On-load tap-changer on-line fault monitoring method and system based on electrical signals
By training the classification model of one-dimensional convolutional neural network, the electrical signals of on-load tap-offs are classified, which solves the inefficiency problem of relying on manual selection of feature indicators in the existing technology, and realizes automated fault monitoring.
Patent Information
- Application Number
- CN202411481525.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-23
- Publication Date
- 2025-02-28
AI Technical Summary
The existing online fault monitoring methods for load tap-offs mainly rely on vibration signal monitoring and require manual selection of characteristic indicators, which are inefficient and not automated enough.
Using an electrical signal-based method, a classification data set of on-load tap-off switch electrical signals with classification labels is obtained, and a classification model is obtained, which is used to classify actual electrical data and realize fault monitoring.
It realizes automatic real-time acquisition of electrical signals and fault monitoring without power outage, and the monitoring process is automated without manual selection of characteristic indicators, which improves the accuracy and efficiency of monitoring.
Smart Images
Figure CN119538027A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of transformer fault monitoring, and in particular to an on-load tap changer online fault monitoring method and system based on electrical signals. Background Art
[0002] The on-load tap changer is an important component of the on-load voltage regulating transformer. Accurately testing its performance is particularly important to ensure the safe, stable and reliable operation of the power system.
[0003] Existing fault monitoring technologies for on-load tap changers are divided into offline monitoring and online monitoring. Offline monitoring is to monitor the relevant performance indicators of the on-load tap changer when it is not in operation, which needs to be performed under power outages; while the online monitoring method in the existing technology generally monitors the vibration signal and manually selects the characteristic indicators to obtain the fault situation. Summary of the invention
[0004] The present invention provides an on-load tap changer online fault monitoring method and system based on electrical signals, which can accurately monitor the faults of the on-load tap changer through electrical signals during grid-connected operation.
[0005] In order to solve the above technical problems, an embodiment of the present invention provides an on-load tap changer online fault monitoring method based on electrical signals, comprising:
[0006] Obtaining a classification dataset of on-load tapchanger electrical signals with classification labels;
[0007] The classification data set is used to train an initial neural network to obtain a classification model; the initial neural network is a one-dimensional convolutional neural network;
[0008] The classification model is used to classify actual electrical data of the on-load tap changer to obtain a fault monitoring result.
[0009] As an improvement of the above solution, the step of obtaining a classification data set of electrical signals of an on-load tap changer with classification labels includes:
[0010] Obtaining an initial data set of on-load factoring switches;
[0011] According to the typical faults of the on-load tap changer, classification labels are added to the initial data set to obtain a classification data set.
[0012] As an improvement of the above solution, the step of obtaining an initial data set of the on-load split switch includes:
[0013] Collect historical data sets of on-load tap-changer electrical signals;
[0014] Construct a simulation model based on the voltage regulation principle, set simulation parameters to conduct simulation experiments, and obtain a simulation data set;
[0015] Set the operating parameters of the on-load tap-changing transformer, conduct actual tap-changing test, and obtain the test data set;
[0016] An initial data set of electrical signals of the on-load tap changer is obtained according to the historical data set, the simulation data set and the test data set.
[0017] As an improvement of the above solution, the initial neural network includes a convolutional layer, a pooling layer and a fully connected layer connected in sequence; the convolutional layer is a one-dimensional neural network.
[0018] As an improvement of the above solution, the classification data set is used to train an initial neural network to obtain a classification model, including:
[0019] Preprocessing the classification data set to obtain a preprocessed classification data set;
[0020] Dividing the preprocessed classification data set into a training set and a validation set;
[0021] Using the training set, iteratively training the initial neural network to obtain a prediction result;
[0022] After each iteration of training is completed, the error of the prediction result is calculated using the verification set;
[0023] When the error of the prediction result is less than a preset error threshold, the iterative training of the initial neural network is stopped, and the trained initial neural network is used as a classification model.
[0024] As an improvement of the above solution, the adopting the training set to iteratively train the initial neural network includes:
[0025] Extracting a feature map of the preprocessed classification data set through the convolutional layer;
[0026] By means of the pooling layer, the size of the feature map is reduced to obtain a pooling feature;
[0027] Through the fully connected layer, the pooled features are mapped to n dimensions to obtain a prediction result.
[0028] As an improvement of the above solution, extracting the feature map of the preprocessed classification data set through the convolution layer includes:
[0029] Performing depth convolution and point-by-point convolution operations on the preprocessed classification data set through the one-dimensional depth convolution kernel of the convolution layer to obtain local features;
[0030] The local features are weighted by the sliding convolution kernel of the convolution layer to obtain a feature map.
[0031] As an improvement of the above solution, after each iteration of training is completed, the error of the prediction result is calculated using the verification set, including:
[0032] Establish a generalized cross entropy loss function that is robust to label noise;
[0033] According to the verification set, a true result is obtained;
[0034] After each iteration of training is completed, the generalized cross entropy loss function is used to calculate the error between the predicted result and the true result.
[0035] As an improvement of the above solution, the classification model is used to classify the actual electrical data of the on-load tap changer to obtain the fault monitoring result, including:
[0036] Collect actual electrical data of on-load tap-changer;
[0037] Preprocessing the actual electrical data to obtain preprocessed actual electrical data;
[0038] The pre-processed actual electrical data is input into the classification model to obtain a fault monitoring result.
[0039] The embodiment of the present invention further provides an on-load tap changer online fault monitoring system based on electrical signals, comprising:
[0040] A classification data set acquisition module, used for acquiring a classification data set of electrical signals of the on-load tap changer with classification labels;
[0041] A classification model building module, used to train an initial neural network using the classification data set to obtain a classification model; the initial neural network is a one-dimensional convolutional neural network;
[0042] The fault monitoring module is used to classify the actual electrical data of the on-load tap changer using the classification model to obtain a fault monitoring result.
[0043] Compared with the prior art, the present invention discloses an on-load tap changer online fault monitoring method and system based on electrical signals, which obtains a classification data set of electrical signals of the on-load tap changer with classification labels; uses the classification data set to train an initial neural network to obtain a classification model; the initial neural network is a one-dimensional convolutional neural network; uses the classification model to classify the actual electrical data of the on-load tap changer to obtain a fault monitoring result. By using the embodiment of the present invention, electrical signals can be automatically collected in real time without power outages, and the classification model obtained by neural network training can be used for fault monitoring. The monitoring process is automatic and simple, and there is no need to manually select feature indicators. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] Figure 1 It is a schematic flow chart of the steps of an on-load tap changer online fault monitoring method based on electrical signals provided by an embodiment of the present invention;
[0045] Figure 2 The present invention is a schematic structural diagram of an on-load tap changer online fault monitoring system based on electrical signals provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0046] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0047] In the description of the specification and claims, it is to be understood that the terms first, second, etc. in the specification and claims are only used for the purpose of describing the same technical features, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated, nor necessarily describing the order or time sequence. The terms are interchangeable where appropriate. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features.
[0048] The embodiment of the present invention provides a schematic flow chart of the steps of an on-load tap changer online fault monitoring method based on electrical signals. Figure 1 In this embodiment, the on-load tap changer online fault monitoring method based on electrical signals is specifically performed through steps S1 to S3:
[0049] S1. Obtain a classification dataset of electrical signals of on-load tap changers with classification labels.
[0050] It should be noted that the classification data set is used to train a classification model. By training the model with the classification data set with the classification label, the model can be trained to judge whether there is a fault and the fault type. In the embodiment of the present invention, the classification label only needs to indicate whether there is a fault and the typical fault type.
[0051] S2. Using the classification data set to train an initial neural network to obtain a classification model; the initial neural network is a one-dimensional convolutional neural network.
[0052] The embodiment of the present invention trains a one-dimensional convolutional neural network to obtain a classification model that can automatically extract the features of electrical data without manually selecting feature indicators. The method of manually selecting feature indicators depends largely on the experience of the operator, but in the rapid update and iteration process of development, the feature indicators of the fault will not remain unchanged. The traditional manual method not only requires high professional knowledge, but also has difficulty adapting to changes.
[0053] It should be noted that the classification model can obtain the classification results corresponding to the classification labels based on the classification data set. If there are several data in the classification data set showing similar characteristics, the classification model can also obtain classification results in addition to the classification labels. New classification results generally require manual verification and confirmation before they can be applied.
[0054] S3. Using the classification model to classify actual electrical data of the on-load tap changer to obtain a fault monitoring result.
[0055] It should be noted that the fault monitoring result is generated based on actual electrical data, and the fault monitoring result may include a classification result obtained by the classification model, and may also include an operating state of the on-load tap changer obtained according to the classification result.
[0056] In the above scheme, unlike the traditional mechanism analysis method which requires manual construction of multiple characteristic indicators for analysis, the professional barriers are reduced. By extracting the characteristics of the transformer electrical signal during the switching process of the on-load tap changer, the neural network is trained to realize the classification function, thereby achieving the goal of judging whether the operating status of the on-load tap changer is normal or faulty based on the actual electrical data.
[0057] As a preferred implementation, step S1, obtaining a classification data set of electrical signals of on-load tap changers with classification labels, comprises:
[0058] Obtaining an initial data set of on-load factoring switches;
[0059] According to the typical faults of the on-load tap changer, classification labels are added to the initial data set to obtain a classification data set.
[0060] Further, preferably, the obtaining of an initial data set of the on-load split switch comprises:
[0061] Collect historical data sets of on-load tap-changer electrical signals;
[0062] Construct a simulation model based on the voltage regulation principle, set simulation parameters to conduct simulation experiments, and obtain a simulation data set;
[0063] Set the operating parameters of the on-load tap-changing transformer, conduct actual tap-changing test, and obtain the test data set;
[0064] An initial data set of electrical signals of the on-load tap changer is obtained according to the historical data set, the simulation data set and the test data set.
[0065] Exemplarily, a simulation model is built according to the voltage regulation principle of an on-load tap-changing transformer, and simulation parameters are set to perform simulation tests to obtain a simulation data set. According to the on-load tap-changing transformer in the laboratory, the operating parameters of the on-load tap-changing transformer are set, and an actual voltage regulation test is performed to obtain a test data set. In some embodiments, the simulation parameters and the test operating parameters are calibrated according to the simulation data set and the test data set to obtain a more accurate data set. The data set is used to train the initial neural network to obtain a classification model on the one hand, and is used to calculate and verify the accuracy of the classification model on the other hand.
[0066] As a preferred implementation, the initial neural network includes a convolutional layer, a pooling layer and a fully connected layer connected in sequence; the convolutional layer is a one-dimensional neural network.
[0067] In some embodiments, the initial neural network further includes a loss function for calculating the error between the predicted output and the actual output.
[0068] It should be noted that when training the initial neural network, the learning rate and the maximum number of iterations need to be preset. Once the preset requirements are met, the initial neural network training can be considered completed.
[0069] In the above scheme, the convolution layer is used to extract the features of the data set to obtain the features corresponding to different fault types; the pooling layer is used to reduce the size of the feature map to reduce the amount of calculation and prevent overfitting; the fully connected layer is used to combine all the outputs of the pooling layer and classify the features.
[0070] Further, as a preferred implementation, step S2, using the classification data set to train an initial neural network to obtain a classification model, includes:
[0071] Preprocessing the classification data set to obtain a preprocessed classification data set;
[0072] Dividing the preprocessed classification data set into a training set and a validation set;
[0073] Using the training set, iteratively training the initial neural network to obtain a prediction result;
[0074] After each iteration of training is completed, the error of the prediction result is calculated using the verification set;
[0075] When the error of the prediction result is less than a preset error threshold, the iterative training of the initial neural network is stopped, and the trained initial neural network is used as a classification model.
[0076] Preferably, the preprocessing of the classification data set to obtain the preprocessed classification data set includes:
[0077] The classification data set is cleaned and normalized to obtain a preprocessed classification data set.
[0078] In the above scheme, data cleaning is performed on the classification data set to remove noise, missing values and outliers from the classification data set; data normalization is performed on the classification data set to make the data distribution more uniform, so that the classification model trained in this way is easier to converge.
[0079] Furthermore, preferably, the adopting the verification set to iteratively train the initial neural network comprises:
[0080] Extracting a feature map of the preprocessed classification data set through the convolutional layer;
[0081] By means of the pooling layer, the size of the feature map is reduced to obtain a pooling feature;
[0082] Through the fully connected layer, the pooled features are mapped to n dimensions to obtain a prediction result.
[0083] As a preferred implementation, extracting the feature graph of the preprocessed classification data set through the convolution layer includes:
[0084] Performing depth convolution and point-by-point convolution operations on the preprocessed classification data set through the one-dimensional depth convolution kernel of the convolution layer to obtain local features;
[0085] The local features are weighted by the sliding convolution kernel of the convolution layer to obtain a feature map.
[0086] Exemplarily, a one-dimensional depthwise separable convolution kernel is applied to perform depthwise convolution and point-wise convolution on the input sequence, and a feature map is generated by sliding the convolution kernel and calculating the weighted sum of the local area. Then, a one-dimensional learnable pooling operation is applied to reduce the size of the feature map, reduce the amount of calculation and prevent overfitting. The features extracted by the convolution layer and the pooling layer are flattened and classified through the fully connected layer, and finally a trained classification model is obtained.
[0087] As a preferred implementation, after each iteration of training is completed, the error of the prediction result is calculated using the validation set, including:
[0088] Establish a generalized cross entropy loss function that is robust to label noise;
[0089] According to the verification set, a true result is obtained;
[0090] After each iteration of training is completed, the generalized cross entropy loss function is used to calculate the error between the predicted result and the true result.
[0091] For example, error prediction is performed using the Generalized Cross-Entropy Loss loss function, which introduces an additional parameter to balance different types of losses compared to the traditional cross entropy loss function, and can provide better performance and adaptability.
[0092] Preferably, step S3, using the classification model to classify actual electrical data of the on-load tap changer to obtain a fault monitoring result, includes:
[0093] Collect actual electrical data of on-load tap-changer;
[0094] Preprocessing the actual electrical data to obtain preprocessed actual electrical data;
[0095] The pre-processed actual electrical data is input into the classification model to obtain a fault monitoring result.
[0096] It should be noted that after the actual electrical data is preprocessed, the trained classification network is called to give a classification result for the input data, so that the operating status of the on-load tap changer can be known.
[0097] It should also be noted that, in the embodiment of the present invention, the method for preprocessing the actual electrical data is consistent with the method for preprocessing the classified data set.
[0098] The on-load tap changer online fault monitoring method based on electrical signals provided by the embodiment of the present invention can automatically collect electrical signals in real time without power outage and use the classification model obtained by neural network training to perform fault monitoring. The monitoring process is automatic and simple, and there is no need to manually select characteristic indicators.
[0099] See also Figure 2 The embodiment of the present invention provides an on-load tap changer online fault monitoring system based on electrical signals. The on-load tap changer online fault monitoring system based on electrical signals comprises a classification data set acquisition module 11, a classification model construction module 12 and a fault monitoring module 13, wherein:
[0100] A classification data set acquisition module 11 is used to acquire a classification data set of electrical signals of the on-load tap changer with classification labels;
[0101] A classification model building module 12 is used to train an initial neural network using the classification data set to obtain a classification model; the initial neural network is a one-dimensional convolutional neural network;
[0102] The fault monitoring module 13 is used to classify the actual electrical data of the on-load tap changer using the classification model to obtain a fault monitoring result.
[0103] As a preferred implementation, the classification data set acquisition module 11 includes:
[0104] An initial data set acquisition unit, used for acquiring an initial data set of the on-load split switch;
[0105] The classification label adding unit is used to add classification labels to the initial data set according to the typical faults of the on-load tap changer to obtain a classification data set.
[0106] Further, preferably, the initial data set acquisition unit includes:
[0107] Collect historical data sets of on-load tap-changer electrical signals;
[0108] Construct a simulation model based on the voltage regulation principle, set simulation parameters to conduct simulation experiments, and obtain a simulation data set;
[0109] Set the operating parameters of the on-load tap-changing transformer, conduct actual tap-changing test, and obtain the test data set;
[0110] An initial data set of electrical signals of the on-load tap changer is obtained according to the historical data set, the simulation data set and the test data set.
[0111] As a preferred implementation, the initial neural network includes a convolutional layer, a pooling layer and a fully connected layer connected in sequence; the convolutional layer is a one-dimensional neural network.
[0112] As a preferred implementation, the classification model building module 12 includes:
[0113] A first preprocessing unit, used to preprocess the classification data set to obtain a preprocessed classification data set;
[0114] A data partitioning unit, used for partitioning the preprocessed classification data set into a training set and a validation set;
[0115] An iterative training unit, used to use the training set to iteratively train the initial neural network to obtain a prediction result;
[0116] An error calculation unit, used to calculate the error of the prediction result using the verification set after each iterative training is completed;
[0117] The classification model generating unit is used to stop the iterative training of the initial neural network when the error of the prediction result is less than a preset error threshold, and use the trained initial neural network as the classification model.
[0118] Further, as a preferred implementation, the iterative training unit is specifically used for:
[0119] Extracting a feature map of the preprocessed classification data set through the convolutional layer;
[0120] By means of the pooling layer, the size of the feature map is reduced to obtain a pooling feature;
[0121] Through the fully connected layer, the pooled features are mapped to n dimensions to obtain a prediction result.
[0122] More preferably, extracting the feature map of the preprocessed classification data set through the convolution layer includes:
[0123] Performing depth convolution and point-by-point convolution operations on the preprocessed classification data set through the one-dimensional depth convolution kernel of the convolution layer to obtain local features;
[0124] The local features are weighted by the sliding convolution kernel of the convolution layer to obtain a feature map.
[0125] As a preferred implementation, the error calculation unit is specifically used to:
[0126] Establish a generalized cross entropy loss function that is robust to label noise;
[0127] According to the verification set, a true result is obtained;
[0128] After each iteration of training is completed, the generalized cross entropy loss function is used to calculate the error between the predicted result and the true result.
[0129] As a preferred implementation, the fault monitoring module 13 includes:
[0130] Actual data acquisition unit, used to collect actual electrical data of the on-load tap-changer;
[0131] Preprocessing the actual electrical data to obtain preprocessed actual electrical data;
[0132] The pre-processed actual electrical data is input into the classification model to obtain a fault monitoring result.
[0133] The on-load tap changer online fault monitoring system based on electrical signals provided by the embodiment of the present invention can automatically collect electrical signals in real time without power outage and use the classification model obtained by neural network training to perform fault monitoring. The monitoring process is automatic and simple, and there is no need to manually select characteristic indicators.
[0134] The present invention may take the form of a computer program product implemented on one or more storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing program code. Computer-readable storage media include permanent and non-permanent, removable and non-removable media, and information storage can be achieved by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include but are not limited to: phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, read-only compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device.
[0135] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.
Claims
1. A method for online fault monitoring of on-load tap changer based on electrical signals, characterized in that: include: Obtaining a classification dataset of on-load tapchanger electrical signals with classification labels; The classification data set is used to train an initial neural network to obtain a classification model; the initial neural network is a one-dimensional convolutional neural network; The classification model is used to classify actual electrical data of the on-load tap changer to obtain a fault monitoring result.
2. The method for online fault monitoring of an on-load tap changer based on electrical signals according to claim 1, characterized in that: The step of obtaining a classification data set of an on-load tap changer electrical signal with a classification label includes: Obtaining an initial data set of on-load factoring switches; According to the typical faults of the on-load tap changer, classification labels are added to the initial data set to obtain a classification data set.
3. The method for online fault monitoring of an on-load tap changer based on electrical signals according to claim 2, characterized in that: The step of obtaining an initial data set of the on-load split switch comprises: Collect historical data sets of on-load tap-changer electrical signals; Construct a simulation model based on the voltage regulation principle, set simulation parameters to conduct simulation experiments, and obtain a simulation data set; Set the operating parameters of the on-load tap-changing transformer, conduct actual tap-changing test, and obtain the test data set; An initial data set of electrical signals of the on-load tap changer is obtained according to the historical data set, the simulation data set and the test data set.
4. The method for online fault monitoring of on-load tap changer based on electrical signals according to claim 1, characterized in that: The initial neural network includes a convolutional layer, a pooling layer and a fully connected layer connected in sequence; the convolutional layer is a one-dimensional neural network.
5. The method for online fault monitoring of on-load tap changer based on electrical signals according to claim 4, characterized in that: The method of using the classification data set to train an initial neural network to obtain a classification model includes: Preprocessing the classification data set to obtain a preprocessed classification data set; Dividing the preprocessed classification data set into a training set and a validation set; Using the training set, iteratively training the initial neural network to obtain a prediction result; After each iteration of training is completed, the error of the prediction result is calculated using the verification set; When the error of the prediction result is less than a preset error threshold, the iterative training of the initial neural network is stopped, and the trained initial neural network is used as a classification model.
6. The method for online fault monitoring of an on-load tap changer based on electrical signals according to claim 5, characterized in that: The adopting the training set to iteratively train the initial neural network to obtain a prediction result includes: Extracting a feature map of the preprocessed classification data set through the convolutional layer; By means of the pooling layer, the size of the feature map is reduced to obtain a pooling feature; Through the fully connected layer, the pooled features are mapped to n dimensions to obtain a prediction result.
7. The method for online fault monitoring of on-load tap changer based on electrical signals according to claim 6, characterized in that: The step of extracting a feature map of the preprocessed classification data set through the convolution layer includes: Performing depth convolution and point-by-point convolution operations on the preprocessed classification data set through the one-dimensional depth convolution kernel of the convolution layer to obtain local features; The local features are weighted by the sliding convolution kernel of the convolution layer to obtain a feature map.
8. The method for online fault monitoring of an on-load tap changer based on electrical signals according to claim 5, characterized in that: After each iteration of training is completed, the error of the prediction result is calculated using the verification set, including: Establish a generalized cross entropy loss function that is robust to label noise; According to the verification set, a true result is obtained; After each iteration of training is completed, the generalized cross entropy loss function is used to calculate the error between the predicted result and the true result.
9. The method for online fault monitoring of on-load tap changer based on electrical signals according to claim 1, characterized in that: The method of using the classification model to classify the actual electrical data of the on-load tap changer to obtain the fault monitoring result includes: Collect actual electrical data of on-load tap-changer; Preprocessing the actual electrical data to obtain preprocessed actual electrical data; The pre-processed actual electrical data is input into the classification model to obtain a fault monitoring result.
10. An on-line fault monitoring system for on-load tap changer based on electrical signals, characterized in that: include: A classification data set acquisition module, used for acquiring a classification data set of an on-load tap changer electrical signal with a classification label; A classification model building module, used to train an initial neural network using the classification data set to obtain a classification model; the initial neural network is a one-dimensional convolutional neural network; The fault monitoring module is used to classify the actual electrical data of the on-load tap changer using the classification model to obtain a fault monitoring result.