Insulation deterioration diagnosis method for switch cabinet
Through the combination of BP neural network and CNN convolutional neural network model, the characteristic gas and image data of the switch cabinet are obtained and the diagnosis is carried out, which solves the problem of low diagnostic accuracy in the prior art, and significantly improves the diagnostic accuracy of switch cabinet insulation deterioration faults.
Patent Information
- Application Number
- CN202411575175.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-06
- Publication Date
- 2025-03-21
AI Technical Summary
The existing fault diagnosis method for insulation deterioration of switch cabinets has characteristic gas analysis that requires a large amount of manual data processing and is not universal. Although deep learning algorithms are universal, they cannot accurately distinguish the deterioration faults of similar characteristic gases, resulting in the inability to guarantee the diagnostic accuracy.
The BP neural network model is used to combine the CNN convolutional neural network model to obtain the characteristic gas data and image data of the switch cabinet for diagnosis. The BP neural network model is used for preliminary diagnosis. If the diagnosis is prone to misjudgment deterioration faults, the image data will be further diagnosed through the CNN model.
The diagnostic accuracy of switch cabinet insulation deterioration faults is significantly improved, especially when the characteristic gas composition and concentration are similar, the possibility of misjudgment is reduced through unique diagnostic result identification and image recognition.
Smart Images
Figure CN119669812A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of switch cabinet fault diagnosis, and in particular to a switch cabinet insulation degradation diagnosis method. Background Art
[0002] Switchgear is an essential and numerous basic equipment in the operation of power systems. When switchgear frequently fails, the stable operation of the power grid is challenged. Therefore, how to ensure the long-term safe operation of switchgear has become a technical problem that must be solved.
[0003] When a switch cabinet has a degradation discharge fault, fault characteristic gases such as O3, NO, NO2, and N2O will be generated inside. Different insulation degradation faults such as corona discharge, surface discharge, air gap discharge, and suspension discharge have different discharge sizes and locations, resulting in different composition and content of characteristic gases. At present, it is possible to roughly identify the type of switch cabinet insulation degradation fault by analyzing the composition and content of characteristic gases. The commonly used methods are deep learning algorithms or data feature analysis algorithms.
[0004] Its shortcomings are: 1. The feature analysis algorithm requires a lot of manual data analysis and processing, which is not universal for different environments and switch cabinets, and is time-consuming and labor-intensive with poor universality. 2. Although the deep learning algorithm can learn autonomously, has strong universality and saves time and effort, it cannot accurately distinguish degradation faults with similar characteristic gases, and the diagnostic accuracy cannot be guaranteed. Summary of the invention
[0005] The present invention discloses a switch cabinet insulation degradation diagnosis method. The specific method is as follows:
[0006] Obtain characteristic data of characteristic gas of switch cabinet;
[0007] Diagnose characteristic data of characteristic gases using BP neural network model;
[0008] If the BP neural network model diagnoses a specific degradation fault category, the degradation fault category is used as the diagnosis result;
[0009] If the BP neural network model diagnoses a degradation fault that is easily misjudged, the components of the degradation fault that is easily misjudged are obtained; according to the components of the degradation fault that is easily misjudged, image data of the corresponding position of the switch cabinet is photographed; the image data is diagnosed with the CNN convolutional neural network model to obtain a diagnosis result.
[0010] The advantage of this embodiment is that two or more different insulation failures may have similar characteristic gas compositions and concentrations due to different discharge amounts. The BP neural network is specially designed with a unique diagnostic result identifier for this situation, which can significantly improve the diagnostic accuracy.
[0011] Further, characteristic gas composition and concentration.
[0012] Furthermore, specific degradation fault categories include:
[0013] Corona discharge, surface discharge, air gap discharge and suspension discharge.
[0014] Furthermore, a BP neural network model is constructed, and the specific method is as follows:
[0015] Conduct insulation degradation test on target switchgear, record test data, classify and process the test data, and obtain training data;
[0016] Initialize the BP neural network model;
[0017] The BP neural network model is trained with training data.
[0018] Furthermore, the BP neural network model is initialized. The specific method is as follows:
[0019] Set the number of input layer nodes, the number of hidden layer nodes, the number of output layer nodes, the input layer to hidden layer weights, the hidden layer to output layer weights, the input layer to hidden layer bias, the hidden layer to output layer bias, the learning rate and the activation function.
[0020] Furthermore, the test data is classified and processed to obtain training data. The specific method is as follows:
[0021] For different insulation degradation faults, discharge is performed at equal intervals, and the characteristic gas composition and concentration of each insulation degradation fault at each level of discharge are recorded. The fault feature array is constructed with the characteristic gas composition and concentration data.
[0022] Clustering algorithm is used to cluster the fault feature array and divide it into several similar fault intervals;
[0023] A preset distance is drawn from the center of each similar fault interval, and the fault feature array belonging to the same similar fault interval within the preset distance range is classified as a misjudged degradation fault, and the misjudged degradation fault composition is marked;
[0024] The fault feature array outside the preset distance of the similar fault interval is labeled with the corresponding degradation fault category;
[0025] The training data is constructed by taking the average value of the fault feature array corresponding to each easily misjudged degradation fault and the fault feature array corresponding to each degradation fault category.
[0026] The advantage of this embodiment is that degradation faults that are easy to misjudge are divided into the same interval through a clustering algorithm, and unified remarks are made, which significantly reduces the possibility of misjudgment.
[0027] Furthermore, when the degradation fault components that are easily misjudged are corona discharge and creeping discharge, the location of the captured image data is on the surface of the cable, bushing and insulator;
[0028] When the degradation fault components that are easily misjudged are corona discharge and air gap discharge, the location where the image data is taken is on the inner surface of the cabinet;
[0029] When the degradation fault components that are easily misjudged are corona discharge and suspension discharge, the location where the image data is taken is at the grounded part;
[0030] When the degradation fault components that are easily misjudged are surface discharge and air gap discharge, the location of the image data is on the surface of the cable, bushing and insulator;
[0031] When the degradation fault components that are easily misjudged are creeping discharge and suspended discharge, the location where the image data is taken is at the grounded part;
[0032] When the degradation fault components that are easily misjudged are air gap discharge and suspended discharge, the position where the image data is captured is at the grounded part.
[0033] The advantage of this embodiment is that, according to the characteristics of the four types of degradation discharge, the exact location of the captured image data is determined, and further diagnosis of degradation faults that are easily misjudged can be completed through image recognition, thereby significantly improving the diagnostic accuracy.
[0034] Furthermore, the CNN convolutional neural network model is constructed as follows:
[0035] Determine input data and output data;
[0036] Determine the number of hidden layers and neurons in each layer;
[0037] Initialize weights;
[0038] Set the activation function and loss function;
[0039] Add optimizer;
[0040] Train CNN convolutional neural network.
[0041] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be achieved and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS
[0042] The accompanying drawings of the present invention are as follows.
[0043] Figure 1 It is a schematic diagram of the overall process of the present invention.
[0044] Figure 2 This is a schematic diagram of the process of training the BP neural network model. DETAILED DESCRIPTION
[0045] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.
[0046] A switch cabinet insulation degradation diagnosis method, such as Figure 1 As shown, the specific steps are as follows:
[0047] S1. Acquire characteristic data of characteristic gas of the switch cabinet.
[0048] Specifically, the characteristic data of the characteristic gas includes characteristic gas composition and concentration.
[0049] In this embodiment, specific degradation fault categories include corona discharge, creeping discharge, air gap discharge and suspension discharge.
[0050] S2. Diagnose characteristic data of characteristic gas using BP neural network model.
[0051] In step S2, the BP neural network model is trained, such as Figure 2 As shown, the specific method is as follows:
[0052] S21. Perform an insulation degradation test on the target switch cabinet, record the test data, classify the test data, and obtain training data.
[0053] Step S21 obtains training data, and the specific steps are as follows:
[0054] S211. For different insulation degradation faults, discharge is performed at equal intervals, and characteristic gas composition and concentration of each insulation degradation fault at each level of library discharge are recorded, and a fault characteristic array is constructed with characteristic gas composition and concentration data.
[0055] For example, the first set of data is corona discharge, the discharge amount is C1 coulomb, the composition and concentration of the characteristic gas are O3: a1, NO: b1, NO2: c1, N2O: d1, and similarly, several fault characteristic arrays are obtained.
[0056] S212: clustering the fault feature array using a clustering algorithm to divide it into several similar fault intervals.
[0057] In this embodiment, the fault feature array constructed by characteristic gas concentration is calculated using Euclidean distance. The Euclidean distance calculation formula is as follows:
[0058]
[0059] S213, a preset distance is drawn with the center of each similar fault interval, and the fault feature array belonging to the same similar fault interval within the preset distance range is classified as a degradation fault that is easy to be misjudged, and the degradation fault composition that is easy to be misjudged is marked.
[0060] S214. The fault feature array outside the preset distance of the similar fault interval is labeled with the corresponding degradation fault category.
[0061] For example, if the surface discharge with a discharge amount of x and the air gap discharge with a discharge amount of y are similar in composition and content of the characteristic gas, they are marked as: easy to misjudge degradation fault - surface discharge with a discharge amount of x and air gap discharge with a discharge amount of y, and the average value of the characteristic gas obtained by the two discharges is used as the characteristic data.
[0062] S215 , constructing training data using the average value of the fault feature array corresponding to each easily misjudged degradation fault and the fault feature array corresponding to each degradation fault category.
[0063] S22, initialize the BP neural network model.
[0064] Initializing the BP neural network model in step S22 includes:
[0065] Set the number of input layer nodes: n
[0066] Number of hidden layer nodes: l
[0067] Number of output layer nodes: m
[0068] Input layer to hidden layer weight: ω ij
[0069] Hidden layer to output layer weight: ω jk
[0070] Input layer to hidden layer bias: a j
[0071] Bias from hidden layer to output layer: a k
[0072] Learning rate: σ
[0073] Activation function:
[0074] The hidden layer output is:
[0075]
[0076] The output of the output layer is:
[0077]
[0078] The misjudgment calculation formula is:
[0079]
[0080] where Y k The expected output is:
[0081] Y k -O k =e k
[0082] but
[0083]
[0084] Back-propagation is used to update weights and biases.
[0085] S22, training the BP neural network model using the training data.
[0086] S3. If the BP neural network model diagnoses a specific degradation fault category, the degradation fault category is used as the diagnosis result.
[0087] S4. If the BP neural network model diagnoses a degradation fault that is prone to misjudgment, take image data of the corresponding position of the switch cabinet and diagnose it using the CNN convolutional neural network model.
[0088] The shooting position is determined as follows:
[0089] When the degradation fault components that are easily misjudged are corona discharge and air gap discharge, the location for capturing image data is on the inner surface of the cabinet.
[0090] When the degradation fault components that are easily misjudged are corona discharge and suspension discharge, the position where the image data is captured is at the grounded part.
[0091] When the degradation fault components that are easily misjudged are surface discharge and air gap discharge, the location where the image data is taken is on the surface of the cable, bushing and insulator.
[0092] When the degradation fault components that are easily misjudged are creeping discharge and floating discharge, the position where the image data is captured is at the grounded part.
[0093] When the degradation fault components that are easily misjudged are air gap discharge and suspended discharge, the position where the image data is captured is at the grounded part.
[0094] For example, the situation obtained is that the degradation fault - x discharge amount is surface discharge and y discharge amount is air gap discharge, so the location of the image data is on the surface of the cable, bushing and insulator.
[0095] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.
Claims
1. A switch cabinet insulation degradation diagnosis method, characterized in that: The specific method is as follows: Obtain characteristic data of characteristic gas of switch cabinet; Diagnose characteristic data of characteristic gases using BP neural network model; If the BP neural network model diagnoses a specific degradation fault category, the degradation fault category is used as the diagnosis result; If the BP neural network model diagnoses a degradation fault that is easily misjudged, the components of the degradation fault that is easily misjudged are obtained; according to the components of the degradation fault that is easily misjudged, image data of the corresponding position of the switch cabinet is photographed; the image data is diagnosed with the CNN convolutional neural network model to obtain a diagnosis result.
2. The switch cabinet insulation degradation diagnosis method according to claim 1, characterized in that: The characteristic data of the characteristic gas includes: characteristic gas composition and concentration.
3. The switch cabinet insulation degradation diagnosis method according to claim 1, characterized in that: Specific degradation fault categories include: Corona discharge, surface discharge, air gap discharge and suspension discharge.
4. The switch cabinet insulation degradation diagnosis method according to claim 1, characterized in that: Construct the BP neural network model. The specific method is as follows: Conduct insulation degradation test on target switchgear, record test data, classify and process the test data, and obtain training data; Initialize the BP neural network model; The BP neural network model is trained with training data.
5. The switch cabinet insulation degradation diagnosis method according to claim 4, characterized in that: Initialize the BP neural network model. The specific method is as follows: Set the number of input layer nodes, the number of hidden layer nodes, the number of output layer nodes, the input layer to hidden layer weights, the hidden layer to output layer weights, the input layer to hidden layer bias, the hidden layer to output layer bias, the learning rate and the activation function.
6. The switch cabinet insulation degradation diagnosis method according to claim 4, characterized in that: Classify the test data and obtain training data. The specific method is as follows: For different insulation degradation faults, discharge is performed at equal intervals, and the characteristic gas composition and concentration of each insulation degradation fault at each level of discharge are recorded. The fault feature array is constructed with the characteristic gas composition and concentration data. Clustering algorithm is used to cluster the fault feature array and divide it into several similar fault intervals; A preset distance is drawn from the center of each similar fault interval, and the fault feature array belonging to the same similar fault interval within the preset distance range is classified as a misjudged degradation fault, and the misjudged degradation fault composition is marked; The fault feature array outside the preset distance of the similar fault interval is labeled with the corresponding degradation fault category; The training data is constructed by taking the average value of the fault feature array corresponding to each easily misjudged degradation fault and the fault feature array corresponding to each degradation fault category.
7. The switch cabinet insulation degradation diagnosis method according to claim 3, characterized in that: When the degradation fault components that are easily misjudged are corona discharge and surface discharge, the location of the image data is on the surface of the cable, bushing and insulator; When the degradation fault components that are easily misjudged are corona discharge and air gap discharge, the location where the image data is taken is on the inner surface of the cabinet; When the degradation fault components that are easily misjudged are corona discharge and suspension discharge, the location where the image data is taken is at the grounded part; When the degradation fault components that are easily misjudged are surface discharge and air gap discharge, the location of the image data is on the surface of the cable, bushing and insulator; When the degradation fault components that are easily misjudged are creeping discharge and suspended discharge, the location where the image data is taken is at the grounded part; When the degradation fault components that are easily misjudged are air gap discharge and suspended discharge, the position where the image data is captured is at the grounded part.
8. The switch cabinet insulation degradation diagnosis method according to claim 1, characterized in that: The CNN convolutional neural network model is constructed as follows: Determine input data and output data; Determine the number of hidden layers and neurons in each layer; Initialize weights; Set the activation function and loss function; Add optimizer; Train CNN convolutional neural network.