CN116011412A - GIS equipment insulation defect evaluation method, system, equipment and medium - Google Patents
GIS equipment insulation defect evaluation method, system, equipment and medium Download PDFInfo
- Publication number
- CN116011412A CN116011412A CN202211708828.3A CN202211708828A CN116011412A CN 116011412 A CN116011412 A CN 116011412A CN 202211708828 A CN202211708828 A CN 202211708828A CN 116011412 A CN116011412 A CN 116011412A
- Authority
- CN
- China
- Prior art keywords
- partial discharge
- vector
- state
- feature
- feature vectors
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000007547 defect Effects 0.000 title claims abstract description 75
- 238000011156 evaluation Methods 0.000 title claims abstract description 63
- 238000009413 insulation Methods 0.000 title claims abstract description 47
- 239000013598 vector Substances 0.000 claims abstract description 219
- 238000009826 distribution Methods 0.000 claims abstract description 43
- 238000000034 method Methods 0.000 claims abstract description 38
- 238000013528 artificial neural network Methods 0.000 claims abstract description 33
- 239000011159 matrix material Substances 0.000 claims description 49
- 230000006870 function Effects 0.000 claims description 35
- 230000009467 reduction Effects 0.000 claims description 22
- 238000012795 verification Methods 0.000 claims description 16
- 238000012545 processing Methods 0.000 claims description 15
- 238000011176 pooling Methods 0.000 claims description 14
- 230000004913 activation Effects 0.000 claims description 13
- 238000004590 computer program Methods 0.000 claims description 11
- 230000008569 process Effects 0.000 claims description 9
- 238000012300 Sequence Analysis Methods 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 6
- 238000013215 result calculation Methods 0.000 claims description 5
- 238000003860 storage Methods 0.000 claims description 2
- 230000000694 effects Effects 0.000 abstract description 10
- 238000001514 detection method Methods 0.000 abstract description 6
- 238000010586 diagram Methods 0.000 description 5
- 102100025142 Beta-microseminoprotein Human genes 0.000 description 4
- 101000576812 Homo sapiens Beta-microseminoprotein Proteins 0.000 description 4
- 238000013527 convolutional neural network Methods 0.000 description 4
- 239000000284 extract Substances 0.000 description 3
- 108010076504 Protein Sorting Signals Proteins 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 238000005065 mining Methods 0.000 description 2
- 238000003909 pattern recognition Methods 0.000 description 2
- 230000000737 periodic effect Effects 0.000 description 2
- 241000287127 Passeridae Species 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000013210 evaluation model Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000012634 fragment Substances 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 238000010845 search algorithm Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
- 238000009966 trimming Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Testing Relating To Insulation (AREA)
Abstract
The invention discloses a GIS equipment insulation defect evaluation method and system, which are characterized in that partial discharge information of GIS equipment is extracted to obtain a plurality of partial discharge feature vectors, historical fault case text of the GIS equipment is extracted to obtain historical fault case text feature vectors, the historical fault case text feature vectors and the plurality of partial discharge feature vectors are summed to obtain a plurality of vector sequences, the plurality of vector sequences are input into an LSTM model to be calculated to obtain state feature vectors, the dimension of the state feature vectors is reduced to the number of states by using a full connection layer, the state feature vectors are converted into probability distribution, the probability distribution comprises a plurality of states, each state corresponds to the respective probability, the state with the first probability rank is output from the probability distribution to be the defect evaluation result of the equipment, and the method can obtain the evaluation effect of better effect by evaluating the defect of the GIS equipment by taking the current collected partial discharge detection data and the historical case data into consideration and using a neural network.
Description
Technical Field
The invention relates to the technical field of GIS equipment insulation defect severity assessment, in particular to a GIS equipment insulation defect assessment method and system.
Background
The gas insulated switchgear (gas insulated switchgear, GIS) is widely applied to the power grid due to the characteristics of small occupied area, high reliability, excellent insulating property and the like. In the engineering of manufacturing, transporting and assembling, the GIS equipment is easy to introduce potential hidden dangers such as particles, fragments and the like and cause insulation defects such as partial discharge (partial discharge, PD) and the like in the equipment, so that the equipment can be subjected to serious faults.
The traditional GIS equipment insulation state evaluation is mainly based on relevant standards and expert experience, and static evaluation is carried out through periodic inspection and power failure experiments. In order to reduce economic loss and human resource waste caused by periodic power outage inspection, students propose to utilize partial discharge statistical feature quantity to combine with expert experience to perform state evaluation on GIS equipment. With the rapid development of information technology and big data technology, intelligent algorithms such as machine learning and deep learning are widely applied to equipment insulation state evaluation. The high-voltage switching equipment reliability evaluation model based on the radial basis function has good learning and classifying capabilities, and is high in training speed and low in error rate; the fractal box dimension-based method can be used for identifying the contact state of the GIS isolating switch for the characteristic quantity; the optimization support vector machine can effectively identify the partial discharge type and the like in GIS equipment based on an improved sparrow group search algorithm. However, besides the fact that the partial discharge characteristic in the GIS can reflect the insulation defect state of the equipment, the current insulation state evaluation result is affected by the historical running condition of the equipment and the condition that the similar equipment has insulation faults. For example, when the partial discharge characteristics are similar to those of the similar equipment having an insulation failure, then sufficient attention should be paid even if the partial discharge in the interior thereof is not serious at present. The current intelligent diagnosis algorithm can only evaluate the insulation defect state of the equipment by analyzing the current acquired partial discharge detection data, and the evaluated intelligent level and accuracy are low.
Disclosure of Invention
The invention provides a GIS equipment insulation defect evaluation method and a GIS equipment insulation defect evaluation system.
In order to solve the technical problems, an embodiment of the present invention provides a method for evaluating insulation defects of a GIS device, including:
the local discharge model is adopted to identify a neural network, a plurality of local discharge characteristic vectors are obtained by extracting the local discharge information of the GIS equipment in a plurality of days, and a historical fault case text characteristic vector of the GIS equipment is obtained by extracting the historical fault case text;
summing the historical fault case text feature vector and a plurality of partial discharge feature vectors to obtain a plurality of vector sequences, and inputting the plurality of vector sequences into a preset LSTM model to calculate to obtain a state feature vector;
and after the dimension of the state feature vector is reduced to the number of states by using the full connection layer, converting the state feature vector into probability distribution by using a Softmax function, and outputting a state with a first probability rank from the probability distribution as a defect evaluation result of the GIS equipment, wherein the probability distribution comprises a plurality of states, and each state corresponds to the probability of the corresponding state.
According to the method, the local discharge model is adopted to identify the neural network, the partial discharge information of the GIS equipment on multiple days is extracted to obtain multiple partial discharge feature vectors, the historical fault case text of the GIS equipment is extracted to obtain the historical fault case text feature vectors, the historical fault case text feature vectors and the multiple partial discharge feature vectors are summed to obtain multiple vector sequences, the multiple vector sequences are input into a preset LSTM model to be calculated to obtain state feature vectors, the full-connection layer is utilized to reduce the dimensionality of the state feature vectors to the number of states, then the state feature vectors are converted into probability distribution by utilizing a Softmax function, the probability distribution comprises multiple states, each state corresponds to each probability, the first state of probability rank is output from the probability distribution to be a defect evaluation result of the GIS equipment, and the defect of the GIS equipment is evaluated by utilizing the neural network by taking the current collected partial discharge detection data and the historical case data into consideration, so that an evaluation effect of better effect can be obtained.
Preferably, the method further comprises:
and after verifying the defect evaluation result by adopting a K-fold cross verification mode, evaluating the defect evaluation result by adopting a confusion matrix.
As a preferred scheme, the partial discharge model identification neural network is adopted to extract the partial discharge information of the GIS equipment in multiple days to obtain a plurality of partial discharge feature vectors, and the method specifically comprises the following steps:
after the partial discharge information is subjected to zero filling, the data subjected to zero filling is input into a convolution layer to be subjected to convolution operation, and then a partial discharge feature matrix is output, wherein the convolution process is as follows:
wherein, sigma (·) is expressed as a nonlinear activation function,
indicating the ith in the output partial discharge characteristic matrix after convolutionValues on row, j;
pooling operation is carried out on the partial discharge feature matrix to obtain a model feature vector, the model feature vector is input into a full-connection layer for classification, the dimensionality of the feature vector is reduced to the preset number of partial discharge modes, and the feature vector after the dimensionality reduction is output, wherein the full-connection layer expresses:
f p ＝σ(W p ·f d +b p )
wherein ,
weights and biases for fully connected layers, +.>
The feature vector is the classification mode feature vector after dimension reduction;
and converting the feature vectors after the dimension reduction into pre-probability distribution by using a Softmax function, and outputting a plurality of partial discharge feature vectors.
As a preferred scheme, extracting the historical fault case text of the GIS equipment to obtain a feature vector of the historical fault case text, specifically:
modeling each character in the historical fault case text by adopting a BERT embedding model, and converting the historical fault case text into an embedding vector matrix;
and (3) performing sequence analysis on word embedded vectors in the embedded vector matrix of the text by using the LSTM neural network, and extracting historical case characteristics of the text.
According to the embodiment, modeling is carried out on each character in the historical fault case text by adopting the BERT embedding model, the historical fault case text is converted into an embedding vector matrix, and after the LSTM neural network is used for carrying out sequence analysis on word embedding vectors in the embedding vector matrix of the text, the historical case characteristics of the text are extracted. The method is used for converting the text of the historical fault case of the equipment into a word embedding matrix, and utilizing a long and short-time memory neural network to conduct information mining on the matrix to obtain the representing feature vector of the historical case, and sequence information can be further mined out by using the method, so that the accuracy of insulation defect severity assessment is improved.
As a preferred scheme, after performing sequence analysis on word embedded vectors in an embedded vector matrix of a text by using an LSTM neural network, extracting historical case features of the text, specifically:
and processing the word embedded vector by utilizing the LSTM neural network to output historical case characteristics, wherein the processing process is as follows:
wherein ,Wf 、W i And W is equal to o Representing weights, b f 、b i And b o Representing the bias component, [. Cndot.,)]Representing the concatenation operation of vectors, f tanh (. Cndot.) is the tanh activation function, W c 、b c For weight and bias, symbols
For the term-wise multiplication of vectors, f t Indicating forgetful door c t-1 Indicating the state of the cell at the previous time, i t Representing an input door, c t Representing the cell state, f hc ＝h l Representing historical case features.
In order to solve the same technical problems, the invention also provides a GIS equipment insulation defect evaluation system which comprises a data processing module, a state characteristic vector calculation module and a defect evaluation result calculation module,
the data processing module is used for extracting the partial discharge information of the GIS equipment in multiple days by adopting the partial discharge model identification neural network to obtain multiple partial discharge feature vectors, and extracting the historical fault case text of the GIS equipment to obtain the historical fault case text feature vectors;
the state feature vector calculation module is used for summing the historical fault case text feature vector and the partial discharge feature vectors to obtain a plurality of vector sequences, and inputting the plurality of vector sequences into a preset LSTM model to calculate to obtain a state feature vector;
the defect evaluation result calculation module is used for reducing the dimensionality of the state feature vector into the number of states by using the full connection layer, converting the state feature vector into probability distribution by using a Softmax function, and outputting a state with a first probability rank from the probability distribution as a defect evaluation result of the GIS equipment, wherein the probability distribution comprises a plurality of states, and each state corresponds to the respective probability.
Preferably, the system also comprises a verification module,
the verification module is used for verifying the defect evaluation result by adopting a K-fold cross verification mode and then evaluating the defect evaluation result by adopting a confusion matrix.
Preferably, the data processing module comprises a convolution unit, a pooling unit and a dimension reduction unit,
the convolution unit is used for inputting the data subjected to zero filling into the convolution layer after the partial discharge information is subjected to zero filling, outputting a partial discharge feature matrix after the convolution operation is performed, and the convolution process is as follows:
wherein, sigma (·) is expressed as a nonlinear activation function,
the numerical value on the ith row and the jth column in the output partial discharge characteristic matrix after convolution is represented;
the pooling unit is used for pooling the partial discharge feature matrix to obtain model feature vectors, inputting the model feature vectors into the full-connection layer for classification, and reducing the dimensionality of the feature vectors to the preset number of partial discharge modes to output the feature vectors after the dimensionality reduction, wherein the full-connection layer expresses:
f p ＝σ(W p ·f d +b p )
wherein ,
weights and biases for fully connected layers, +.>
The feature vector is the classification mode feature vector after dimension reduction;
the dimension reduction unit is used for converting the feature vectors after dimension reduction into pre-probability distribution by using a Softmax function and outputting a plurality of partial discharge feature vectors.
The invention also provides an electronic device, which comprises a memory and a processor, wherein the memory is used for storing a computer program, and the processor is used for realizing the GIS device insulation defect evaluation method shown in the embodiment when executing the computer program.
The invention also provides a computer program which is executed by a processor to realize the steps of the GIS equipment insulation defect evaluation method according to the embodiment.
The invention has the following beneficial effects:
the method comprises the steps of identifying a neural network by adopting a partial discharge model to extract partial discharge information of GIS equipment on multiple days to obtain multiple partial discharge feature vectors, extracting historical fault case text of the GIS equipment to obtain historical fault case text feature vectors, summing the historical fault case text feature vectors with the multiple partial discharge feature vectors to obtain multiple vector sequences, inputting the multiple vector sequences into a preset LSTM model to calculate to obtain a state feature vector, reducing the dimensionality of the state feature vector into the number of states by utilizing a full-connection layer, converting the state feature vector into probability distribution by utilizing a Softmax function, wherein the probability distribution comprises multiple states, each state corresponds to the respective probability, outputting a state with the first probability rank from the probability distribution as a defect evaluation result of the GIS equipment, and evaluating the defect of the GIS equipment by utilizing the neural network by considering the partial discharge detection data and the historical case data which are acquired currently, so that a better effect evaluation effect can be obtained.
Drawings
Fig. 1: a flow diagram of one embodiment of the GIS equipment insulation defect evaluation method provided by the invention;
fig. 2: the method flow diagram of one embodiment of the GIS equipment insulation defect evaluation method is provided by the invention;
fig. 3: the confusion matrix schematic diagram of one embodiment of the GIS equipment insulation defect evaluation method provided by the invention;
fig. 4: the invention provides a system structure schematic diagram of another embodiment of a GIS equipment insulation defect evaluation method.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Example 1
Referring to fig. 1, in order to provide a method for evaluating insulation defects of a GIS device according to an embodiment of the present invention, the method for evaluating insulation defects of a GIS device includes steps 101 to 103, where each step is specifically as follows:
step 101: and adopting the partial discharge model to identify a neural network, extracting the partial discharge information of the GIS equipment for a plurality of days to obtain a plurality of partial discharge feature vectors, and extracting the historical fault case text of the GIS equipment to obtain the historical fault case text feature vectors.
Optionally, the partial discharge model identification neural network is used for extracting the partial discharge information of the GIS equipment for a plurality of days to obtain a plurality of partial discharge feature vectors, which specifically comprises:
after the partial discharge information is subjected to zero filling, the data subjected to zero filling is input into a convolution layer to be subjected to convolution operation, and then a partial discharge feature matrix is output, wherein the convolution process is as follows:
wherein, sigma (·) is expressed as a nonlinear activation function,
the numerical value on the ith row and the jth column in the output partial discharge characteristic matrix after convolution is represented;
pooling operation is carried out on the partial discharge feature matrix to obtain a model feature vector, the model feature vector is input into a full-connection layer for classification, the dimensionality of the feature vector is reduced to the preset number of partial discharge modes, and the feature vector after the dimensionality reduction is output, wherein the full-connection layer expresses:
f p ＝σ(W p ·f d +b p )
wherein ,
weights and biases for fully connected layers, +.>
The feature vector is the classification mode feature vector after dimension reduction;
and converting the feature vectors after the dimension reduction into pre-probability distribution by using a Softmax function, and outputting a plurality of partial discharge feature vectors.
Optionally, extracting the historical fault case text of the GIS equipment to obtain a feature vector of the historical fault case text, which specifically includes:
modeling each character in the historical fault case text by adopting a BERT embedding model, and converting the historical fault case text into an embedding vector matrix;
and (3) performing sequence analysis on word embedded vectors in the embedded vector matrix of the text by using the LSTM neural network, and extracting historical case characteristics of the text.
Optionally, after performing sequence analysis on word embedded vectors in an embedded vector matrix of the text by using an LSTM neural network, extracting historical case features of the text, specifically:
and processing the word embedded vector by utilizing the LSTM neural network to output historical case characteristics, wherein the processing process is as follows:
wherein ,Wf 、W i And W is equal to o Representing weights, b f 、b i And b o Representing the bias component, [. Cndot.,)]Representing the concatenation operation of vectors, f tanh (. Cndot.) is the tanh activation function, W c 、b c For weight and bias, symbols
For the term-wise multiplication of vectors, f t Indicating forgetful door c t-1 Indicating the state of the cell at the previous time, i t Representing an input door, c t Representing the cell state, f hc ＝h l Representing historical case characteristics, i t Represents an input gate, o t Representing an output gate.
In this embodiment, as shown in fig. 2, firstly, the partial discharge mode features of the device are extracted, and since the different partial discharge modes have different influences on the insulation state of the device, the partial discharge feature information of the device has a larger influence on the accuracy of state evaluation, and the PRPS data of the power device is essentially a two-dimensional matrix, which has a certain similarity with the picture format. Therefore, the invention uses the neural network (Convolutional Neural Network, CNN) to identify the partial discharge mode inside the GIS equipment and extracts the partial discharge characteristic vector of the equipment for nearly 5 days. The partial discharge pattern recognition CNN adopted by the invention consists of 2 convolution layers, 2 pooling layers, 1 full-connection layer and 1 normalized exponential function layer (Softmax), and comprises the following specific steps:
1) Zero padding is carried out on PRPS data so as to ensure that the formats of the output data of the subsequent convolution layers are consistent;
2) The PRPS data after zero padding is input into a convolution layer for convolution operation, wherein the convolution layer is the core of a CNN neural network and is provided with a plurality of convolution kernels, the local features of an outgoing signal can be extracted, and one convolution kernel of the convolution layer is omega s×s The input normalized partial discharge PRPS data is
The convolution operation of the convolution layer may be represented by the following equation:
wherein, sigma (·) is expressed as a nonlinear activation function,
the numerical value on the ith row and the jth column in the output partial discharge characteristic matrix after convolution is represented;
3) And (3) pooling the partial discharge feature matrix, and reducing the size of the data while maintaining the features of the original data and reducing the overfitting condition of the model. The mode feature vector of the partial discharge data is finally obtained through two convolutions and pooling
4) The mode feature vectors are input into a full connection layer for classification, and the dimension of the feature vectors is reduced to the number d of partial discharge modes 2 The full connection layer may be represented by the following formula:
f p ＝σ(W p ·f d +b p )
wherein ,
weights and biases for fully connected layers, +.>
The feature vector is the classification mode feature vector after dimension reduction;
5) Output f of fully connected layer using Softmax function p And converting into a probability distribution so as to realize pattern recognition of the partial discharge data.
And secondly, extracting text characteristics of the historical fault cases. Converting the historical fault case text of the equipment into a word embedding matrix, and utilizing a long and short time memory neural network to carry out information mining on the matrix to obtain a representation feature vector of a historical case, wherein the method comprises the following specific steps of:
modeling each character in the text by adopting a BERT embedding model, and representing each character by using a word embedding vector so as to convert the historical fault case text into an embedding vector matrix, wherein the length and the direction of the word embedding vector in an embedding space can reflect the semantics of a certain character to a certain extent, the characters with similar semantics have similar two corresponding embedding vectorsLength and direction, on the other hand, characters with opposite semantics have larger distances between their embedded vectors. As an example of this embodiment, the historical failure case text of GIS is T hc ＝<w 1 ,w 2 ,…,w l>, wherein wi For the ith character in the text, l is the character length of the text, the BERT representation model can be expressed as:
Φ＝f bert (T hc )
wherein ,fbert (. Cndot.) represents the BERT embedding model,
embedded representation matrix representing GIS history cases, column vector in phi +.>
A word embedding vector corresponding to the ith character in the representing text, and the dimension of the word embedding vector is d e 。
The LSTM neural network is used for extracting the historical case characteristics, the sequence of characters in the historical fault case text also affects the semantic information of the whole text, and different arrangement sequences of the same characters often express different meanings, so that the sequence analysis is required to be carried out on word embedded vectors in an embedded representation matrix of the text, the historical case characteristic vectors of the text are extracted, and the word embedded vectors input at the moment t are made to be
Output is h t The cell state is c t Forgetting door f of LSTM at time t t Input gate i t Output door o t Can be represented by the following formula: />
wherein ,Wf 、W i And W is equal to o Representing weights, b f 、b i And b o Representing the bias component, [. Cndot.,)]Representing the concatenation operation of vectors, f tanh (. Cndot.) is the tanh activation function, W c 、b c For weight and bias, symbols
For the term-wise multiplication of vectors, f t Indicating forgetful door c t-1 Indicating the state of the cell at the previous time, i t Representing an input door, c t Representing the cell state, f hc ＝h l Representing historical case characteristics, i t Represents an input gate, o t Representing an output gate.
Therefore, the forgetting gate screens the cell state at the last moment, selects which information needs to be forgotten, the input gate updates the screened information to the current memory state, namely the cell state, according to the current input information, and finally, the feature vector of the whole historical fault case text is the last corresponding output f in the word embedded vector sequence hc ＝h l The vector contains the information of the whole text, and the GIS state is evaluated together with the partial discharge feature vector.
Step 102: and summing the historical fault case text feature vector and the partial discharge feature vectors to obtain a plurality of vector sequences, and inputting the plurality of vector sequences into a preset LSTM model to calculate to obtain a state feature vector.
In this embodiment, the case-representative feature vector is summed with the resulting 5 partial-discharge feature vectors. The development trend of partial discharge of GIS equipment also has an effect on the insulation defect state of the equipment to a certain extent, and when the partial discharge is more serious, the insulation state of the equipment is also more critical. Therefore, the state of the device needs to be evaluated according to the partial discharge signal sequence within 5 days, and is integrated into the device history case information.
The added vector sequence is input into LSTM. After the following implementation steps, the state feature vector of the equipment is finally obtained, and the implementation steps are as follows:
wherein ,Wf 、W i And W is equal to o Representing weights, b f 、b i And b o Representing the bias component, [. Cndot.,)]Representing the concatenation operation of vectors, f tanh (. Cndot.) is the tanh activation function, W c 、b c For weight and bias, symbols
For the term-wise multiplication of vectors, f t Indicating forgetful door c t-1 Indicating the state of the cell at the previous time, i t Representing an input door, c t Representing the cell state, f hc ＝h l Representing historical case characteristics, i t Represents an input gate, o t Representing an output gate.
As an example of this embodiment, the state of the device is estimated according to the partial discharge signal sequence within 5 days and is incorporated into the device history case information, so that the case feature vectors are summed with 5 partial discharge feature vectors, respectively, to finally obtain 5 state feature vectors.
Step 103: and after the dimension of the state feature vector is reduced to the number of states by using the full connection layer, converting the state feature vector into probability distribution by using a Softmax function, and outputting a state with a first probability rank from the probability distribution as a defect evaluation result of the GIS equipment, wherein the probability distribution comprises a plurality of states, and each state corresponds to the probability of the corresponding state.
In this embodiment, the dimension of the feature vector is reduced to the number of states by using the full connection layer, and the feature vector is converted into a probability distribution by using the Softmax function, where the probability distribution includes four elements, each element represents the probability that the GIS device belongs to 4 states, the sum of the 4 elements is 1, and the final severity of the GIS is determined according to which state in the probability distribution has the highest probability, and the highest one is the severity of the insulation defect of the device.
Optionally, the method further comprises the step of evaluating the defect evaluation result by adopting a K-fold cross validation mode and then adopting a confusion matrix.
In this embodiment, the proposed method for evaluating the severity of insulation defect of GIS device is verified by adopting a K-fold cross verification method, and s=3, d is set by a parameter scanning method 1 ＝256，d 2 ＝5，d e =768. On the other hand, the LSTM layer number for the history case feature extraction is 1, the hiding size is 256, the LSTM layer number for the state feature extraction is 1, the hiding size is 128, the activation function sigma (-) is an attenuation trimming linear unit, the attenuation rate is-0.2, the loss function of the model is a cross entropy function, and for further verification, the method is differentThe preference degree of the insulation state data is evaluated by adopting an confusion matrix, and the confusion matrix of the verification results on 1200 verification set data is used, as shown in fig. 3, and from the experimental results of the confusion matrix, the method can accurately and effectively evaluate different equipment insulation defect severity degrees.
The method can accurately identify samples with different defect severity on the verification set. The accuracy of the model on 4 samples in the insulating states is 91.50%, 86.45%, 86.64% and 88.01%, and the recall rates are 93.33%, 89.33%, 84.33% and 85.67%, respectively, and the method can effectively extract continuous characteristics in the samples, and treat the severity level of the insulating defect of the GIS device as continuous rather than discrete, accords with common evaluation common sense of people, improves the F measure of the traditional model by 19.39%, and can equally treat samples in various insulating defect states under the addition of a historical fault case text.
The method has the advantages that the partial discharge information of the GIS equipment on multiple days is extracted by adopting the partial discharge model to identify the neural network to obtain multiple partial discharge feature vectors, the historical fault case text of the GIS equipment is extracted to obtain the historical fault case text feature vectors, the historical fault case text feature vectors and the multiple partial discharge feature vectors are summed to obtain multiple vector sequences, the multiple vector sequences are input into a preset LSTM model to calculate to obtain state feature vectors, the dimension of the state feature vectors is reduced to the number of states by utilizing the full connection layer, the state feature vectors are converted into probability distribution by utilizing the Softmax function, the probability distribution comprises multiple states, each state corresponds to the respective probability, the state with the first probability is output from the probability distribution to serve as a defect evaluation result of the GIS equipment, and the method can obtain a better evaluation effect by evaluating the defect of the GIS equipment by taking the acquired partial discharge detection data and the historical case data into consideration and utilizing the neural network.
Example two
Accordingly, referring to fig. 4, fig. 4 is a schematic structural diagram of a GIS device insulation defect evaluation system provided by the present invention, as shown in the drawing, the GIS device insulation defect evaluation system includes a data processing module 401, a state feature vector calculating module 402 and a defect evaluation result calculating module 403,
the data processing module 401 is configured to extract partial discharge information of a GIS device on multiple days by using a partial discharge model to identify a neural network to obtain multiple partial discharge feature vectors, and extract a historical fault case text of the GIS device to obtain a historical fault case text feature vector;
the state feature vector calculation module 402 is configured to sum the historical fault case text feature vector and the partial discharge feature vectors to obtain a plurality of vector sequences, and input the plurality of vector sequences into a preset LSTM model to perform calculation to obtain a state feature vector;
the defect evaluation result calculation module 403 is configured to reduce the dimension of the state feature vector to the number of states by using the full connection layer, convert the state feature vector to probability distribution by using a Softmax function, and output a state with a first probability rank from the probability distribution as a defect evaluation result of the GIS device, where the probability distribution includes a plurality of states, and each state corresponds to a respective probability.
Optionally, a verification module 404 is also included,
the verification module 404 is configured to verify the defect evaluation result by using a K-fold cross verification method, and then evaluate the defect evaluation result by using a confusion matrix.
Optionally, the data processing module 401 further includes a convolution unit 4011, a pooling unit 4012 and a dimension reduction unit 4013,
the convolution unit 4011 is configured to perform zero padding on the partial discharge information, input the data after zero padding into a convolution layer, perform convolution operation, and output a partial discharge feature matrix, where the convolution process is:
wherein, sigma (·) is expressed as a nonlinear activation function,
the numerical value on the ith row and the jth column in the output partial discharge characteristic matrix after convolution is represented;
the pooling unit 4012 is configured to pool the partial discharge feature matrix to obtain a model feature vector, input the model feature vector into a full-connection layer for classification, and reduce the dimension of the feature vector to a preset number of partial discharge modes to output the feature vector after dimension reduction, where the full-connection layer expresses:
f p ＝σ(W p ·f d +b p )
wherein ,
weights and biases for fully connected layers, +.>
The feature vector is the classification mode feature vector after dimension reduction;
the dimension reduction unit 4013 is configured to convert the feature vectors after dimension reduction into a pre-probability distribution by using a Softmax function, and then output a plurality of partial discharge feature vectors.
Optionally, the invention further provides an electronic device, a memory and a processor, which are used for storing a computer program and implementing the GIS device insulation defect evaluation method according to the first embodiment when the computer program is executed.
Optionally, the present invention further provides a storage medium, where a computer program is stored, where the computer program implements the steps of the GIS device insulation defect evaluation method according to the first embodiment when the computer program is executed by a processor.
The more detailed working principle and the step flow of this embodiment can be, but not limited to, those described in the related embodiment one.
Compared with the prior art, the method has the advantages that partial discharge information of GIS equipment on multiple days is extracted through the partial discharge model identification neural network to obtain multiple partial discharge feature vectors, historical fault case text of the GIS equipment is extracted to obtain historical fault case text feature vectors, the historical fault case text feature vectors and the multiple partial discharge feature vectors are summed to obtain multiple vector sequences, the multiple vector sequences are input into a preset LSTM model to be calculated to obtain state feature vectors, the dimension of the state feature vectors is reduced to the number of states through the full connection layer, the state feature vectors are converted into probability distribution through the Softmax function, the probability distribution comprises multiple states, each state corresponds to each probability, the first state of probability rank is output from the probability distribution to serve as a defect evaluation result of the GIS equipment, and the defect of the GIS equipment is evaluated through the neural network by taking the partial discharge detection data and the historical case data collected at present into consideration, so that a better effect evaluation effect can be obtained.
The foregoing embodiments have been provided for the purpose of illustrating the general principles of the present invention, and are not to be construed as limiting the scope of the invention. It should be noted that any modifications, equivalent substitutions, improvements, etc. made by those skilled in the art without departing from the spirit and principles of the present invention are intended to be included in the scope of the present invention.
Claims (10)
1. The GIS equipment insulation defect evaluation method is characterized by comprising the following steps of:
the local discharge model is adopted to identify a neural network, a plurality of local discharge characteristic vectors are obtained by extracting the local discharge information of the GIS equipment in a plurality of days, and a historical fault case text characteristic vector of the GIS equipment is obtained by extracting the historical fault case text;
summing the historical fault case text feature vector and the partial discharge feature vectors to obtain a plurality of vector sequences, and inputting the plurality of vector sequences into a preset LSTM model to calculate to obtain a state feature vector;
and after the dimensionality of the state feature vector is reduced to the number of states by using a full connection layer, converting the state feature vector into probability distribution by using a Softmax function, and outputting a state with a first probability ranking from the probability distribution as a defect evaluation result of the GIS equipment, wherein the probability distribution comprises a plurality of states, and each state corresponds to the respective probability.
2. The GIS device insulation defect evaluation method of claim 1, further comprising:
and after verifying the defect evaluation result by adopting a K-fold cross verification mode, evaluating the defect evaluation result by adopting a confusion matrix.
3. The method for evaluating insulation defects of a GIS device according to claim 1, wherein the local discharge model identification neural network is used for extracting the partial discharge information of the GIS device for a plurality of days to obtain a plurality of partial discharge feature vectors, and specifically comprises:
after the partial discharge information is subjected to zero filling, the data subjected to zero filling is input into a convolution layer to be subjected to convolution operation, and then a partial discharge feature matrix is output, wherein the convolution process is as follows:
wherein σ (·) is expressed as a nonlinear activation function, P d (i, j) represents the numerical value on the ith row and the jth column in the output partial discharge feature matrix after convolution;
performing pooling operation on the partial discharge feature matrix to obtain a model feature vector, inputting the model feature vector into a full-connection layer for classification, and reducing the dimensionality of the feature vector to the preset number of partial discharge modes to output the feature vector after the dimensionality reduction, wherein the full-connection layer expresses:
f p ＝σ(W p ·f d +b p )
wherein ,
weights and biases for fully connected layers, +.>
For classification after dimension reductionA pattern feature vector;
and converting the feature vectors after the dimension reduction into pre-probability distribution by using a Softmax function, and outputting a plurality of partial discharge feature vectors.
4. The method for evaluating insulation defects of a GIS device according to claim 1, wherein the extracting the historical fault case text of the GIS device to obtain the feature vector of the historical fault case text specifically comprises:
modeling each character in the historical fault case text by adopting a BERT embedding model, and converting the historical fault case text into an embedding vector matrix;
and (3) performing sequence analysis on word embedded vectors in the embedded vector matrix of the text by using the LSTM neural network, and extracting historical case characteristics of the text.
5. The method for evaluating insulation defects of GIS equipment according to claim 4, wherein the sequence analysis of word embedded vectors in the embedded vector matrix of the text by using the LSTM neural network is performed, and then historical case characteristics of the text are extracted, specifically:
and processing the word embedded vector by utilizing the LSTM neural network to output historical case characteristics, wherein the processing process is as follows:
wherein ,Wf 、W i And W is equal to o Representing weights, b f 、b i And b o Representing the bias component, [. Cndot.,)]Representing the concatenation operation of vectors, f tanh (. Cndot.) is the tanh activation function, W c 、b c For weight and bias, symbols
For the term-wise multiplication of vectors, f t Indicating forgetful door c t-1 Indicating the state of the cell at the previous time, i t Representing an input door, c t Representing the cell state, f hc ＝h l Representing historical case characteristics, i t Represents an input gate, o t Representing an output gate.
6. A GIS equipment insulation defect evaluation system is characterized by comprising a data processing module, a state characteristic vector calculation module and a defect evaluation result calculation module,
the data processing module is used for extracting the partial discharge information of the GIS equipment on multiple days by adopting the partial discharge model recognition neural network to obtain multiple partial discharge feature vectors, and extracting the historical fault case text of the GIS equipment to obtain the historical fault case text feature vectors;
the state feature vector calculation module is used for summing the historical fault case text feature vector and the partial discharge feature vectors to obtain a plurality of vector sequences, and inputting the plurality of vector sequences into a preset LSTM model to calculate to obtain a state feature vector;
the defect evaluation result calculation module is used for reducing the dimensionality of the state feature vector into the number of states by using a full connection layer, converting the number of states into probability distribution by using a Softmax function, and outputting a defect evaluation result of which the first state of probability ranking is GIS equipment from the probability distribution, wherein the probability distribution comprises a plurality of states, and each state corresponds to the respective probability.
7. The GIS equipment insulation defect evaluation system of claim 6, further comprising a verification module,
the verification module is used for verifying the defect evaluation result by adopting a K-fold cross verification mode and then evaluating the defect evaluation result by adopting a confusion matrix.
8. The GIS equipment insulation defect evaluation system according to claim 6, wherein the data processing module comprises a convolution unit, a pooling unit and a dimension reduction unit,
the convolution unit is used for inputting the data subjected to zero filling into the convolution layer after the partial discharge information is subjected to zero filling, outputting a partial discharge feature matrix, and the convolution process is as follows:
wherein, sigma (·) is expressed as a nonlinear activation function,
the numerical value on the ith row and the jth column in the output partial discharge characteristic matrix after convolution is represented;
the pooling unit is used for pooling the partial discharge feature matrix to obtain model feature vectors, inputting the model feature vectors into a full-connection layer for classification, and reducing the dimensionality of the feature vectors to the preset number of partial discharge modes to output the feature vectors with reduced dimensionality, wherein the full-connection layer expresses:
f p ＝σ(W p ·f d +b p )
wherein ,
weights and biases for fully connected layers, +.>
The feature vector is the classification mode feature vector after dimension reduction;
the dimension reduction unit is used for converting the dimension reduced feature vector into a pre-probability distribution by using a Softmax function and outputting a plurality of partial discharge feature vectors.
9. An electronic device, comprising:
a memory for storing a computer program;
a processor for implementing the GIS device insulation defect evaluation method according to any one of claims 1 to 5 when executing the computer program.
10. A storage medium having stored thereon a computer program which, when executed by a processor, implements the steps of the GIS device insulation defect evaluation method according to any one of claims 1 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211708828.3A CN116011412A (en) | 2022-12-29 | 2022-12-29 | GIS equipment insulation defect evaluation method, system, equipment and medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211708828.3A CN116011412A (en) | 2022-12-29 | 2022-12-29 | GIS equipment insulation defect evaluation method, system, equipment and medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116011412A true CN116011412A (en) | 2023-04-25 |
Family
ID=86024338
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211708828.3A Pending CN116011412A (en) | 2022-12-29 | 2022-12-29 | GIS equipment insulation defect evaluation method, system, equipment and medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116011412A (en) |
-
2022
- 2022-12-29 CN CN202211708828.3A patent/CN116011412A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20220358363A1 (en) | Engine surge fault prediction system and method based on fusion neural network model | |
| Van et al. | Bearing‐fault diagnosis using non‐local means algorithm and empirical mode decomposition‐based feature extraction and two‐stage feature selection | |
| US8935258B2 (en) | Identification of sample data items for re-judging | |
| CN112418292B (en) | Image quality evaluation method, device, computer equipment and storage medium | |
| CN113343581B (en) | Transformer fault diagnosis method based on graph Markov neural network | |
| CN113220886A (en) | Text classification method, text classification model training method and related equipment | |
| CN112199536A (en) | Cross-modality-based rapid multi-label image classification method and system | |
| CN110019822B (en) | Few-sample relation classification method and system | |
| CN114048468A (en) | Intrusion detection method, intrusion detection model training method, device and medium | |
| CN115392357A (en) | Classification model training and labeled data sample spot inspection method, medium and electronic equipment | |
| Chang et al. | Improved VMD‐KFCM algorithm for the fault diagnosis of rolling bearing vibration signals | |
| CN111080168A (en) | Power communication network equipment reliability evaluation method based on capsule network | |
| CN112559741B (en) | Nuclear power equipment defect record text classification method, system, medium and electronic equipment | |
| CN115795038B (en) | Intent recognition method and device based on localization deep learning framework | |
| CN116011412A (en) | GIS equipment insulation defect evaluation method, system, equipment and medium | |
| CN116089605A (en) | Text emotion analysis method based on transfer learning and improved word bag model | |
| CN113052060B (en) | Bearing residual life prediction method and device based on data enhancement and electronic equipment | |
| CN113435321A (en) | Method, system and equipment for evaluating state of main shaft bearing and readable storage medium | |
| CN112818164A (en) | Music type identification method, device, equipment and storage medium | |
| CN110569331A (en) | Context-based relevance prediction method and device and storage equipment | |
| CN116432120A (en) | Transformer fault type classification method based on knowledge distillation | |
| Zhang et al. | Bearing Fault Diagnosis based on Convolution Neural Network with Logistic Chaotic Map | |
| CN116882392A (en) | Power transformer fault identification method and system based on improved BERT-CNN | |
| CN117763491A (en) | Non-invasive load monitoring method, equipment and medium for deep integrated learning | |
| CN116152582A (en) | Image classification method and device, storage medium and electronic equipment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |