CN115906520A - Low-voltage transformer health degree evaluation method based on Mahalanobis distance and self-encoder - Google Patents
Low-voltage transformer health degree evaluation method based on Mahalanobis distance and self-encoder Download PDFInfo
- Publication number
- CN115906520A CN115906520A CN202211663923.6A CN202211663923A CN115906520A CN 115906520 A CN115906520 A CN 115906520A CN 202211663923 A CN202211663923 A CN 202211663923A CN 115906520 A CN115906520 A CN 115906520A
- Authority
- CN
- China
- Prior art keywords
- data
- encoder
- mahalanobis distance
- self
- health
- 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
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
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to the technical field of power equipment monitoring, and discloses a low-voltage transformer health degree evaluation method based on Mahalanobis distance and a self-encoder, wherein electrical parameter data and temperature parameter data of a low-voltage transformer which runs stably for a long time are acquired from a transformer historical database, and the acquired transformer data are preprocessed, including missing value mean value filling and standardization processing, so that the specific value difference of the parameter data is reduced, and each data is ensured to have homology and equivalence; carrying out PCA (principal component analysis) dimension reduction analysis based on the preprocessed data, and solving a data mean value and a covariance matrix by using the dimension reduced data so as to obtain the Mahalanobis distance of the sample data; meanwhile, a self-encoder model is established by utilizing the preprocessed data, and the reconstruction error of the sample data is obtained after the model is trained; and based on the Mahalanobis distance and the reconstruction error, converting the Markov distance and the reconstruction error into a health degree score by using a quantization function, and performing weighting fusion on the health degree score to obtain a final health degree score of the low-voltage transformer.
Description
Technical Field
The invention relates to the technical field of power equipment monitoring, in particular to a low-voltage transformer health degree evaluation method based on Mahalanobis distance and an autoencoder.
Background
The power transformer is one of key devices in the process of transmitting electric energy in a power system, and has very important significance in ensuring safe and stable operation of the transformer. When a transformer is put into use, the transformer may be subjected to an excessive load, and may be damaged due to insulation deterioration of a material or a natural disaster. Once the transformer is damaged, the electric energy cannot be normally transported, and huge loss can be brought to national economy. In order to guarantee the reliability and economy of the operation of the power system, a diagnosis model capable of accurately evaluating the health degree of the transformer in real time is indispensable.
Currently, the evaluation of the running state of the transformer includes three aspects: fault diagnosis, health state evaluation and health degree evaluation. Common methods for fault diagnosis include analysis of dissolved gas in oil, expert systems, machine learning, and deep learning methods. However, these methods have certain disadvantages, for example, the traditional analysis method for dissolved gas in oil has the problem of incomplete coding, and the rule between characteristic gas and fault type cannot be faithfully expressed; the expert system cannot learn autonomously, needs expert experience to intervene, and is low in working efficiency; machine learning and deep learning have strong learning ability, but a large amount of sample data is needed for training, the over-parameter adjustment is complex, the learning period is long, and the local optimum value is easy to fall into. In addition, the transformer fault diagnosis requires a large amount of fault data, but the labeled fault data is difficult to obtain in reality. A transformer health state evaluation method based on the fuzzy theory, the cloud model and the set pair analysis belongs to a qualitative method, the transformer state is divided into a plurality of grades by the methods, and when the health states of a plurality of devices need to be compared, particularly when the devices are in the same health level, the qualitative evaluation result is useless. Therefore, it is necessary to quantify the health status of the transformer device to reflect the real operation status of the transformer device. For cost reasons, low voltage transformers generally only acquire basic electrical and temperature parameters, which also limits the use of the above-described methods. In view of the above, a low-voltage transformer health degree evaluation method based on mahalanobis distance and a self-encoder is provided by using key parameter indexes representing health degree.
Disclosure of Invention
Technical problem to be solved
Aiming at the defects of the prior art, the invention provides a low-voltage transformer health degree evaluation method based on Mahalanobis distance and a self-encoder, aiming at solving the problem that the running state of a low-voltage transformer is evaluated under the condition that the low-voltage transformer lacks a fault sample and lacks core measurement parameters, obtaining a health degree score which quantitatively reflects the running state of the transformer and providing a reliable basis for operation and maintenance personnel to correctly judge the running state of the transformer.
(II) technical scheme
In order to achieve the purpose, the invention provides the following technical scheme: the low-voltage transformer health degree evaluation method based on the Mahalanobis distance and the self-encoder comprises the following steps of:
1): acquiring electrical parameter data and temperature parameter data of the low-voltage transformer in long-term stable operation from a transformer historical database;
2): preprocessing the acquired transformer data to ensure that each data has homology and equivalence;
3): carrying out PCA (principal component analysis) dimensionality reduction on the basis of the preprocessed data, calculating the Mahalanobis distance of the sample data by using the data subjected to dimensionality reduction, and converting the Mahalanobis distance into a Mahalanobis distance health score by using a quantization function;
4): establishing a self-encoder model by utilizing the preprocessed data, obtaining a reconstruction error of sample data after training the model, and converting the reconstruction error into a self-encoder health score by utilizing a quantization function;
5): and based on the Mahalanobis distance health degree and the self-encoder health degree, weighting and fusing the health degree scores to obtain the final health degree score of the low-voltage transformer.
Preferably, the preprocessing of the data in step 2 includes: filling the parameter data mean value for the problem that the parameter characteristic data is missing; in order to make the parameter characteristics comparable, the detected data of each parameter characteristic is normalized.
Preferably, the calculation of converting the quantization function into the mahalanobis distance health degree in step 3 includes performing PCA dimension reduction on the normalized data, specifically, solving a covariance matrix of the normalized data, performing eigenvalue decomposition on the obtained covariance mean to obtain eigenvalues and corresponding eigenvectors, sorting the obtained eigenvalues from large to small, selecting eigenvectors corresponding to the first K eigenvalues with an accumulated contribution degree greater than 90% to form a load matrix based on the principal component contribution degree, performing dimension reduction on the normalized data based on the load matrix, obtaining a mean vector and a covariance matrix of the data after dimension reduction, solving a mahalanobis distance and a mahalanobis distance threshold of sample data based on the obtained mean vector and covariance matrix, and performing quantization on the mahalanobis distance to convert the mahalanobis distance into the health degree.
Preferably, the calculation of converting the quantization function into the self-Encoder health degree in the step 4 includes randomly initializing a parameter matrix and a bias vector of a self-Encoder and a Decoder, inputting the standardized data into the Encoder to obtain a compressed representation of the input data, decoding the compressed representation by the Decoder to output a reconstructed sample of the original input, and minimizing a loss function through a back propagation algorithm and a gradient descent method to obtain an optimal parameter of the self-Encoder. And obtaining the reconstruction error and the reconstruction error threshold of the sample data by using the trained self-encoder, and quantizing and converting the reconstruction error into a health degree.
Preferably, the calculating of the weighted health degree in step 5 includes: the method provides reliable basis for objectively and accurately judging the running state of the transformer for operation and maintenance personnel, and obtains the final health degree of the low-voltage transformer by performing weighted fusion on the health degree of the Mahalanobis distance and the health degree of the self-encoder, wherein the fusion formula is shown as the formula (1).
s final ＝w mh s mh +w ae s ae (1)
Wherein s is final For weighting the degree of health, i.e. the final degree of health of the low-voltage transformer, s mh And s ae The Mahalanobis distance health degree and the self-encoder health degree of the low-voltage transformer, w mh And w ae Weights, w, for Mahalanobis distance health and self-encoder health, respectively mh ,w ae ∈[0,1]And w is mh +w ae ＝1。
(III) advantageous effects
The invention provides a low-voltage transformer health degree evaluation method based on Mahalanobis distance and a self-encoder, which has the following beneficial effects:
the method comprises the steps that electrical parameter data and temperature parameter data of the low-voltage transformer in long-term stable operation are obtained from a transformer historical database, and the obtained transformer data are preprocessed, wherein the preprocessing comprises missing value mean value filling and standardization processing, so that the specific value difference of the parameter data is reduced, and the data are ensured to have the homology and the equivalence; carrying out PCA dimension reduction analysis based on the preprocessed data, and solving a data mean value and a covariance matrix by using the dimension-reduced data so as to obtain the Mahalanobis distance of the sample data; meanwhile, a self-encoder model is established by utilizing the preprocessed data, and the reconstruction error of the sample data is obtained after the model is trained; and based on the Mahalanobis distance and the reconstruction error, converting the Markov distance and the reconstruction error into a health degree score by using a quantization function, and performing weighting fusion on the health degree score to obtain a final health degree score of the low-voltage transformer. The method calculates the health degree of the low-voltage transformer in a model fusion mode, and provides reliable basis for operation and maintenance personnel to correctly judge the running condition of the transformer
Drawings
Fig. 1 is a flowchart of the low-voltage transformer health assessment based on mahalanobis distance and self-encoder provided by the present invention;
FIG. 2 is a health assessment process based on PCA-Mahalanobis distance proposed by the present invention;
fig. 3 is a flow chart of health assessment based on an auto-encoder according to the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without making any creative effort based on the embodiments in the present invention, belong to the protection scope of the present invention.
Referring to fig. 1-3, a low voltage transformer health assessment method based on mahalanobis distance and self-encoder includes the following steps:
1): acquiring electrical parameter data and temperature parameter data of the low-voltage transformer in long-term stable operation from a transformer historical database;
2): preprocessing the acquired transformer data to ensure that each data has homology and equivalence;
3): carrying out PCA (principal component analysis) dimensionality reduction on the basis of the preprocessed data, calculating the Mahalanobis distance of the sample data by using the dimensionality reduced data, and converting the Mahalanobis distance into a Mahalanobis distance health score by using a quantization function;
4): establishing a self-encoder model by utilizing the preprocessed data, obtaining a reconstruction error of sample data after training the model, and converting the reconstruction error into a self-encoder health score by utilizing a quantization function;
5): and based on the Mahalanobis distance health degree and the self-encoder health degree, weighting and fusing the health degree scores to obtain the final health degree score of the low-voltage transformer.
Preferably, the preprocessing of the data in step 2 includes: filling the parameter data mean value for the problem that the parameter characteristic data is missing; in order to make the parameter characteristics comparable, the detected data of each parameter characteristic is normalized.
Preferably, the calculation of converting the quantization function into the mahalanobis distance health degree in step 3 includes performing PCA dimension reduction on the normalized data, specifically, solving a covariance matrix of the normalized data, performing eigenvalue decomposition on the obtained covariance mean to obtain eigenvalues and corresponding eigenvectors, sorting the obtained eigenvalues from large to small, selecting eigenvectors corresponding to the first K eigenvalues with an accumulated contribution degree greater than 90% to form a load matrix based on the principal component contribution degree, performing dimension reduction on the normalized data based on the load matrix, obtaining a mean vector and a covariance matrix of the data after dimension reduction, solving a mahalanobis distance and a mahalanobis distance threshold of sample data based on the obtained mean vector and covariance matrix, and performing quantization on the mahalanobis distance to convert the mahalanobis distance into the health degree.
Preferably, the calculation of converting the quantization function into the self-Encoder health degree in the step 4 includes randomly initializing a parameter matrix and a bias vector of a self-Encoder and a Decoder, inputting the standardized data into the Encoder to obtain a compressed representation of the input data, decoding the compressed representation by the Decoder to output a reconstructed sample of the original input, and minimizing a loss function through a back propagation algorithm and a gradient descent method to obtain an optimal parameter of the self-Encoder. And obtaining the reconstruction error and the reconstruction error threshold of the sample data by using the trained self-encoder, and quantizing and converting the reconstruction error into a health degree.
Preferably, the calculating of the weighted health degree in step 5 includes: the method provides reliable basis for objectively and accurately judging the running state of the transformer for operation and maintenance personnel, and obtains the final health degree of the low-voltage transformer by performing weighted fusion on the health degree of the Mahalanobis distance and the health degree of the self-encoder, wherein the fusion formula is shown as the formula (1).
s final ＝w mh s mh +w ae s ae (1)
Wherein s is final For weighting the degree of health, i.e. the final degree of health of the low-voltage transformer, s mh And s ae The Mahalanobis distance health degree and the self-encoder health degree of the low-voltage transformer, w mh And w ae Weights, w, for Mahalanobis distance health and self-encoder health, respectively mh ,w ae ∈[0,1]And w is mh +w ae ＝1。
1. Data acquisition:
and inquiring the historical data of the electrical parameters and the historical data of the temperature parameters of the transformer in the MySQL database according to the id of the transformer. The electrical parameters comprise A-phase current IA, B-phase current IB, C-phase current IC, A-phase voltage UA, B-phase voltage UB, C-phase voltage UC, ab-phase voltage UAB, bc-phase voltage UBC, ca-phase voltage UCA, A-phase active power PA, B-phase active power PB, C-phase active power UC, A-phase reactive power QA, B-phase reactive power QB and C-phase reactive power QC; the temperature parameters include upper contact A temperature WDSA, upper contact B temperature WDSB, upper contact C temperature WDSC, lower contact A temperature WDXA, lower contact B temperature WDXB, and lower contact C temperature WDXC. The data acquisition interval is 10 seconds, the data which runs stably for four months is obtained, and the original data is recorded as X belongs to R M*N Where M is the historical data sample size and N is the total number of parameters, where N =21.
2. Data preprocessing:
(1) Missing value processing: the inevitable problem of lack in the data acquisition process is due to sensor and network transmission problems. All data are data of long-term stable operation of the transformer, so missing data under each parameter are filled by adopting the mean value of the parameter data. (2) normalization treatment: for the same equipment, the numerical intervals and units of different parameter characteristics are mostly different, and in order to make the characteristics characterized by different parameter characteristics comparable, the detection data of each parameter characteristic needs to be standardized. The normalized formula is formula (1).
Wherein X * Is the parameter characteristic data after X standardization, mu 0 And σ 0 Is the mean vector and standard deviation vector of each parameter feature.
3. Model training
(1) And constructing a health degree evaluation model based on the Mahalanobis distance.
The principle of calculating the health degree of the low-voltage transformer based on the Mahalanobis distance is that the performance of the low-voltage transformer is gradually degraded along with the continuous aging of the low-voltage transformer equipment under the influence of various factors in the operation process, and the running mode distribution inconsistent with the normal and stable running can be shown on each monitoring parameter of the low-voltage transformer. Therefore, the mahalanobis distance distribution of the normal stable operation data is calculated using the long-term stable operation data, and an appropriate distance threshold is selected as the normal-abnormal boundary point based on the mahalanobis distance distribution of the normal data. And for newly acquired data, calculating the Mahalanobis distance between the newly acquired data and normal data through the mean value and the covariance mean value of the normal data, then comparing the newly acquired data with a distance threshold, and if the distance exceeds the threshold, considering that the performance of the test point is degraded, wherein the more serious the degradation is, the greater the degree of the Mahalanobis distance exceeding the threshold is.
The obtained low-voltage transformer parameters have more dimensions, redundancy exists among the parameters, the training of the model is influenced, and the data are interfered by various noises in the collection process, so that the data need to be subjected to dimension reduction processing. Through dimension reduction, on one hand, the dimension of data is reduced, the model training cost is reduced, and on the other hand, the influence of noise is eliminated while the main information of the data is kept.
Principal Component Analysis (PCA) is a widely applied dimensionality reduction algorithm for solving the problems that sample data variables are more and unknown relations exist among parameter variables. The PCA is used for analyzing new variables by projecting the original data to ensure the information content contained in the original data to the maximum extent, the new variables and the original variables have linear relation, and the new variables are mutually independent.
The evaluation steps of the health degree of the low-voltage transformer based on the PCA-Mahalanobis distance are as follows:
a. calculating the normalized data X by the equation (2) * Covariance matrix sigma * ；
b. Pass equation (3) for covariance matrix ∑ * Decomposing the eigenvalue to obtain an eigenvalue lambda 1 、λ 2 、...、λ N And corresponding feature vectors p 1 、p 2 、...、p N ；
(∑ * -λE)p＝0 (3)
c. Sorting the eigenvalues from large to small, calculating the principal component contribution degree based on the formula (4), and selecting the eigenvectors corresponding to the first K maximum eigenvalues to form a load matrix P = [ P ] 1 、p 2 、...、p K ]∈R N*K And is made of
d. Obtaining the data X after dimensionality reduction by the formula (5) 1 ；
X 1 ＝X * P∈R N*K (5)
e. After the original data is subjected to dimensionality reduction, the data X subjected to X dimensionality reduction is utilized 1 The mahalanobis distance distribution of the normal data is calculated by equation (6):
wherein dist i Is the data X after the dimensionality reduction 1 Mahalanobis distance, x, of the ith sample i Is data X after dimensionality reduction 1 I =1,2, 3.., M, μ is the post-dimensionality reduction data X 1 Mean vector of (3), Σ being the reduced-dimension data X 1 The covariance matrix of (2).
f. The Mahalanobis distance distribution dist = [ dist ] of normal data is obtained 1 ,dist 2 ,...,dist M ]Then, the Mahalanobis distance threshold is obtained by the equation (7) 1 ：
threshold 1 ＝mean(dist)+3*std(dist) (7)
Where mean (dist) is the mean of the mahalanobis distance dist and std (dist) is the standard deviation of the mahalanobis distance dist.
g. In order to quantitatively describe the health degree of the low-voltage transformer, the mahalanobis distance is converted into a [0, 100] interval by a quantization function shown in an equation (8), and the larger the value is, the higher the health degree of the low-voltage transformer is, and the healthier the equipment is.
Wherein s is mh,i Is data X after dimensionality reduction 1 The mahalanobis distance health of the ith sample, exp (.) is an exponential function, and α is an adjustment coefficient, where α =3.
And calculating the health degree of the newly acquired data through the formulas (6) and (8) after PCA dimensionality reduction, so as to obtain the March distance health degree score after the low-voltage transformer is quantized at the acquisition time. The health degree evaluation model process based on the Mahalanobis distance is shown in the attached figure 2 in the specification.
The low-voltage transformer health degree assessment based on the Mahalanobis distance is not influenced by transformer monitoring parameter dimensions, the Mahalanobis distance between two points is irrelevant to a measurement unit of original data, the Mahalanobis distance has the function of amplifying variables with small changes, and the low-voltage transformer health degree assessment method has strong detection capability on initial degradation; however, the mahalanobis distance requires that the parameter variables conform to normal distribution, and only linear correlation relations among the respective variables can be considered, and the low-voltage transformer parameters do not necessarily satisfy normality and linear correlation under the actual degradation condition, so that the health degree of the low-voltage transformer is evaluated secondarily by adopting a self-coding neural network model in order to more objectively evaluate the health degree of the low-voltage transformer.
(2) And constructing a health degree evaluation model based on an autoencoder.
The self-Encoder is an unsupervised neural network learning model and comprises an Encoder Encoder and a Decoder Decode, the self-Encoder and the multilayer perceptron have similar structures and are provided with an Input layer Input, an Output layer Output and one or more hidden layers connecting the Input layer Input and the Output layer Output, but the number of nodes of the Output layer of the self-Encoder is the same as that of the nodes of the Input layer, and the goal of the self-Encoder is to reconstruct the Input of the self-Encoder. The role of Encoder is to find a compressed representation of a given high-dimensional data, and Decoder is to reconstruct the original input from the compressed representation as much as possible. The AutoEncoder is similar to PCA, but the AutoEncoder uses a nonlinear activation function to overcome the linear limitation of the PCA, the AutoEncoder also plays a role in noise reduction in the process from input to reconstruction output, and simultaneously, the learned low-dimensional representation of the AutoEncoder in the training process captures a hidden mode of data, so that the AutoEncoder can be applied to anomaly detection and further converts the anomaly degree into equipment health degree.
The automatic Encoder network is trained by normal long-term stable operation data, in the dimension reduction process, the automatic Encoder learns the interaction among various parameter characteristics and can reconstruct the interaction back to an original input variable, namely, standardized data X ^ is input into the Encoder, and after data compression, the Decoder outputs reconstruction (X ^ of X ^ so that X ^ is approximately equal to (X ^ or). As the performance of the monitored equipment degrades, this will affect the interaction between the variables, and when this occurs, the errors will increase significantly when the AutoEncoder reconstructs the input variables, and by calculating the reconstruction errors, an indication of the health of the monitored equipment can be obtained. The training process of the AutoEncoder is as follows:
a. randomly initializing a parameter matrix W and a bias vector b of the Encoder, defining an activation function of the Encoder as f, taking the activation function as a rule activation function, if the activation function is f (x) = max (0, x), after the input data is encoded by the Encoder, outputting a compression expression z, wherein the z is obtained by a formula (9):
wherein z is i For a compressed representation of the ith input sample,
is the normalized ith input sample;
b. randomly initializing a parameter matrix W ' and a bias vector b ' of the Decoder, defining that an activation function f ' of the Decoder is still a rule function, z i Outputting original input sample after decoding by Decoder
Is based on the reconstructed sample->
Obtained from formula (10):
the training goal of the autoencoder is to make the reconstruction error of the reconstructed samples and the input samples as small as possible by finding the optimal parameter θ = { W, b, W ', b' }, i.e., minimize the loss function L (W, b, W ', b'), which is expressed as formula (11):
d. the loss function L (W, b, W ', b') is minimized through an error back propagation algorithm and a gradient descent method, and then the optimal parameter theta of the AutoEncoder is obtained * ＝{W,b,W′,b′}。
e. Obtaining the reconstruction error distribution of normal data by using the trained AutoEncodererr＝[J 1 ,J 2 ,...,J M ]Then, a reconstruction error threshold is obtained by equation (13) 2 ：：
threshold 2 ＝mean(err)+3*std(err) (13)
Wherein J i Is the reconstruction error of the ith input sample, mean (err) is the mean of the reconstruction error distribution err, std (err) is the standard deviation of the reconstruction error distribution err;
f. in order to quantitatively describe the health degree of the low-voltage transformer, the reconstruction error is converted into a [0, 100] interval through a quantization function shown in an equation (14), and the larger the value is, the higher the health degree of the low-voltage transformer is, and the healthier the equipment is.
Wherein s is ae,i Is the self-encoder health of the ith input sample, exp (.) is an exponential function, β is the adjustment coefficient, where β =3 is taken.
And calculating the health degree of the newly acquired data through an equation (12) and an equation (14) to obtain the score of the self-encoder after the low-voltage transformer is quantized at the acquisition time. The health degree evaluation model flow based on the self-encoder is shown in the attached figure 3 of the specification.
(3) And fusing the health degree based on the Mahalanobis distance health degree and the self-encoder health degree.
The method provides reliable basis for objectively and accurately judging the running state of the transformer for operation and maintenance personnel, and obtains the final health degree of the low-voltage transformer by performing weighted fusion on the health degree of the Mahalanobis distance and the health degree of the self-encoder, wherein the fusion formula is shown as the formula (15).
s final ＝w hn s mh +w ae s ae (15)
Wherein s is final For weighting the degree of health, i.e. the final degree of health of the low-voltage transformer, s mh And s ae The Mahalanobis distance health degree and the self-encoder health degree of the low-voltage transformer, w mh And w ae Respectively mahalanobis distance health degree and self-organizationWeight of encoder health, w mh ,w ae ∈[0,1]And w is a mh +w ae ＝1。w mh And w ae The setting of (a) needs to be adjusted according to the actual situation, and w is set without loss of generality mh ＝w ae ＝0.5。
For the equipment for implementing the health degree assessment, the assessment is not only carried out at a certain moment, but is carried out for a plurality of times regularly or irregularly in the whole life cycle, so that a plurality of health degree assessment results arranged according to the assessment time are obtained. In order to find out the rule from the evaluation results, the evaluation results are stored in the form of a health degree file. For qualitative health level, the health level obtained through evaluation all the time roughly reflects the change rule of the equipment health degree; for quantitative health degree, the health degree obtained through evaluation of the previous time can form a health degree curve, so that the change rule of the health degree of the equipment can be reflected more truly.
It is noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.
Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and alterations can be made in these embodiments without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.
Claims (5)
1. The low-voltage transformer health degree assessment method based on the Mahalanobis distance and the self-encoder is characterized by comprising the following steps of:
1): acquiring electrical parameter data and temperature parameter data of the low-voltage transformer in long-term stable operation from a transformer historical database;
2): preprocessing the acquired transformer data to ensure that each data has homology and equivalence;
3): carrying out PCA (principal component analysis) dimensionality reduction on the basis of the preprocessed data, calculating the Mahalanobis distance of the sample data by using the data subjected to dimensionality reduction, and converting the Mahalanobis distance into a Mahalanobis distance health score by using a quantization function;
4): establishing a self-encoder model by utilizing the preprocessed data, obtaining a reconstruction error of sample data after training the model, and converting the reconstruction error into a self-encoder health score by utilizing a quantization function;
5): and based on the Mahalanobis distance health degree and the self-encoder health degree, weighting and fusing the health degree scores to obtain the final health degree score of the low-voltage transformer.
2. The mahalanobis distance and self-encoder based low voltage transformer health assessment method of claim 1, wherein: the preprocessing of the data in the step 2 comprises: filling the parameter data mean value for the problem that the parameter characteristic data is missing; in order to make the parameter characteristics comparable, the detected data of each parameter characteristic is normalized.
3. The mahalanobis distance and self-encoder based low voltage transformer health assessment method of claim 1, wherein: the calculation of converting the quantization function into the mahalanobis distance health degree in the step 3 includes performing PCA dimension reduction on the standardized data, specifically, solving a covariance matrix of the standardized data, performing eigenvalue decomposition on the obtained covariance mean value to obtain eigenvalues and corresponding eigenvectors, sorting the obtained eigenvalues from large to small, selecting eigenvectors corresponding to the first K eigenvalues with the cumulative contribution degree larger than 90% to form a load matrix based on the principal component contribution degree, performing dimension reduction on the standardized data based on the load matrix, obtaining a mean vector and a covariance matrix of the data after dimension reduction, solving the mahalanobis distance and the mahalanobis distance threshold of the sample data based on the obtained mean vector and the covariance matrix, and performing quantization conversion on the mahalanobis distance to the health degree.
4. The mahalanobis distance and self-encoder based low voltage transformer health assessment method of claim 1, wherein: the calculation of converting the quantization function into the self-Encoder health degree in the step 4 comprises randomly initializing parameter matrixes and offset vectors of the self-Encoder Encoder and the Decoder, inputting standardized data into the Encoder to obtain compressed representation of the input data, decoding the compressed representation by the Decoder to output an originally input reconstruction sample, minimizing a loss function through a back propagation algorithm and a gradient descent method to obtain optimal parameters of the self-Encoder, obtaining a reconstruction error and a reconstruction error threshold of sample data by using the trained self-Encoder, and quantizing and converting the reconstruction error into the health degree.
5. The mahalanobis distance and self-encoder based low voltage transformer health assessment method of claim 1, wherein: the calculating of the weighted health degree in the step 5 comprises: the method provides reliable basis for objectively and accurately judging the running state of the transformer for operation and maintenance personnel, and obtains the final health degree of the low-voltage transformer by performing weighted fusion on the health degree of the Mahalanobis distance and the health degree of the self-encoder, wherein the fusion formula is as follows (1):
s final ＝w mh s mh +w ae s ae (1)
wherein s is final For weighting the degree of health, i.e. the final degree of health of the low-voltage transformer, s mh And s ae The Mahalanobis distance health degree and the self-encoder health degree of the low-voltage transformer, w mh And w ae Are respectively asWeight of Mahalanobis distance health and self-encoder health, w mh ，w ae ∈[0，1]And w is mh +w ae ＝1。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211663923.6A CN115906520A (en) | 2022-12-23 | 2022-12-23 | Low-voltage transformer health degree evaluation method based on Mahalanobis distance and self-encoder |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211663923.6A CN115906520A (en) | 2022-12-23 | 2022-12-23 | Low-voltage transformer health degree evaluation method based on Mahalanobis distance and self-encoder |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115906520A true CN115906520A (en) | 2023-04-04 |
Family
ID=86479544
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211663923.6A Pending CN115906520A (en) | 2022-12-23 | 2022-12-23 | Low-voltage transformer health degree evaluation method based on Mahalanobis distance and self-encoder |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115906520A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117290670A (en) * | 2023-11-27 | 2023-12-26 | 南京中鑫智电科技有限公司 | Transformer bushing insulation state estimation method based on enhancement filter algorithm |
-
2022
- 2022-12-23 CN CN202211663923.6A patent/CN115906520A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117290670A (en) * | 2023-11-27 | 2023-12-26 | 南京中鑫智电科技有限公司 | Transformer bushing insulation state estimation method based on enhancement filter algorithm |
| CN117290670B (en) * | 2023-11-27 | 2024-01-26 | 南京中鑫智电科技有限公司 | Transformer bushing insulation state estimation method based on enhancement filter algorithm |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111666982B (en) | Electromechanical equipment fault diagnosis method based on deep neural network | |
| CN111813084B (en) | Mechanical equipment fault diagnosis method based on deep learning | |
| CN112834211A (en) | Fault early warning method for transmission system of wind turbine generator | |
| CN108763729A (en) | Process industry Mechatronic Systems couple state appraisal procedure based on network structure entropy | |
| CN110570030A (en) | Wind power cluster power interval prediction method and system based on deep learning | |
| CN115906520A (en) | Low-voltage transformer health degree evaluation method based on Mahalanobis distance and self-encoder | |
| CN115563563A (en) | Fault diagnosis method and device based on transformer oil chromatographic analysis | |
| CN113554361B (en) | Comprehensive energy system data processing and calculating method and processing system | |
| CN114186379A (en) | Transformer state evaluation method based on echo network and deep residual error neural network | |
| CN112580198A (en) | Improved optimization classification method for transformer state evaluation | |
| CN117074961A (en) | Variation self-encoder power battery abnormality detection method, system, device and storage medium | |
| CN112883634A (en) | DC measurement system state prediction method and system based on multi-dimensional analysis | |
| CN116840765B (en) | Voltage transformer error state evaluation method based on multivariate time sequence analysis | |
| CN114358116A (en) | Oil-immersed transformer fault diagnosis method and system and readable storage medium | |
| CN114021758A (en) | Operation and maintenance personnel intelligent recommendation method and device based on fusion of gradient lifting decision tree and logistic regression | |
| CN113642244A (en) | Power metering equipment fault prediction method based on artificial intelligence | |
| CN114118460A (en) | Low-voltage transformer area line loss rate abnormity detection method and device based on variational self-encoder | |
| CN113779879A (en) | Medium-and-long-term electricity utilization abnormity detection method based on LSTM-seq2seq-attention model | |
| CN116776209A (en) | Method, system, equipment and medium for identifying operation state of gateway metering device | |
| CN112633331A (en) | Self-encoder-based thermal power plant auxiliary machine health state assessment method | |
| CN116430164A (en) | Cable online monitoring method based on distributed temperature measurement and fault waveform analysis | |
| CN114186618A (en) | Method for judging working state of converter valve | |
| CN114581699A (en) | Transformer state evaluation method based on deep learning model in consideration of multi-source information | |
| CN114358160A (en) | Data anomaly detection method in power system | |
| Tian et al. | Subsystem Measurement-Based Condition Assessment for Power Transformers via Joint Inference of Data and Knowledge |
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 |