On-load tap changer operating state detection method and system

Abstract

This invention relates to a method and system for detecting the operating status of on-load tap changers. The method for detecting the operating status of on-load tap changers includes the following steps: acquiring vibration signals of the on-load tap changer, and detecting the operating status based on a trained recognition model and the acquired vibration signals; wherein, the recognition model uses an expanded dataset of fault vibration signals of the on-load tap changer as training input, specifically: randomly selecting a fault sample point as the original sample point, searching for the K closest sample points centered on the original sample point, randomly selecting one of the searched sample points, generating a new sample between the original sample point and the searched sample point, and merging the generated new sample with the original fault sample as the fault dataset for training input.

Description

Technical Field

[0001] This invention relates to the field of transformer technology, specifically to a method and system for detecting the operating status of on-load tap changers. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] On-load tap-change (OLTC) is a crucial component of transformers, playing vital roles in stabilizing load center voltage, increasing grid flexibility, and improving power quality. After being transported, assembled, and put into use, the OLTC is subjected to significant electrical and mechanical shocks, leading to a gradual decline in performance. This can range from minor voltage regulation failures to severe transformer burnout. Such failures inevitably cause damage to electrical equipment and waste of human and material resources, resulting in adverse economic and social impacts on the power system.

[0004] Currently, OLTC fault detection mainly uses a series of point sensors such as accelerometers. However, these sensors are susceptible to electromagnetic interference due to their metal components, and their poor corrosion resistance, as well as the fact that they can only test a limited number of points, greatly reduce their effectiveness in application. Summary of the Invention

[0005] To address the technical problems mentioned above, this invention provides a method and system for detecting the operating status of an on-load tap changer. The method involves acquiring vibration signals from an on-load tap changer (OLTC) under fault conditions using a distributed optical fiber acoustic sensor (DAS), artificially synthesizing new minority samples based on the acquired vibration signals to expand the sample pool, training a detection model using the expanded samples, and finally using the trained detection model to achieve status detection.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] The first aspect of the present invention provides a method for detecting the operating status of an on-load tap changer, comprising the following steps:

[0008] The vibration signal of the on-load tap changer is acquired, and the operating status is detected based on the trained recognition model and the acquired vibration signal.

[0009] The identification model uses an augmented dataset of fault vibration signals from on-load tap changers as training input, specifically:

[0010] A faulty sample point is randomly selected as the original sample point. Using this sample point as the center, the K nearest sample points are searched. A new sample is generated between the original sample point and the searched sample point. The generated new sample and the original faulty sample are merged as the faulty dataset for training input.

[0011] The training process of the recognition model includes:

[0012] Fault samples and normal samples are acquired using distributed fiber optic acoustic sensors arranged on the top of the on-load tap changer and in the chassis. After preprocessing, fault datasets and normal datasets are obtained.

[0013] The fault dataset and the normal dataset are processed sequentially through convolution, pooling, and classification to output the recognition results;

[0014] The model is trained using backpropagation, and the weights and biases of each layer are adjusted using stochastic gradient descent until the optimal recognition rate is found.

[0015] Preprocessing includes data segmentation and data augmentation. Specifically, data segmentation involves dividing faulty samples and normal samples into sample points of the same matrix size.

[0016] The data expansion specifically includes:

[0017] x n =x i +β(x zi -x i )

[0018] Where, x n The data to be generated, x i Then represents an existing data point in the dataset, β represents the Euclidean distance range between the generated data point and the original data point, and x represents the distance between the generated data point and the original data point. zi This represents another point in the original dataset.

[0019] The faulty sample, after being expanded, contains the same amount of data as the normal sample.

[0020] A second aspect of the present invention provides a system for implementing the above-described method, comprising:

[0021] The signal acquisition module is configured to acquire the vibration signal of the on-load tap changer.

[0022] The detection module is configured to detect the running status based on the trained recognition model and the acquired vibration signal.

[0023] The identification model uses an augmented dataset of fault vibration signals from on-load tap changers as training input, specifically:

[0024] A faulty sample point is randomly selected as the original sample point. Using this sample point as the center, the K nearest sample points are searched. A new sample is generated between the original sample point and the searched sample point. The generated new sample and the original faulty sample are merged as the faulty dataset for training input.

[0025] A third aspect of the present invention provides a computer-readable storage medium.

[0026] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the on-load tap changer operation status detection method described above.

[0027] A fourth aspect of the present invention provides a computer device.

[0028] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the on-load tap changer operation status detection method described above.

[0029] Compared with existing technologies, one or more of the above technical solutions have the following beneficial effects:

[0030] 1. Since the failure of an on-load tap changer has a huge impact on the transformer, the amount of fault sample data is relatively small. Vibration signals of the on-load tap changer (OLTC) under fault conditions are obtained by using a distributed optical fiber acoustic sensor (DAS). New samples are synthesized by an algorithm to expand the sample. The expanded samples are used to train the detection model. Expanding the fault sample event dataset can improve the accuracy of the identification model and effectively extract the deep features of the data.

[0031] 2. During the training of the recognition model, the one-dimensional vibration signal collected by the distributed optical fiber acoustic sensor (DAS) is used as the input, thereby effectively avoiding the feature loss problem that is easily caused during the data conversion process. Attached Figure Description

[0032] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0033] Figure 1 This is a schematic diagram of the CNN-LSTM network structure provided in one or more embodiments of the present invention;

[0034] Figure 2 It is the confusion matrix before training set processing provided in one or more embodiments of the present invention;

[0035] Figure 3 It is a confusion matrix of the training set provided in one or more embodiments of the present invention after being augmented by the SMOTE algorithm;

[0036] Figure 4 This is an overall flowchart of the SMOTE-CNN-LSTM recognition algorithm provided in one or more embodiments of the present invention. Detailed Implementation

[0037] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0038] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0039] SMOTE, short for Synthetic Minority Over-Sampling Technique, is a method that uses algorithms to artificially synthesize new minority samples instead of directly resampling the minority class.

[0040] LSTM, short for Long Short Term Memory, is a special type of recurrent neural network used to avoid long-term dependency problems and can remember information for a long time.

[0041] Figure 1 In the network structure, there are Input layer, Convolution layer, Pooling layer, LSTM layer, Output layer, and CNN layer.

[0042] Figure 2 and Figure 3 In the matrix, the horizontal axis represents Predicted labels, and the vertical axis represents true labels.

[0043] Figure 4 In the process:

[0044] Data acquired by DAS (Data Acquired Based on Distributed Fiber Optic Acoustic Sensor);

[0045] Data segmentation (SMOTE) (Data segmentation based on the SMOTE method)

[0046] Data and labels, training set, test set;

[0047] Training the CNN-LSTM classifier

[0048] Well-trained SMOTE-CNN-LSTM classifier.

[0049] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0050] As described in the background section, on-load tap-change switches (OLTCs) gradually degrade in performance as transformers operate, leading to various faults. Currently, when point sensors are used to obtain OLTC fault status, the sensors themselves are susceptible to electromagnetic interference due to their metal components, and are also prone to corrosion due to their material properties. Furthermore, only a limited number of points can be tested during testing, resulting in unsatisfactory performance.

[0051] Distributed optical fiber acoustic sensors (DAS) are characterized by their resistance to electromagnetic interference, corrosion resistance, and high sensitivity, making them suitable for real-time monitoring and applicable to a wide range of fields. Existing technologies can be based on machine learning and deep learning, and can be used in conjunction with DAS to complete detection and fault identification in certain fields. However, because OLTC faults have a significant impact on transformers and the consequences of faults are severe, the number of fault samples collected is extremely small. In contrast, the number of samples under normal conditions can be infinitely extracted. Machine learning and deep learning require a huge amount of data for training, and the lack of fault data makes the fault detection effect of machine learning and deep learning unsatisfactory.

[0052] Therefore, the following embodiments provide a method and system for detecting the operating status of on-load tap changers. Vibration signals of on-load tap changers (OLTC) under fault conditions are obtained through distributed optical fiber acoustic sensors (DAS). Based on the obtained vibration signals, some new minority samples are artificially synthesized to expand the sample. The expanded samples are used to train the detection model. Finally, the trained detection model is used to realize the status detection.

[0053] Example 1:

[0054] The method for detecting the operating status of an on-load tap changer includes the following steps:

[0055] The vibration signal of the on-load tap changer is acquired, and the operating status is detected based on the trained recognition model and the acquired vibration signal.

[0056] The identification model uses an augmented dataset of fault vibration signals from on-load tap changers as training input, specifically:

[0057] A faulty sample point is randomly selected as the original sample point. Using this sample point as the center, the K nearest sample points are searched. A new sample is generated between the original sample point and the searched sample point. The generated new sample and the original faulty sample are merged as the faulty dataset for training input.

[0058] Specifically:

[0059] Vibration signals from an on-load tap changer (OLTC) are acquired using a distributed optical fiber acoustic sensor (DAS). The operational status of the OLTC is then detected using a trained recognition model. This process includes the following steps: Figure 4 As shown:

[0060] (1) A distributed fiber optic acoustic sensor (DAS) system is built in an on-load tap changer (OLTC). In this embodiment, the optical fiber is distributed on the top of the OLTC and in the chassis.

[0061] (2) Use DAS to collect fault information and normal information of OLTC.

[0062] (3) The fault information collected by DAS is expanded using the SMOTE algorithm.

[0063] (4) The collected data is fed into the CNN-LSTM network for training.

[0064] Step (2) involves data acquisition, data segmentation and preprocessing, and database construction. In this embodiment, data acquisition is performed on an OLTC transformer, collecting data for both normal and fault conditions. The acquired backscattered Rayleigh signals are segmented using Python, resulting in a final data matrix size of 20000*1.

[0065] Step (3) uses the SMOTE algorithm to balance the collected data, expanding the amount of fault information data to the same amount as normal information data.

[0066] A faulty sample point is randomly selected as the original sample point. Using this sample point as the center, the K nearest sample points are searched. A new sample is generated between the original sample point and the searched sample point. The generated new sample and the original faulty sample are merged as the faulty dataset for training input.

[0067] The sampling formula for the SMOTE algorithm is as follows:

[0068] x n =x i +β(x zi -x i )

[0069] Where x n The data to be generated, x i This represents an existing data point in the dataset, and β represents the Euclidean distance between the generated data point and the original data point, ranging from 0 to 1. zi This represents another point in the original dataset.

[0070] Step (4): The collected data and the expanded dataset are fed into the CNN-LSTM network structure for recognition using the LSTM algorithm, and the recognition result is finally obtained.

[0071] The CNN-LSTM network model mainly consists of four parts, such as Figure 1 As shown:

[0072] A. The parameters of the CNN-LSTM network model are shown in Table 1. The four parts are the input layer, convolutional layer, pooling layer, and output layer.

[0073] Table 1: CNN-LSTM network model parameters

[0074]

[0075] B. Based on the actual input situation, the original one-dimensional vibration signal collected by DAS is used as input. The input one-dimensional vibration signal first undergoes a one-dimensional convolution operation with a kernel size of 3 and a number of 32. The one-dimensional convolutional layer can be specifically represented as:

[0076]

[0077] In the formula: l represents the number of network layers; M represents the input x l-1 The number of channels; ω i denoted by ; b represents the weight of the i-th convolutional kernel; δ represents the bias function; δ is a non-linear activation function, designed to make the model approximate any non-linear function; and * represents the convolution operation.

[0078] C. After convolution, the data enters the pooling layer, whose main function is feature dimensionality reduction. Global average pooling is selected, as shown in the following formula:

[0079]

[0080] In the formula, C l+1 For output; s is the pooling kernel compensation; L is the pooling kernel size; a l (n) represents the l-th neuron.

[0081] D. The final output uses LSTM and a Softmax classifier. The mathematical formula is as follows:

[0082]

[0083] Where z is i The output value of the i-th node, where C is the number of output nodes, i.e., the number of categories. The Softmax function can be used to convert the output values ​​of multi-class classification into a probability distribution ranging from [0,1] with a sum of 1.

[0084] The model is trained using backpropagation. During training, the initial learning rate is set to 0.01, and the number of iterations is set to 100. ReLU is selected as the activation function in the model, and its function expression is as follows:

[0085] ReLU(x) = max(0, 1)

[0086] The weights and biases of each layer of the network are adjusted using stochastic gradient descent, and the cross-entropy loss function is used.

[0087]

[0088] Where y is the true value; The predicted values ​​are saved from the trained model.

[0089] In step (4), during the training of the CNN-LSTM classifier, the RBF radial basis function is selected, and its function expression is as follows:

[0090]

[0091] Where, ‖x-x'‖ 2 It can be viewed as the squared Euclidean distance between two feature vectors; σ is a free parameter. The iteration count is set to infinitely many times until the optimal recognition rate is found.

[0092] Test data with unknown event types is first input into a trained CNN-LSTM network, which outputs feature vectors. Finally, these vectors are fed into a trained CNN-LSTM classifier for recognition, thus verifying the model's recognition performance.

[0093] In this embodiment, the confusion matrix after the data without SMOTE algorithm is fed into the CNN-LST M network is as follows: Figure 2 As shown, the confusion matrix after the data processed by the SMOTE algorithm is fed into the CNN-LSTM network is as follows. Figure 3 As shown, the data processed by the SMOTE algorithm performs better on the validation set.

[0094] Because the failure of an on-load tap changer has a huge impact on the transformer, the amount of fault sample data is relatively small. Expanding the fault sample event dataset can improve the accuracy of the identification model and effectively extract the deep features of the data.

[0095] Since the backscattered Rayleigh signal acquired by DAS is a one-dimensional vibration signal, it can be used as input after sample augmentation by the SMOT E algorithm, which can effectively avoid the feature loss problem caused by data transformation.

[0096] The recognition model uses LSTM instead of the softmax layer used for classification in the original network, thus achieving higher recognition accuracy for fault events.

[0097] Example 2:

[0098] A system for implementing the above method includes:

[0099] The signal acquisition module is configured to acquire the vibration signal of the on-load tap changer.

[0100] The detection module is configured to detect the running status based on the trained recognition model and the acquired vibration signal.

[0101] The identification model uses an augmented dataset of fault vibration signals from on-load tap changers as training input, specifically:

[0102] A faulty sample point is randomly selected as the original sample point. Using this sample point as the center, the K nearest sample points are searched. A new sample is generated between the original sample point and the searched sample point. The generated new sample and the original faulty sample are merged as the faulty dataset for training input.

[0103] Example 3:

[0104] This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in the on-load tap changer operation status detection method as described in Embodiment 1 above.

[0105] Example 4:

[0106] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the on-load tap changer operation status detection method as described in Embodiment 1 above.

[0107] The steps or modules involved in Embodiments 2 to 4 above correspond to those in Embodiment 1. For specific implementation details, please refer to the relevant description section of Embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood as including any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and enabling the processor to perform any of the methods in this invention.

[0108] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for detecting the operating state of a load tap changer, characterized in that The method comprises the following steps: The vibration signal of the on-load tap-changing transformer is acquired by the distributed optical fiber acoustic sensor, the optical fiber is distributed on the top and the cabinet of the OLTC, and the running state detection is realized based on the trained identification model and the acquired vibration signal; The identification model takes the expanded data set of the fault vibration signal of the on-load tap-changing transformer as the training input, and specifically: A fault sample point is randomly determined as an original sample point, K sample points closest to the original sample point are searched based on the original sample point as the center, a searched sample point is randomly selected, a new sample is generated between the original sample point and the searched sample point, and the generated new sample and the original fault sample are combined as the fault data set for training input. wherein, x n is the data to be generated, x i represents an original data point in the data set, β represents the range of the Euclidean distance between the generated data point and the original data point, x zi represents another point in the original data set.

2. The method for detecting the operating status of an on-load tap changer as described in claim 1, characterized in that, The training process of the identification model comprises the following steps:

3. The method of claim 2, wherein the operating state of the on-load tap changer is determined by comparing the first and second signals. The training process of the identification model further comprises the following steps:

4. The method of claim 3, wherein the operating state of the on-load tap changer is determined by the step of: The training process of the identification model further comprises the following steps:

5. The method for detecting the operating status of an on-load tap changer as described in claim 2, characterized in that, The pre-processing comprises data segmentation and data expansion, and the data segmentation specifically comprises the following steps:

6. The method for detecting the operating status of an on-load tap changer as described in claim 1, characterized in that, The expanded fault sample contains the same amount of data as the normal sample.

7. A system for detecting the operating state of a voltage regulating tap changer, characterized in that The method comprises the following steps: The detection module is configured to acquire the vibration signal of the on-load tap-changing transformer by the distributed optical fiber acoustic sensor, distribute the optical fiber on the top and the cabinet of the OLTC, and realize the running state detection based on the trained identification model and the acquired vibration signal. The identification model takes the expanded data set of the fault vibration signal of the on-load tap-changing transformer as the training input, and specifically: A fault sample point is randomly determined as an original sample point, K sample points closest to the original sample point are searched based on the original sample point as the center, a searched sample point is randomly selected, a new sample is generated between the original sample point and the searched sample point, and the generated new sample and the original fault sample are combined as the fault data set for training input. wherein, x n is the data to be generated, x i represents an original data point in the data set, β represents the range of the Euclidean distance between the generated data point and the original data point, x zi represents another point in the original data set. 8.A computer readable storage medium having stored thereon a computer program, which, when executed by a processor, implements the steps of the on-load tap-changing transformer running state detection method according to any one of claims 1 to 6. 9.A computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps of the on-load tap-changing transformer running state detection method according to any one of claims 1 to 6 when executing the program.

Images