Damp detection method and device for transformer oil paper insulating sleeve and computer program product

Abstract

The invention discloses a damp detection method and device for a transformer oil paper insulating sleeve and a computer program product, and the method comprises the steps: S1, collecting the temperature, humidity and electric field intensity data of the interior of the transformer oil paper insulating sleeve; s2, sequentially carrying out data preprocessing operation on the collected data; s3, utilizing a frequency domain dielectric spectrum method in combination with temperature, humidity and electric field intensity data to extract characteristic parameters reflecting a damp position; s4, constructing a DNN machine learning model based on the extracted characteristic parameters, and training the DNN machine learning model; and S5, inputting to-be-evaluated transformer oil paper insulating sleeve data into the trained DNN machine learning model, outputting a damp position evaluation result, and predicting a future damp position based on historical data and a current trend. According to the invention, the evaluation precision and prediction capability of the damp position in the transformer oil paper insulating sleeve are improved, the operation is simple and convenient, the real-time performance is high, the applicability is wide, and powerful support is provided for the maintenance and safe operation of the transformer.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and in particular to a method and device for detecting moisture of an oil-paper insulation bushing of a transformer, and a computer program product. Background Art

[0002] In the power system architecture, transformers play a core role in voltage conversion, power transmission and distribution. Their operational stability and reliability are the cornerstones for ensuring safe and efficient operation of the power grid. The oil-paper insulation bushing inside the transformer is a key component for connection and protection. Its insulation performance is directly related to the long-term service life and operational safety of the transformer. Unfortunately, in actual operating scenarios, the insulation bushing is susceptible to multiple external factors such as environmental humidity, temperature changes, and electric field strength, resulting in frequent moisture. This condition will significantly weaken the insulation performance and induce serious accidents such as partial discharge and even insulation breakdown, posing a major threat to the stability of the power grid.

[0003] For the moisture detection of transformer oil-paper insulation bushings, the current technical system adopts a data-driven feature extraction and pattern recognition strategy. This strategy accurately collects key data during the operation of the transformer, such as dielectric response spectrum, partial discharge signal, etc., and uses advanced pattern recognition algorithms (such as random forest algorithms) to effectively extract moisture features and accurately predict moisture locations. The advantage of this method is that it realizes non-destructive detection, shows high flexibility, and can be compatible with multiple signal inputs, which greatly simplifies the manual analysis process and improves detection efficiency.

[0004] However, this technology has also exposed certain limitations in practical applications. Its core challenge lies in its strong reliance on high-quality data: the accuracy of the prediction results is directly subject to the quality and integrity of data acquisition, requiring a large number of data samples covering a wide range of working conditions as support to build and optimize the prediction model. In addition, due to the differences in structure, materials and operating characteristics of transformers of different types, the versatility of this method is limited, and it is difficult to directly apply it to all types of transformers and achieve universal applicability across models. Therefore, exploring a moisture detection technology that can maintain high-precision prediction capabilities while having good adaptability and generalization capabilities has become a technical problem that needs to be solved urgently. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a method, device and computer program product for detecting moisture of transformer oil-paper insulation bushing, so as to have good adaptability and generalization ability while maintaining high-precision prediction ability.

[0006] In order to solve the above technical problems, the present invention provides a method for detecting moisture of an oil-paper insulation bushing of a transformer, comprising:

[0007] Step S1, collecting temperature, humidity and electric field strength data inside the oil-paper insulation bushing of the transformer;

[0008] Step S2, performing data preprocessing operations on the collected data in sequence;

[0009] Step S3, using frequency domain dielectric spectroscopy combined with temperature, humidity and electric field strength data to extract characteristic parameters reflecting the damp position;

[0010] Step S4, constructing a DNN machine learning model based on the extracted feature parameters and training it;

[0011] Step S5, input the transformer oil-paper insulated bushing data to be evaluated into the trained DNN machine learning model, output the damp position evaluation result, and predict the future damp position based on historical data and current trends.

[0012] Preferably, the step S2 specifically includes:

[0013] Identify outliers through the statistical standard deviation method, first calculate the mean and standard deviation of the data, set the threshold, retain the data within the upper and lower limits, and remove outliers;

[0014] The median filter method is used to set the window size, perform rolling processing on the data, calculate the median of the data in each window to obtain a new data column after removing the noise, and use the forward filling method to supplement the missing values;

[0015] The Z-score standardization method is used to create a standardization tool object, standardize the data, and organize it into a suitable format.

[0016] Preferably, the characteristic parameters reflecting the dampened location extracted in step S3 include the electric field intensity gradient change, the humidity gradient change and the characteristic spectral lines of the dielectric constant changing with frequency. The electric field intensity gradient change is obtained by calculating the electric field intensity data along the coordinate axis direction using a gradient calculation method, and the humidity gradient change is obtained by calculating the difference of the humidity data changing with time and dividing it by the average value of the time interval.

[0017] Preferably, constructing a DNN machine learning model in step S4 specifically includes:

[0018] Use the Keras framework to determine the model input layer based on the input data shape;

[0019] If the data is a grid structure, add a convolutional layer first and then flatten it. Otherwise, directly use the fully connected layer to build the network structure and add a Dropout layer in the middle.

[0020] Set the number of hidden layers and neurons according to the complexity of the task, determine the output layer and activation function, select the optimizer and loss function as well as the evaluation metric, and print an overview of the model structure;

[0021] Use the prepared training data to train the model, set the training rounds, batch size, and validation set ratio, and set and print the training process parameters.

[0022] Preferably, the transformer oil-paper insulation bushing data to be evaluated in step S5 includes data collected in step S1, pre-processed in step S2, and feature extracted in step S3.

[0023] The present invention also provides a moisture detection device for transformer oil-paper insulation bushing, comprising:

[0024] The acquisition module is used to collect the temperature, humidity and electric field strength data inside the oil-paper insulation bushing of the transformer;

[0025] A preprocessing module is used to perform data preprocessing operations on the collected data in sequence;

[0026] The feature extraction module is used to extract characteristic parameters reflecting the damp position by combining the temperature, humidity and electric field strength data with the frequency domain dielectric spectroscopy method;

[0027] Model building module, used to build a DNN machine learning model based on the extracted feature parameters and train it;

[0028] The detection module is used to input the data of the transformer oil-paper insulated bushing to be evaluated into the trained DNN machine learning model, output the evaluation results of the moisture-affected location, and predict the future moisture-affected location based on historical data and current trends.

[0029] Preferably, the feature extraction module extracts characteristic parameters reflecting the dampened location, including the electric field intensity gradient change, the humidity gradient change and the characteristic spectral lines of the dielectric constant changing with frequency. The electric field intensity gradient change is obtained by calculating the electric field intensity data along the coordinate axis by a gradient calculation method, and the humidity gradient change is obtained by calculating the difference of the humidity data changing with time and dividing it by the average value of the time interval.

[0030] Preferably, the model building module builds a DNN machine learning model, specifically including:

[0031] Use the Keras framework to determine the model input layer based on the input data shape;

[0032] If the data is a grid structure, add a convolutional layer first and then flatten it. Otherwise, directly use the fully connected layer to build the network structure and add a Dropout layer in the middle.

[0033] Set the number of hidden layers and neurons according to the complexity of the task, determine the output layer and activation function, select the optimizer and loss function as well as the evaluation metric, and print an overview of the model structure;

[0034] Use the prepared training data to train the model, set the training rounds, batch size, and validation set ratio, and set and print the training process parameters.

[0035] The present invention also provides a moisture detection device for transformer oil-paper insulation bushing, comprising:

[0036] one or more processors;

[0037] Memory;

[0038] One or more applications, wherein the one or more applications are stored in the memory and are configured to be executed by the one or more processors, and the one or more applications are configured to execute the moisture detection method for the transformer oil-paper insulation bushing.

[0039] The present invention also provides a computer program product, comprising computer instructions, wherein the computer instructions instruct a computer device to execute operations corresponding to the method.

[0040] The implementation of the present invention has the following beneficial effects: the present invention greatly improves the assessment accuracy and prediction ability of the damp position in the oil-paper insulation bushing of the transformer, accurately locates the damp area and predicts the development trend, and effectively ensures the safe and stable operation of the transformer. Secondly, it is easy to operate and has good real-time performance. It can be widely used in the damp monitoring of oil-paper insulation bushings of various transformers. By collecting key data such as electric field strength, temperature, humidity, etc. in real time, and analyzing it through machine learning algorithms, it can efficiently evaluate moisture migration and damp conditions, reduce the burden of manual inspections, significantly improve maintenance efficiency, detect potential faults in advance, and reduce the risk of equipment downtime. Furthermore, it can be effectively applied to the life management of power equipment, control the aging and dampness of equipment based on predicted data, reasonably adjust the operation strategy, maintain stable insulation performance, extend the service life of equipment, reduce operation and maintenance costs, lay a solid foundation for the intelligent management and control of power equipment, comprehensively enhance the safety and reliability of the power system, and promote the efficient development of the power industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0042] Figure 1The present invention is a flowchart of a method for detecting moisture on an oil-paper insulation bushing of a transformer according to an embodiment of the present invention. DETAILED DESCRIPTION

[0043] The following descriptions of the embodiments refer to the accompanying drawings to illustrate specific embodiments in which the present invention may be implemented.

[0044] Please refer to Figure 1 As shown, an embodiment of the present invention provides a method for detecting moisture of an oil-paper insulation bushing of a transformer, comprising:

[0045] Step S1, collecting temperature, humidity and electric field strength data inside the oil-paper insulation bushing of the transformer;

[0046] Step S2, performing data preprocessing operations on the collected data in sequence;

[0047] Step S3, using frequency domain dielectric spectroscopy combined with temperature, humidity and electric field strength data to extract characteristic parameters reflecting the damp position;

[0048] Step S4, constructing a DNN machine learning model based on the extracted feature parameters and training it;

[0049] Step S5, input the transformer oil-paper insulated bushing data to be evaluated into the trained DNN machine learning model, output the damp position evaluation result, and predict the future damp position based on historical data and current trends.

[0050] Through the above steps, it can be seen that the embodiment of the present invention can comprehensively obtain the internal environmental information of the bushing by installing multiple sensors to collect data, providing a rich data source for subsequent precise analysis; the data preprocessing step effectively improves the data quality, reduces error interference, and ensures the accuracy and reliability of the analysis; the extraction of comprehensive feature parameters combines multiple key data to accurately locate the moisture characteristics, making the evaluation more targeted; the DNN model constructed and trained based on this can not only accurately evaluate the current moisture position in the oil-paper insulation bushing of the transformer, but also effectively predict the future moisture position based on historical data and trends, prevent potential faults in advance, and improve the safety and stability of power system operation.

[0051] Specifically, in the embodiment of the present invention, in step S1, a temperature sensor, a humidity sensor and an electric field strength sensor are installed on the oil-paper insulation bushing of the transformer to monitor the changes in temperature, humidity and electric field strength inside the bushing in real time, and the collected data is transmitted to the data processing unit.

[0052] Step S2 performs preprocessing operations such as cleaning, denoising and standardization on the collected data to eliminate outliers and noise interference and improve data quality. It specifically includes:

[0053] (1) Outlier cleaning and processing

[0054] Due to factors such as sensor failure and signal interference, outliers may appear in the collected data. This method uses the statistical standard deviation method to identify outliers. First, the average value of the data is calculated, then the standard deviation is calculated, and a threshold (for example, 3 times the standard deviation) is set. The upper and lower limits are determined by adding and subtracting the product of the threshold and the standard deviation from the average value. Finally, the data within the upper and lower limits are retained and the outliers are removed.

[0055] import numpy as np

[0056] import pandas as pd

[0057] mean_val = df['data'].mean()

[0058] std_dev = df['data'].std()

[0059] threshold=3

[0060] lower_bound=mean_val-threshold*std_dev

[0061] upper_bound=mean_val+threshold*std_dev

[0062] df_cleaned=df[(df['data']>=lower_bound)&(df['data']<=upper_bound)]

[0063] print(df_cleaned)

[0064] (2) Noise Removal

[0065] The present invention uses the median filtering method to remove random noise in the data: a window size is set (for example, set to 3), the data is rolled according to the window size, and the median of the data in each window is calculated to obtain a new data column after the noise is removed. The missing values ​​are supplemented by the forward filling method.

[0066] window_size=3

[0067] df['data_filtered']=df['data'].rolling(window=window_size,min_periods=1).median()

[0068] df['data_filtered']=df['data_filtered'].fillna(method='ffill')

[0069] print(df[['data','data_filtered']])

[0070] (3) Data standardization

[0071] In order to make data of different dimensions comparable and analyzed in the same framework, the data needs to be standardized. The present invention realizes data standardization through a Z-score standardization method: firstly, a standardized tool object is created, and then the object is used to perform standardized conversion on the data, and the converted data is organized into a suitable format for subsequent use.

[0072] from sklearn.preprocessing import StandardScaler

[0073] scaler = StandardScaler()

[0074] df['data_scaled']=scaler.fit_transform(df[['data']])

[0075] df['data_scaled']=pd.Series(df['data_scaled'].ravel(),index=df.index)

[0076] print(df[['data','data_scaled']])

[0077] Step S3 uses the frequency domain dielectric spectroscopy method to analyze the dielectric properties of the oil-paper insulation material, extracts characteristic parameters of the damp state from the frequency domain spectrum, and extracts comprehensive characteristic parameters that can reflect the damp position by combining temperature, humidity and electric field strength data. Specifically, it includes the following aspects:

[0078] (1) Electric field intensity gradient change

[0079] Using the gradient calculation method, the gradient change of the electric field strength data along the corresponding coordinate axis direction (such as X, Y, and Z directions) is calculated. Then, the appropriate image size is set through the drawing tool (Matplotlib), and the calculated X-direction electric field strength gradient change is presented in a visual form, and elements such as color bars and titles are added to make the display effect clearer and more intuitive.

[0080] grad_E_x,grad_E_y,grad_E_z=gradient(E_field,axis=(0,1,2))

[0081] plt.figure(figsize=(8,6))

[0082] plt.imshow(grad_E_x[:,:,5],cmap='viridis')

[0083] plt.colorbar()

[0084] plt.title('Gradient of Electric Field Intensity in X Direction')

[0085] plt.show()

[0086] (2) Humidity gradient change (time series)

[0087] First, calculate the difference in humidity data over time, and then divide it by the average of the time interval to get the humidity gradient change value. Then use the drawing tool (Matplotlib) to set the appropriate image size and plot the humidity gradient change under the time series into a line graph. At the same time, add elements such as titles, axis labels, and grid lines to the chart to make the chart clearer.

[0088] humidity_gradient=np.diff(humidity) / np.diff(time).mean()

[0089] plt.figure(figsize=(8,4))

[0090] plt.plot(time[:-1],humidity_gradient)

[0091] plt.title('Humidity Gradient(Change Rate)')

[0092] plt.xlabel('Time')

[0093] plt.ylabel('Gradient of Humidity')

[0094] plt.grid(True)

[0095] plt.show()

[0096] (3) Characteristic spectrum of dielectric constant changing with frequency

[0097] With the help of drawing tools (Matplotlib), set the appropriate image size, and draw the corresponding relationship between the dielectric constant and the frequency into a line graph. At the same time, add elements such as titles, axis labels, and grid lines to the chart to clearly show the characteristic spectral lines of the dielectric constant changing with frequency.

[0098] plt.figure(figsize=(8,4))

[0099] plt.plot(frequencies,dielectric_constants)

[0100] plt.title('Dielectric Constant vs Frequency')

[0101] plt.xlabel('Frequency(Hz)')

[0102] plt.ylabel('Dielectric Constant')

[0103] plt.grid(True)

[0104] plt.show()

[0105] The above code shows how to extract the characteristics of the electric field intensity gradient, humidity gradient, and dielectric constant changing with frequency from the simulation data, and can be visualized by using Matplotlib.

[0106] In step S4, a DNN (deep neural network) machine learning model is constructed based on the extracted feature parameters to evaluate the moisture-affected position in the oil-paper insulation bushing.

[0107] (1) Model construction

[0108] Use Keras to build a DNN model: first, clarify the shape of the input data (assuming that the training data X_train and the corresponding label y_train are ready), determine the input layer of the model according to the characteristics of the input data, if the input is data with a grid structure such as an image, you can add a convolutional layer first and then flatten it. If this is not the case, directly use the fully connected layer (Dense layer) to build the network structure, add a Dropout layer in the middle to prevent overfitting, and reasonably set the number of hidden layers and the number of neurons in each layer according to the complexity of the task. Finally, determine the output layer (if the regression problem outputs a continuous value of the degree of moisture, use the sigmoid activation function; if it is a classification problem, use the softmax activation function and set the appropriate number of output units). After building the model, select a suitable optimizer (here use the Adam optimizer and set the learning rate), set the loss function and the corresponding evaluation metric according to the task type (assuming it is a regression problem, choose the mean square error as the loss function), and finally print out the structural overview of the model.

[0109] import numpy as np

[0110] from tensorflow.keras.models import Sequential

[0111] from tensorflow.keras.layers import Dense,Dropout,Flatten

[0112] from tensorflow.keras.optimizers import Adam

[0113] #Assume that X_train and y_train are the prepared training data and labels

[0114] #The shape of X_train should be (number of samples, number of features) or (number of samples, height, width, number of channels)

[0115] #The shape of y_train should be (number of samples,) or (number of samples, number of output categories)

[0116] #Build the model

[0117] model=Sequential([

[0118] #If the input is an image or data with a grid structure, you can add a Conv2D layer before Flatten

[0119] #For example: Conv2D(32,(3,3),activation='relu',input_shape=(height,width,channels)),

[0120] #Flatten(),

[0121] #Otherwise, use the Dense layer directly

[0122] Dense(128,activation='relu',input_shape=(X_train.shape[1],)),#Assume that the input layer has 128 neurons

[0123] Dropout(0.5),#Add Dropout to prevent overfitting

[0124] Dense(64,activation='relu'),#Hidden layer 1

[0125] Dropout(0.5),

[0126] Dense(32,activation='relu'),#Hidden layer 2, adjust the number of layers and neurons according to the complexity of the task

[0127] Dense(1,activation='sigmoid')#Output layer, assuming it is a regression problem, output the continuous value of moisture degree

[0128] #If it is a classification problem, use the softmax activation function and set the appropriate number of output units])

[0129] # Compile the model

[0130] model.compile(optimizer=Adam(learning_rate=0.001),

[0131] loss = 'mse', #assuming it is a regression problem, use mean square error

[0132] metrics = ['accuracy']) # Note: For regression problems, 'accuracy' may not be the most appropriate metric

[0133] #If it is a classification problem, you can change it to 'categorical_accuracy' or other suitable metrics

[0134] #Print model structure

[0135] model.summary()

[0136] (2) Model training

[0137] Use the prepared training data X_train and the corresponding label y_train to train the constructed model, set the number of training rounds (here set to 100 rounds), the size of each batch of data (the batch size is set to 32), and set aside 20% of the data from the training data as a validation set. At the same time, set the relevant parameters for printing the training process and let the model start training.

[0138] history=model.fit(X_train,y_train,

[0139] epochs=100,# training rounds

[0140] batch_size=32, #batch size

[0141] validation_split=0.2, # set aside 20% of the data as a validation set

[0142] verbose=1)#Print the training process

[0143] (3) Model evaluation and prediction

[0144] Input the test data X_test into the trained model, calculate the loss value and accuracy through the model (note that 'accuracy' may not be the most appropriate metric for regression problems), and output the corresponding test loss and test accuracy. In addition, when there is a new data sample X_new, the trained model can be used to predict it and obtain the corresponding prediction result.

[0145] loss,accuracy=model.evaluate(X_test,y_test)

[0146] print(f'Test Loss:{loss},Test Accuracy:{accuracy}')

[0147] predictions = model.predict(X_new) #X_new is a new data sample

[0148] Finally, in step S5, the relevant data of the transformer oil-paper insulation bushing to be evaluated is input into the trained model. The model will output the evaluation results of the damp location and make corresponding predictions on the future damp location conditions based on historical data and current damp trends.

[0149] Corresponding to the method for detecting moisture of the transformer oil-paper insulation bushing described in the first embodiment of the present invention, the second embodiment of the present invention further provides a device for detecting moisture of the transformer oil-paper insulation bushing, comprising:

[0150] The acquisition module is used to collect the temperature, humidity and electric field strength data inside the oil-paper insulation bushing of the transformer;

[0151] A preprocessing module is used to perform data preprocessing operations on the collected data in sequence;

[0152] The feature extraction module is used to extract characteristic parameters reflecting the damp position by combining the temperature, humidity and electric field strength data with the frequency domain dielectric spectroscopy method;

[0153] Model building module, used to build a DNN machine learning model based on the extracted feature parameters and train it;

[0154] The detection module is used to input the data of the transformer oil-paper insulated bushing to be evaluated into the trained DNN machine learning model, output the evaluation results of the moisture-affected location, and predict the future moisture-affected location based on historical data and current trends.

[0155] Preferably, the feature extraction module extracts characteristic parameters reflecting the dampened location, including the electric field intensity gradient change, the humidity gradient change and the characteristic spectral lines of the dielectric constant changing with frequency. The electric field intensity gradient change is obtained by calculating the electric field intensity data along the coordinate axis by a gradient calculation method, and the humidity gradient change is obtained by calculating the difference of the humidity data changing with time and dividing it by the average value of the time interval.

[0156] Preferably, the model building module builds a DNN machine learning model, specifically including:

[0157] Use the Keras framework to determine the model input layer based on the input data shape;

[0158] If the data is a grid structure, add a convolutional layer first and then flatten it. Otherwise, directly use the fully connected layer to build the network structure and add a Dropout layer in the middle.

[0159] Set the number of hidden layers and neurons according to the complexity of the task, determine the output layer and activation function, select the optimizer and loss function as well as the evaluation metric, and print an overview of the model structure;

[0160] Use the prepared training data to train the model, set the training rounds, batch size, and validation set ratio, and set and print the training process parameters.

[0161] Corresponding to the method for detecting moisture of the transformer oil-paper insulation bushing described in the first embodiment of the present invention, the third embodiment of the present invention further provides a device for detecting moisture of the transformer oil-paper insulation bushing, comprising:

[0162] one or more processors;

[0163] Memory;

[0164] One or more applications, wherein the one or more applications are stored in the memory and are configured to be executed by the one or more processors, and the one or more applications are configured to execute the moisture detection method for the transformer oil-paper insulation bushing.

[0165] Corresponding to the method for detecting moisture of the transformer oil-paper insulation bushing described in the aforementioned embodiment 1 of the present invention, embodiment 4 of the present invention also provides a computer program product, including computer instructions, and the computer instructions instruct a computer device to perform operations corresponding to the method for detecting moisture of the transformer oil-paper insulation bushing described in the aforementioned embodiment 1 of the present invention.

[0166] Preferably, the processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor, or the processor may be any conventional processor. The processor is the control center of the device, and various parts of the device are connected using various interfaces and lines.

[0167] The memory mainly includes a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function, etc., and the data storage area can store related data, etc. In addition, the memory can be a high-speed random access memory, or a non-volatile memory, such as a plug-in hard disk, a smart memory card (SmartMedia Card, SMC), a secure digital (Secure Digital, SD) card, and a flash card (Flash Card), etc., or the memory can also be other volatile solid-state storage devices.

[0168] It should be noted that the above-mentioned device may include but is not limited to a processor and a memory, which can be understood by those skilled in the art.

[0169] For the working principle and process of this embodiment, please refer to the description of the aforementioned embodiment 1 of the present invention, which will not be repeated here.

[0170] Compared with the prior art, the beneficial effect brought by the embodiments of the present invention is that the present invention greatly improves the assessment accuracy and prediction ability of the damp position in the oil-paper insulation bushing of the transformer, accurately locates the damp area and predicts the development trend, and effectively ensures the safe and stable operation of the transformer. Secondly, it is easy to operate and has good real-time performance. It can be widely used in the damp monitoring of oil-paper insulation bushings of various transformers. By collecting key data such as electric field strength, temperature, humidity, etc. in real time, and analyzing it through machine learning algorithms, it can efficiently evaluate moisture migration and damp conditions, reduce the burden of manual inspections, significantly improve maintenance efficiency, detect potential faults in advance, and reduce the risk of equipment shutdown. Furthermore, it can be effectively applied to the life management of power equipment, control the aging and dampness of equipment based on predicted data, reasonably adjust the operation strategy, maintain the stability of insulation performance, extend the service life of equipment, reduce operation and maintenance costs, lay a solid foundation for the intelligent management and control of power equipment, comprehensively enhance the safety and reliability of the power system, and promote the efficient development of the power industry.

[0171] The above disclosure is only the preferred embodiment of the present invention, which certainly cannot be used to limit the scope of the present invention. Therefore, equivalent changes made according to the claims of the present invention are still within the scope of the present invention.

Claims

1. A method for detecting moisture in transformer oil-paper insulation bushing, characterized in that: include: Step S1, collecting temperature, humidity and electric field strength data inside the oil-paper insulation bushing of the transformer; Step S2, performing data preprocessing operations on the collected data in sequence; Step S3, using frequency domain dielectric spectroscopy combined with temperature, humidity and electric field strength data to extract characteristic parameters reflecting the damp position; Step S4, constructing a DNN machine learning model based on the extracted feature parameters and training it; Step S5, input the transformer oil-paper insulated bushing data to be evaluated into the trained DNN machine learning model, output the damp position evaluation result, and predict the future damp position based on historical data and current trends.

2. The method according to claim 1, characterized in that The step S2 specifically includes: Identify outliers through the statistical standard deviation method, first calculate the mean and standard deviation of the data, set the threshold, retain the data within the upper and lower limits, and remove outliers; The median filter method is used to set the window size, perform rolling processing on the data, calculate the median of the data in each window to obtain a new data column after removing the noise, and use the forward filling method to supplement the missing values; The Z-score standardization method is used to create a standardization tool object, standardize the data, and organize it into a suitable format.

3. The method according to claim 2, characterized in that In step S3, characteristic parameters reflecting the dampened position are extracted, including the electric field intensity gradient change, the humidity gradient change and the characteristic spectrum line of the dielectric constant changing with frequency. The electric field intensity gradient change is obtained by calculating the electric field intensity data along the coordinate axis by a gradient calculation method, and the humidity gradient change is obtained by calculating the difference of the humidity data changing with time and dividing it by the average value of the time interval.

4. The method according to claim 3, characterized in that The step S4 constructs a DNN machine learning model, specifically including: Use the Keras framework to determine the model input layer based on the input data shape; If the data is a grid structure, add a convolutional layer first and then flatten it. Otherwise, directly use the fully connected layer to build the network structure and add a Dropout layer in the middle. Set the number of hidden layers and neurons according to the complexity of the task, determine the output layer and activation function, select the optimizer and loss function as well as the evaluation metric, and print an overview of the model structure; Use the prepared training data to train the model, set the training rounds, batch size, and validation set ratio, and set and print the training process parameters.

5. The method according to claim 4, characterized in that The transformer oil-paper insulation bushing data to be evaluated in step S5 includes data collected in step S1, pre-processed in step S2, and feature extracted in step S3.

6. A moisture detection device for transformer oil-paper insulation bushing, characterized in that: include: The acquisition module is used to collect the temperature, humidity and electric field strength data inside the oil-paper insulation bushing of the transformer; A preprocessing module is used to perform data preprocessing operations on the collected data in sequence; The feature extraction module is used to extract characteristic parameters reflecting the damp position by combining the temperature, humidity and electric field strength data with the frequency domain dielectric spectroscopy method; Model building module, used to build a DNN machine learning model based on the extracted feature parameters and train it; The detection module is used to input the data of the transformer oil-paper insulated bushing to be evaluated into the trained DNN machine learning model, output the evaluation results of the moisture-affected location, and predict the future moisture-affected location based on historical data and current trends.

7. The device according to claim 6, characterized in that The feature extraction module extracts characteristic parameters reflecting the dampened location, including the electric field intensity gradient change, the humidity gradient change and the characteristic spectrum lines of the dielectric constant changing with frequency. The electric field intensity gradient change is obtained by calculating the electric field intensity data along the coordinate axis direction using a gradient calculation method, and the humidity gradient change is obtained by calculating the difference of the humidity data changing with time and dividing it by the average value of the time interval.

8. The device according to claim 7, characterized in that The model building module builds a DNN machine learning model, specifically including: Use the Keras framework to determine the model input layer based on the input data shape; If the data is a grid structure, add a convolutional layer first and then flatten it. Otherwise, directly use the fully connected layer to build the network structure and add a Dropout layer in the middle. Set the number of hidden layers and neurons according to the complexity of the task, determine the output layer and activation function, select the optimizer and loss function as well as the evaluation metric, and print an overview of the model structure; Use the prepared training data to train the model, set the training rounds, batch size, and validation set ratio, and set and print the training process parameters.

9. A moisture detection device for transformer oil-paper insulation bushing, characterized in that: include: one or more processors; Memory; One or more applications, wherein the one or more applications are stored in the memory and are configured to be executed by the one or more processors, and the one or more applications are configured to execute the moisture detection method for the transformer oil-paper insulation bushing according to any one of claims 1 to 5.

10. A computer program product, characterized in that The method comprises computer instructions, wherein the computer instructions instruct a computer device to execute operations corresponding to the method according to any one of claims 1 to 5.

Images