performance of the validation set in order to improve training efficiency.

[0060] Early stopping is used to prevent overfitting and avoid a model that performs well on the training set but degrades on the validation set.

[0061] During training, the loss value and accuracy for each epoch are recorded for subsequent analysis.

[0062] To further optimize this technical solution, in step S6, grid search or Bayesian optimization methods are used to find the optimal combination of hyperparameters. The verification process focuses on multiple evaluation metrics, such as accuracy, recall, and F1 score, to ensure that the model performs in a balanced manner in different aspects.

[0063] Meanwhile, the stability and generalization ability of the model are verified using cross-validation.

[0064] To further optimize this technical solution, in step S7, the confusion matrix is ​​as follows:

[0065] ;

[0066] In the structure of the confusion matrix:

[0067] True cases: The number of samples that the model correctly predicts as positive.

[0068] False positives: The number of samples that the model incorrectly predicts as positive.

[0069] True negative examples: the number of samples that the model correctly predicts as negative.

[0070] False negatives: The number of samples that the model incorrectly predicts as negative.

[0071] To further optimize this technical solution, the anomaly detection mechanism in step S8 is as follows:

[0072] ;

[0073] in,

[0074] This indicates an anomaly flag, with 1 for an anomaly and 0 for a normal condition.

[0075] For input samples;

[0076] The mean vector of the samples is calculated through clustering or historical data;

[0077] The distance between a sample and the mean vector can be calculated using either Euclidean distance or Mahalanobis distance.

[0078] The set threshold is used to determine the boundary of anomalies;

[0079] This mechanism determines whether a sample is an anomaly by calculating the distance between each input sample and the mean vector;

[0080] When the distance of the sample Exceeding the set threshold At that time, the sample was marked as an anomaly. Otherwise, mark as normal. .

[0081] To further optimize this technical solution, the adaptive learning mechanism in step S8 is as follows:

[0082] ;

[0083] in,

[0084] and These are the updated model parameters and the current model parameters, respectively.

[0085] The learning rate;

[0086] This is a real label;

[0087] These are the model's predicted values;

[0088] For input samples The gradient;

[0089] This mechanism enables the model to update its parameters in real time when new data arrives, adapting to changes in the environment.

[0090] 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. .

[0091] Compared with existing technologies, this invention provides a deep learning-based method for analyzing transformer current-boosting test data, which has the following advantages:

[0092] This deep learning-based data analysis method for transformer current boosting tests aims to improve the accuracy and robustness of transformer current boosting tests through comprehensive data analysis. It combines deep learning with features engineering, model selection and architecture design, anomaly detection and adaptive learning techniques. This not only enhances the model's ability to generalize to unseen data but also allows it to adapt to environmental changes in real time, ensuring its effectiveness in practical applications and providing stronger technical support for the safe operation of transformers. Attached Figure Description

[0093] Figure 1 This is a flowchart illustrating a deep learning-based method for analyzing transformer current-boosting test data proposed in this invention.

[0094] Figure 2 This is a schematic diagram of the data collection and preprocessing process in a deep learning-based transformer current-boosting test data analysis method proposed in this invention.

[0095] Figure 3 This is a flowchart illustrating the feature generation model in a deep learning-based transformer current-boosting test data analysis method proposed in this invention.

[0096] Figure 4 This is a flowchart illustrating the deep learning model in the transformer current-boosting test data analysis method proposed in this invention.

[0097] Figure 5 This is a flowchart illustrating the confusion matrix in a deep learning-based transformer current-boosting test data analysis method proposed in this invention.

[0098] Figure 6 This is a flowchart illustrating step S8 in a deep learning-based transformer current-boosting test data analysis method proposed in this invention. Detailed Implementation

[0099] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0100] Example:

[0101] Please see Figure 1 A deep learning-based method for analyzing transformer current-boosting test data includes the following steps:

[0102] S1. Data Collection and Preprocessing

[0103] Data from transformer current-boosting tests were collected from different data sources and preprocessed to improve the accuracy and robustness of subsequent deep learning models.

[0104] like Figure 2 As shown, in this embodiment, in step S1:

[0105] Data sources include sensors, monitoring devices, and historical records;

[0106] The data includes multiple dimensions such as current, voltage, temperature, and vibration.

[0107] Preprocessing includes cleaning, denoising, and normalizing the data;

[0108] Meanwhile, a denoising autoencoder is used to eliminate measurement noise and the data is standardized to ensure that data of different dimensions do not affect the model's learning performance during training.

[0109] S2, Feature Engineering

[0110] By employing feature selection and feature extraction methods, and based on a feature generation model derived from feature engineering, multiple features are combined to capture the complex relationships of transformer operating states, thereby selecting features that have a significant impact on transformer performance.

[0111] In this embodiment, the feature generation model in step S2 is as follows:

[0112] ;

[0113] in, For the newly generated features; For current; Voltage; For temperature; The frequency of the signal; The rate of change of current; These are adjustable weighting coefficients;

[0114] sine function It is used to simulate the periodic changes of current, and captures instantaneous characteristics and oscillation modes based on the frequency response characteristics of transformers during operation.

[0115] logarithmic rate of change Nonlinear characteristics used to capture current changes reflect the dynamic behavior of the system; sharp changes in current are used to indicate potential fault risks.

[0116] The product of temperature and current Used to reflect the effect of temperature on current performance, revealing the impact of thermal effects on transformer efficiency and safety;

[0117] The degree of influence of each feature is determined by the weighting coefficient. Adjustments are made through data-driven methods, such as training linear regression or deep learning models, to ensure that the new features better reflect the actual performance of the transformer.

[0118] like Figure 3 As shown, when using the model:

[0119] Feature Calculation: In the data preprocessing stage, the required basic features are first extracted, such as current, voltage, temperature, and their rates of change. Then, the features are calculated using a predefined formula. .

[0120] Model training: The calculated new features As one of the input features, it is fed into the deep learning model along with other original features for training. By training the model, the weight coefficients are adjusted to optimize the influence of the features.

[0121] Model Validation and Evaluation: When evaluating a model using a validation set, it is important to focus on observing new features. Impact on model performance. If this feature significantly improves the model's accuracy and robustness, further hyperparameter tuning can be performed to ensure optimal model configuration.

[0122] Feature analysis: After model training and validation, analysis is performed using methods such as SHAP values. Its importance in the model decision-making process, and whether it can indeed improve the interpretability and effectiveness of the model.

[0123] This provides a new perspective for data analysis in transformer current boosting tests, helping to more comprehensively understand the various factors affecting transformer performance.

[0124] Principal component analysis (PCA) can be used to reduce feature dimensionality 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. Furthermore, based on domain knowledge, derived features, such as the relationship between temperature and current, can be constructed to further improve the model's performance.

[0125] S3, Dataset Partitioning

[0126] The preprocessed dataset is divided into a training set, a validation set, and a test set, with a ratio of 70% for training, 15% for validation, and 15% for testing. The validation set is used for parameter tuning, and the test set is used for evaluating the final model performance.

[0127] In this embodiment, in step S3:

[0128] When partitioning the dataset, ensure that each subset represents the characteristics of the overall data;

[0129] Stratified sampling is used to ensure that different types of data are reasonably distributed in each subset;

[0130] Stratified sampling ensures the randomness of the data after partitioning to prevent overfitting.

[0131] S4. Model Selection and Architecture Design

[0132] Based on transformer current boosting test data, a suitable deep learning model was selected, and in terms of architecture design, the number of network layers, the number of neurons in each layer, and the hyperparameters of the activation function were determined.

[0133] 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 the time-series data. The deep learning model architecture formula is shown below:

[0134] ;

[0135] in,

[0136] For model output, such as transformer state prediction;

[0137] For input data, such as the time-series characteristics of current, voltage, and temperature;

[0138] These are the weight parameters of the convolutional layer;

[0139] These are the weight parameters of the LSTM layer;

[0140] These are the weight parameters for the output layer;

[0141] For activation functions, such as ReLU or Sigmoid;

[0142] In deep learning models,

[0143] Input layer: Input data Including time-series features such as current, voltage, and temperature, the data is preprocessed to ensure an input format suitable for CNN and LSTM.

[0144] Convolutional Neural Networks (CNNs): Convolutional layers are used to extract local features from data. Multiple convolutional operations are defined, using convolutional kernels. Convolution is performed on the input data to generate feature maps. This indicates that the input data is processed through a convolutional layer. This yields feature representations. The advantage of using convolutional layers is that they can effectively extract local features and reduce data dimensionality, thereby reducing the complexity of subsequent calculations.

[0145] Long Short-Term Memory (LSTM) networks: LSTM layers are applied to the feature maps output by convolutional layers to capture long-term dependencies in time series data using memory capabilities. This means that, based on convolutional features, the data is further processed to capture temporal dynamics; the introduction of LSTM enables the model to understand dynamic changes in time series data, especially abrupt changes and patterns in the current-boosting process of transformers.

[0146] Output layer: The output of the LSTM is passed through a fully connected layer. The process is performed to obtain the final prediction result. The activation function is set according to the task requirements. For example, Sigmoid is used for binary classification and Softmax is used for multi-class classification, which ensures that the model can output results that meet the actual application requirements.

[0147] Techniques such as skip connections can also be considered to enhance the model's learning ability. Furthermore, the model's complexity needs to be reasonably controlled based on the amount of data to prevent overfitting.

[0148] like Figure 4 As shown, when using this model:

[0149] Model building: When designing the model, determine the hyperparameters such as the number of convolutional and LSTM layers, the number of neurons in each layer, and the size of the convolutional kernels based on the specific task requirements (e.g., classification or regression). These parameters can be fine-tuned experimentally to achieve optimal performance.

[0150] Model Training: The constructed model is trained using the preprocessed dataset. During training, the weight parameters are optimized using the backpropagation algorithm. , and Adaptive learning rate optimization algorithms (such as Adam) can be used to accelerate the training process.

[0151] Model Validation and Evaluation: The model is evaluated using a validation set. Generalization ability is assessed by monitoring metrics such as loss function and accuracy. Simultaneously, the training and validation losses are monitored to prevent overfitting.

[0152] Model Improvement: After evaluating the model, adjust the necessary hyperparameters 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.

[0153] Results Analysis: By analyzing the model's prediction results, interpretability tools (such as SHAP or LIME) are used to gain a deeper understanding of the model's decision-making process and verify its practical applicability to transformer performance.

[0154] By combining the advantages of CNN and LSTM, it provides powerful feature extraction and time series modeling capabilities for transformer current boosting test data analysis, and can more effectively capture the complex features of the data.

[0155] S5, Model Training

[0156] The selected model is trained using the training set data, and the network parameters are optimized using the backpropagation algorithm.

[0157] In this embodiment, in step S5:

[0158] During training, set appropriate hyperparameters such as learning rate, batch size, and number of training epochs;

[0159] A learning rate scheduling strategy is used to dynamically adjust the learning rate based on the performance of the validation set in order to improve training efficiency.

[0160] Early stopping is used to prevent overfitting and avoid a model that performs well on the training set but degrades on the validation set.

[0161] During training, the loss value and accuracy for each epoch are recorded for subsequent analysis.

[0162] S6. Model Validation and Hyperparameter Tuning

[0163] After the model is trained, the model is evaluated using a validation set, and the hyperparameters are adjusted based on the evaluation results.

[0164] In this embodiment, in step S6, grid search or Bayesian optimization methods are used to find the optimal combination of hyperparameters. The verification process focuses on multiple evaluation metrics, such as accuracy, recall, and F1 score, to ensure that the model performs in a balanced manner in different aspects.

[0165] Meanwhile, the stability and generalization ability of the model are verified using cross-validation.

[0166] S7. Model Testing and Evaluation

[0167] After the model hyperparameters are tuned, it is evaluated on the test set. The confusion matrix is ​​used to analyze the model’s classification performance, identify potential biases and weaknesses, and generate ROC curves and AUC values ​​based on the test results.

[0168] In this embodiment, the confusion matrix in step S7 is as follows:

[0169] ;

[0170] In the structure of the confusion matrix:

[0171] True cases: The number of samples that the model correctly predicts as positive.

[0172] False positives: The number of samples that the model incorrectly predicts as positive.

[0173] True negative examples: the number of samples that the model correctly predicts as negative.

[0174] False negatives: The number of samples that the model incorrectly predicts as negative.

[0175] In this matrix, assuming a specific three-class classification problem (such as normal, overload, fault), by organizing the prediction results of each class into a confusion matrix, we can intuitively understand the model's performance on different classes.

[0176] This matrix not only records the TP, FP, and FN for each class, but also provides a clearer visualization of multi-class classification through the interaction of rows and columns.

[0177] For example, the first row could represent the prediction for the "normal" category, the second row for "overload", and the third row for "failure".

[0178] Several evaluation metrics can be derived from the confusion matrix:

[0179] Accuracy:

[0180]

[0181] Precision:

[0182]

[0183] Recall:

[0184]

[0185] F1 score:

[0186]

[0187] These metrics can provide a more comprehensive view of the model's overall performance, especially in cases of class imbalance.

[0188] like Figure 5As shown, this matrix is ​​used as follows:

[0189] Model evaluation: During the model testing phase, predictions are made using the test set to generate a confusion matrix. The confusion matrix can be calculated using the `confusion_matrix` function from the `sklearn` library in Python.

[0190] Visualize the confusion matrix: The generated confusion matrix is ​​visualized for intuitive understanding. Heatmaps and other methods can be used to display the prediction results for each category, helping to identify the types of confusion.

[0191] Calculate evaluation metrics: Based on the confusion matrix, calculate evaluation metrics (accuracy, precision, recall, F1-score) for each category and compare them. This process can reveal the strength of the model across different categories, especially in critical fault detection tasks.

[0192] Performance analysis: By analyzing the confusion matrix and its derived metrics, potential biases in the model can be identified. For example, if the model has a high false negative rate (FN) in the "fault" category, it may be necessary to specifically adjust the model to improve the accuracy of fault detection.

[0193] Results Summary and Improvement: Based on the analysis results, summarize the model and identify areas for improvement. If the performance of certain categories is significantly insufficient, consider adding relevant data or conducting experiments using different model architectures.

[0194] By using a confusion matrix, a more detailed and comprehensive evaluation perspective can be provided in the classification task of transformer current boosting test data, ensuring the effectiveness of the model in practical applications.

[0195] S8, Anomaly Detection and Adaptive Learning

[0196] Based on the anomaly detection mechanism, an unsupervised learning method is used to identify and process anomalous samples; an adaptive learning mechanism is established to allow the model to learn online and update model parameters when new data arrives, so as to adapt to the constantly changing environment.

[0197] In this embodiment, the anomaly detection mechanism in step S8 is as follows:

[0198] ;

[0199] in,

[0200] This indicates an anomaly flag, with 1 for an anomaly and 0 for a normal condition.

[0201] For input samples;

[0202] The mean vector of the samples is calculated through clustering or historical data;

[0203] The distance between a sample and the mean vector can be calculated using either Euclidean distance or Mahalanobis distance.

[0204] The set threshold is used to determine the boundary of anomalies;

[0205] This mechanism determines whether a sample is an anomaly by calculating the distance between each input sample and the mean vector;

[0206] When the distance of the sample Exceeding the set threshold At that time, the sample was marked as an anomaly. Otherwise, mark as normal. .

[0207] This mechanism can help identify abnormalities that may occur during current boosting tests in a timely manner, such as sudden changes in current or temperature, which may indicate potential equipment failure.

[0208] In step S8, the adaptive learning mechanism is as follows:

[0209] ;

[0210] in,

[0211] and These are the updated model parameters and the current model parameters, respectively.

[0212] The learning rate;

[0213] This is a real label;

[0214] These are the model's predicted values;

[0215] For input samples The gradient;

[0216] This mechanism enables the model to update its parameters in real time when new data arrives, adapting to changes in the environment.

[0217] 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. .

[0218] This mechanism ensures that the model can adjust itself according to the actual situation, thereby improving its adaptability to new environments.

[0219] like Figure 6 As shown, the two mechanisms are used as follows:

[0220] Data preprocessing and clustering: Before applying the model, cluster analysis is performed on the historical data to calculate the mean vector. and distance threshold Clustering methods can use unsupervised learning methods such as K-means or DBSCAN to ensure that the mean represents the features of normal samples.

[0221] Anomaly detection: When new data arrives, the above anomaly detection mechanism is used to calculate an anomaly label for each sample. If abnormal samples are found, they can be handled separately to avoid negatively impacting model training.

[0222] Online learning: For normal samples, the model parameters are updated in real time according to an adaptive learning mechanism. As new data continuously arrives, the model can gradually adjust to adapt to the new operating conditions. This process can be achieved through mini-batch online learning methods.

[0223] Model Evaluation and Monitoring: Regularly evaluate the model's performance on new datasets, ensuring its ability to adapt to changing environments by monitoring the proportion of outliers and the model's prediction accuracy. If a performance degradation is observed, consider retraining or parameter tuning.

[0224] Feedback mechanism: The detected abnormal data and their 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.

[0225] It can provide continuous monitoring and real-time adjustment capabilities for data analysis in transformer current boosting tests, thereby ensuring the effectiveness and reliability of the model in practical applications.

[0226] S9. Results Interpretation and Model Interpretability

[0227] Interpreting results in deep learning models involves using SHAP values ​​or LIME methods to analyze the model's decision-making process and understand the contribution of features to the final prediction results.

[0228] In this embodiment, the interpretability of results is crucial in deep learning models, especially in industrial applications. This not only helps verify the reliability of the model but also assists technicians in making decisions in subsequent work. Furthermore, comparing the model's predictions with actual monitoring data helps identify performance discrepancies under specific conditions and provides directions for improvement.

[0229] S10, Continuous Monitoring and Model Updates

[0230] Based on the effectiveness of deep learning models in practical applications, a continuous monitoring mechanism is established to track the model's performance in real time.

[0231] This embodiment includes collecting new data, periodically evaluating model performance, and updating the model as needed. During the data collection phase, the data flow can be automated to ensure the model always receives the latest input. Furthermore, a threshold should be set to trigger an automatic retraining or model update mechanism when the model's performance declines within a specific time period. This process ensures that the deep learning model can adapt to various changes that may occur during transformer current-increasing experiments, maintaining efficient predictive capabilities.

[0232] The beneficial effects of the present invention are:

[0233] This deep learning-based data analysis method for transformer current boosting tests aims to improve the accuracy and robustness of transformer current boosting tests through comprehensive data analysis. It combines deep learning with features engineering, model selection and architecture design, anomaly detection and adaptive learning techniques. This not only enhances the model's ability to generalize to unseen data but also allows it to adapt to environmental changes in real time, ensuring its effectiveness in practical applications and providing stronger technical support for the safe operation of transformers.

[0234] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0235] While 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 these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for analyzing transformer current-boosting test data based on deep learning, characterized in that, Includes the following steps: S1. Data collection and preprocessing; Data from transformer current-boosting tests were collected from different data sources and preprocessed to improve the accuracy and robustness of subsequent deep learning models. S2, Feature Engineering; By employing feature selection and feature extraction methods, and based on a feature generation model derived from feature engineering, multiple features are combined to capture the complex relationships of transformer operating states, thereby selecting features that have a significant impact on transformer performance. feature The generative model is shown below: ； in, For the newly generated features; For current; Voltage; For temperature; The frequency of the signal; The rate of change of current; These are adjustable weighting coefficients; sine function It is used to simulate the periodic changes of current, and captures instantaneous characteristics and oscillation modes based on the frequency response characteristics of transformers during operation. logarithmic rate of change Nonlinear characteristics used to capture current changes reflect the dynamic behavior of the system; sharp changes in current are used to indicate potential fault risks. The product of temperature and current Used to reflect the effect of temperature on current performance, revealing the impact of thermal effects on transformer efficiency and safety; The degree of influence of each feature is determined by the weighting coefficient. Adjustments are made using data-driven methods, such as training linear regression or deep learning models, to ensure that the new features better reflect the actual performance of the transformer. S3. Dataset partitioning; The preprocessed dataset is divided into a training set, a validation set, and a test set, with a ratio of 70% for training, 15% for validation, and 15% for testing; the validation set is used for parameter tuning, and the test set is used for evaluating the final model performance. S4. Model selection and architecture design; Based on transformer current boosting test data, a deep learning model was used, and the number of network layers, the number of neurons in each layer, and the hyperparameters of the activation function were determined in the architecture design. S5, Model Training; The selected model is trained using the training set data, and the network parameters are optimized using the backpropagation algorithm. S6. Model validation and hyperparameter tuning; After the model is trained, the model is evaluated using a validation set, and the hyperparameters are adjusted based on the evaluation results. S7. Model Testing and Evaluation; After the model hyperparameters are tuned, they are evaluated on the test set. The confusion matrix is ​​used to analyze the classification performance of the model, identify the biases and weaknesses of the model, and generate ROC curves and AUC values ​​based on the test results. S8, Anomaly Detection and Adaptive Learning; Based on the anomaly detection mechanism, an unsupervised learning method is used to identify and process anomalous samples; an adaptive learning mechanism is established to allow the model to learn online and update model parameters when new data arrives, so as to adapt to the constantly changing environment. The anomaly detection mechanism is as follows: ； in, This indicates an anomaly flag, with 1 for an anomaly and 0 for a normal condition. For input samples; The mean vector of the samples is calculated through clustering or historical data; The distance between a sample and the mean vector can be calculated using either Euclidean distance or Mahalanobis distance. The set threshold is used to determine the boundary of anomalies; This mechanism determines whether a sample is an anomaly by calculating the distance between each input sample and the mean vector; When the distance of the sample Exceeding the set threshold At that time, the sample was marked as an anomaly. Otherwise, mark as normal. ; When new data arrives, the aforementioned anomaly detection mechanism is used to calculate anomaly markers for each sample. If abnormal samples are found, they can be handled separately to avoid negatively impacting model training. For normal samples, the model parameters are updated in real time according to the adaptive learning mechanism; S9. Results Interpretation and Model Interpretability; Interpreting results in deep learning models involves using 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 updates; Based on the effectiveness of deep learning models in practical applications, a continuous monitoring mechanism is established to track the model's performance in real time.

2. The method for analyzing transformer current-boosting test data 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 such as current, voltage, temperature, and vibration. Preprocessing includes cleaning, denoising, and normalizing the data; Meanwhile, a denoising autoencoder is used to eliminate measurement noise and the data is standardized to ensure that data of different dimensions do not affect the model's learning performance during training.

3. The method for analyzing transformer current-boosting test data based on deep learning according to claim 1, characterized in that, In step S3: When partitioning the dataset, ensure that each subset represents the characteristics of the overall data; Stratified sampling is used to ensure that different types of data are reasonably distributed in each subset; Stratified sampling ensures the randomness of the data after partitioning to prevent overfitting.

4. The method for analyzing transformer current-boosting test data based on deep learning according to claim 1, 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 the time-series data. The deep learning model architecture formula is shown below: ； in, The model output is the transformer state prediction; The input data includes the time-series characteristics of current, voltage, and temperature. These are the weight parameters of the convolutional layer; These are the weight parameters of the LSTM layer; These are the weight parameters for the output layer; The activation function is either ReLU or Sigmoid. In deep learning models, Input layer: Input data It includes time-series features such as current, voltage, and temperature, and the data is preprocessed to ensure an input format suitable for CNN and LSTM. Convolutional Neural Networks (CNNs): Convolutional layers are used to extract local features from data. Multiple convolutional operations are defined, using convolutional kernels. Convolution is performed on the input data to generate feature maps. This indicates that the input data is processed through a convolutional layer. The feature representation is obtained; Long Short-Term Memory (LSTM) networks: LSTM layers are applied to the feature maps output by convolutional layers to capture long-term dependencies in time series data using memory capabilities. This indicates that, based on convolutional features, the data is further processed to capture temporal dynamics; Output layer: The output of the LSTM is passed through a fully connected layer. The process is performed to obtain the final prediction result. The activation function is set according to the task requirements: Sigmoid is used for binary classification, and Softmax is used for multi-class classification.

5. The method for analyzing transformer current-boosting test data based on deep learning according to claim 1, characterized in that, In step S5: During training, set appropriate hyperparameters such as learning rate, batch size, and number of training epochs; A learning rate scheduling strategy is used to dynamically adjust the learning rate based on the performance of the validation set in order to improve training efficiency. Early stopping is used to prevent overfitting and avoid a model that performs well on the training set but degrades on the validation set. During training, the loss value and accuracy for each epoch are recorded for subsequent analysis.

6. The method for analyzing transformer current-boosting test data based on deep learning according to claim 1, characterized in that, In step S6, grid search or Bayesian optimization methods are used to find the best combination of hyperparameters. The verification process focuses on multiple evaluation metrics, such as accuracy, recall, and F1 score, to ensure that the model performs in a balanced manner in different aspects. Meanwhile, the stability and generalization ability of the model are verified using cross-validation.

7. The method for analyzing transformer current-boosting test data based on deep learning according to claim 1, characterized in that, In step S7, the confusion matrix is ​​shown below: ； In the structure of the confusion matrix: True cases: The number of samples that the model correctly predicts as positive. False positives: The number of samples that the model incorrectly predicts as positive. True negative examples: the number of samples that the model correctly predicts as negative. False negatives: The number of samples that the model incorrectly predicts as negative.

8. The method for analyzing transformer current-boosting test data based on deep learning according to claim 1, characterized in that, In step S8, the adaptive learning mechanism is as follows: ； in, and These are the updated model parameters and the current model parameters, respectively. The learning rate; This is a real label; These are the model's predicted values; For input samples The gradient; This mechanism enables the model to update its parameters in real time when new data arrives, adapting to changes in the environment. 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