CN113011464A - Comprehensive prediction method for running state of transformer based on multi-dimensional data evaluation - Google Patents
Comprehensive prediction method for running state of transformer based on multi-dimensional data evaluation Download PDFInfo
- Publication number
- CN113011464A CN113011464A CN202110208048.1A CN202110208048A CN113011464A CN 113011464 A CN113011464 A CN 113011464A CN 202110208048 A CN202110208048 A CN 202110208048A CN 113011464 A CN113011464 A CN 113011464A
- Authority
- CN
- China
- Prior art keywords
- signal
- state
- prediction
- transformer
- data
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/50—Testing of electric apparatus, lines, cables or components for short-circuits, continuity, leakage current or incorrect line connections
- G01R31/62—Testing of transformers
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
Abstract
The invention provides a comprehensive prediction method for the running state of a transformer based on multi-dimensional data evaluation, and relates to the technical field of state evaluation of electrical equipment. The method comprises the steps of constructing a training set by acquiring multi-dimensional historical data corresponding to transformers with different operation ages, and carrying out RBF kernel function training, MSE parameter optimization and prediction result error judgment; according to calculation and prediction, Euclidean distance and data information entropy weight of the multidimensional data are obtained; according to the prediction result of the multidimensional data, predicting the state of the transformer component and predicting the overall operation state of the transformer; the full view perception and early warning of the running state of the transformer are realized.
Description
Technical Field
The invention relates to the technical field of state evaluation of electrical equipment, in particular to a comprehensive prediction method for an operating state of a transformer based on multi-dimensional data evaluation.
Background
With the large-scale construction and popularization of ultrahigh voltage engineering in China, higher requirements on power supply reliability are provided while the capacity of a power grid is improved. As a key junction device of a power system, the operation state of a large-scale transformer is directly related to the operation reliability and stability of a power grid. Because the transformer is connected with power grids with different voltage grades, if the transformer fails, major power accidents such as voltage fluctuation, equipment damage, large-area power failure and the like can be caused, and serious economic loss and social influence are caused. Therefore, the method can accurately detect and extract the characteristic information of the transformer, carry out state assessment and effective state prediction, is beneficial to improving the operation maintenance and overhaul level of the transformer, prevents and reduces the occurrence risk of transformer faults, and is the premise for realizing state overhaul of the power system.
The evaluation of the running state of the transformer is to evaluate the corresponding running health state grade by analyzing the working condition of the transformer; the transformer operation state prediction is to predict the change trend of the transformer operation state in a period of time in the future according to the transformer operation historical data, and provide quick and accurate information for operation and maintenance personnel. Transformer condition prediction is a key to transformer condition maintenance. However, since the deterioration process of the operating state is complicated due to many factors affecting the state of the transformer, the prediction of the state of the transformer is a complicated and important subject. With the development of a large overhaul system by a national power grid company, technologies and means capable of comprehensively predicting the operation state of the on-site transformer are urgently needed. The comprehensive prediction method for researching the running state of the transformer has important theoretical research value and engineering application significance for reducing the occurrence risk of transformer faults and ensuring the running safety of the transformer faults.
In the field of the running state of the transformer at home and abroad at present, research on a comprehensive prediction method of the running state of the transformer is not developed, and an effective comprehensive prediction model is also lacked. Therefore, for a transformer which does not have a fault or has an unobvious fault trend, the existing research results cannot predict the future deformation state of the running transformer winding. If the development and change trend of the characteristic indexes of the transformer winding can be effectively predicted through a certain prediction means, the running condition of the winding can be mastered in advance. For operating transformers, studies for building comprehensive predictive models based on periodically continuous historical data are lacking. The above factors severely limit the research and on-line application of the transformer winding deformation state prediction method.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a comprehensive prediction method for the running state of a transformer based on multi-dimensional data evaluation.
A comprehensive prediction method for the running state of a transformer based on multi-dimensional data evaluation comprises the following steps:
step 1: acquiring multi-dimensional historical data corresponding to transformers with different operation years to construct a training set and a prediction model;
the transformer multi-dimensional historical data comprises a wire outlet end temperature signal, a winding vibration signal, a short-circuit reactance signal, a grounding current signal, an iron core magnetic field intensity signal, an oil temperature signal, an oil level signal, a micro-water content signal, a partial discharge signal, an oil chromatography signal, a dielectric loss signal, an insulation resistance signal, a box body metal content signal, a tap switch vibration signal and a fan vibration signal; setting a data set Xi (t operation time limit), and setting Yi (t operation time limit corresponding transformer multidimensional historical data), wherein t is 1, 2. Taking n data in front of a training sample set as training samples, and taking the rest data as test samples; performing phase space reconstruction on the training sample and the test sample, converting the operation age sequence into a matrix form, and using the matrix form for LS.SVM learning of a least square support vector machine;
the least squares support vector machine LS.SVM learned samples are as follows:
where m is the embedding dimension of the model and n is the number of training samples.
Establishing a mapping relation between a training sample and a prediction model:
wherein the content of the first and second substances,and predicting the value of the multidimensional historical data of the running years in advance.
Let training sample D { (x)k,yk) 1, 2., N }, where x isk∈Rn,yke.R are input and output data, xkIndicates the running age, y, of the transformer corresponding to the test datakRefers to the multidimensional data of the transformer obtained by testing, and yk=f(xk) Where f (x) is a non-linear mapping function expressed as:
where ω is the weight vector of the feature space, b is the offset,predicting a least square fitting function of the model, wherein N is the number of samples;
then the least squares support vector machine regression estimation solves the following optimization problem:
in the formula, ek∈RnIs an error variable and gamma is an adjustment parameter. ω is the weight vector of the feature space, J (ω, e) is the least squares estimation function, and b is the offset.
Introducing Lagrangian functionsαkE, R is a Lagrange multiplier, and the following linear equation set is obtained according to the optimization theory:
wherein y is [ y ═ y1,...,yN]T,1v=[1,...,1]T,α=[α1,...,αN]T,k,l=1,...,N,K(xk,xl) I is a Lagrange function matrix parameter for a kernel function meeting the Mexican condition;
and (3) solving a and b by using a least square method, and then performing regression estimation on the least square support vector machine to obtain a formula expressed as a prediction model:
step 2: carrying out RBF kernel function training and MSE parameter optimization on the model;
the RBF kernel function training is to carry out cross validation method optimization on a penalty factor c and an RBF kernel function parameter gamma, wherein the RBF kernel is as follows:
K(xi,xj)=exp(-||xi-xj||2/2σ2)
sigma is a width parameter of model optimization;
is provided withThe prediction accuracy optimization problem is transformed into the following minimization problem,
the minimum value of the above equation depends on the choice of the parameters (c, γ), and the best parameter is chosen to maximize SVM regression performance.
Training a least square support vector machine by using the determined parameters and the training samples to obtain a prediction model; and (3) normalizing the data by adopting range normalization, wherein the normalized value is in the range of [0, 1 ]:
in the formula, X (i) is a value to be normalized in a certain column of a transformer multi-dimensional data sample, maxx (i) and minx (i) are maximum and minimum values of the column of the sample respectively, Y (i) is an actual value at the moment i, maxy (i) and miny (i) are actual maximum and minimum values respectively, and X '(i) and Y' (i) are normalized values respectively;
and step 3: to pairRegression function y of training samples according totTraining and judging the error of a prediction result:
ɑkis a Lagrange multiplier, K is a kernel function satisfying the Mexican condition, and b is an offset;
parameters corresponding to the time to be predicted, specifically including a leading-out terminal temperature signal, a winding vibration signal, a short-circuit reactance signal, a grounding current signal, an iron core magnetic field intensity signal, an oil temperature signal, an oil level signal, a micro-water content signal, a partial discharge signal, an oil chromatography signal, a dielectric loss signal, an insulation resistance signal, a box body metal content signal, a tap switch vibration signal and a fan vibration signal) are input into a trained least square support vector machine model for prediction, and a prediction result is obtained; to test the validity of the model, the prediction results were evaluated using a prediction error, which is calculated as:
in the formula, xiCorresponding to the actual value of the multidimensional data of the transformer for a certain operation age sequence;a calculated value for the prediction data; the prediction error is large, which indicates that the precision of the prediction method is low, and the prediction process is continued until the prediction precision meets the preset threshold;
if the prediction error is larger than the given error range, returning to the step 2 until the prediction result is smaller than the given error range of the model; the result obtained by prediction is used as new input of a prediction model of the least square support vector machine, and the corresponding value of the target operation age can be obtained by prediction, so that multi-step prediction is realized; respectively carrying out multi-dimensional data prediction of the next target operation year sequence by adopting a trained prediction model to obtain a prediction result of the multi-dimensional data of the transformer, namely finally obtaining the predicted value of the (n + l) th step:
and 4, step 4: calculating and predicting Euclidean distance of the multi-dimensional data and data information entropy weight;
and calculating the quantized values of the multidimensional data one by one according to the multidimensional data obtained by prediction. Because the evaluation indexes have different dimensions and magnitude levels, the evaluation indexes cannot be directly compared, the evaluation indexes need to be standardized, each index in the index layer is an intermediate index, and the quantization method is shown as the formula:
wherein P is normalized index value, x is measured index value, a1And a2A minimum value and a maximum value representing the index;
and (3) calculating the quantized value of the multi-dimensional data prediction result and the Euclidean distance between the quantized value of the standard multi-dimensional data when the transformer leaves the factory, wherein the calculation formula is as follows:
in the formula ktPredicting a quantized value, k, of a result for multidimensional dataiAnd D is the Euclidean distance of the multidimensional data.
And evaluating the state of the multidimensional data according to the calculated Euclidean distance of the multidimensional data as the state value of the multidimensional data, and calculating the information entropy of each multidimensional data, namely:
Hj=-PlnP
wherein i is 1,2, …, n; j is 1,2, …, m.
Calculating the entropy weight w of each multidimensional data, namely:
and 5: according to the prediction result of the multidimensional data, predicting the state of the transformer component and predicting the overall operation state of the transformer;
in the component state prediction layer, transformer component state prediction is carried out, and after quantitative calculation is carried out according to a leading-out terminal temperature signal, a winding vibration signal, a short-circuit reactance signal, a grounding current signal, an iron core magnetic field intensity signal, an oil temperature signal, an oil level signal, a micro-water content signal, a partial discharge signal, an oil chromatography signal, a dielectric loss signal, an insulation resistance signal, a box metal content signal, a tap switch vibration signal and a fan vibration signal which are obtained through prediction in the multi-dimensional data prediction layer, the transformer component state (a winding state, an iron core state, a transformer oil state, an insulation component state, a sleeve state and an accessory state) is predicted by adopting the following formula.
In the formula, DjState values for the transformer component states (winding state, core state, transformer oil state, insulation component state, bushing state, accessory state).The characteristic data is a set of Euclidean distances of multidimensional data, namely a leading-out terminal temperature signal, a winding vibration signal, a short-circuit reactance signal, a grounding current signal, an iron core magnetic field intensity signal, an oil temperature signal, an oil level signal, a micro-water content signal, a partial discharge signal, an oil chromatography signal, a dielectric loss signal, an insulation resistance signal, a box body metal content signal, a tap switch vibration signal and a fan vibration signal. w is aiAnd the weight of the entropy value corresponding to the Euclidean distance of the multidimensional data.
And in the equipment state prediction layer, predicting the overall operation state of the transformer. And predicting the overall operation state of the transformer according to the prediction result of the transformer component states (winding states, iron core states, transformer oil states, insulating component states, sleeve states and accessory states) predicted in the component prediction layer.
In the formula, DmThe state value of the overall running state of the transformer is obtained.Is a collection of state values for the transformer component states (winding state, core state, transformer oil state, insulation component state, bushing state, accessory state). w is ajThe entropy weight is corresponding to the state value of the component state (winding state, iron core state, transformer oil state, insulating component state, sleeve state and accessory state).
Adopt the produced beneficial effect of above-mentioned technical scheme to lie in:
the invention provides a comprehensive prediction method for the running state of a transformer based on multi-dimensional data evaluation, which can realize panoramic perception and early warning of the running state of the transformer, break through the limitation of single criterion in the past state monitoring, improve the running stability and the running life of the power transformer, promote the break through and the progress of scientific technology in the field of state prediction of power equipment, and improve the state overhaul efficiency and the operation and maintenance level of the power industry in China.
Drawings
FIG. 1 is a flow chart of a comprehensive prediction method for the operation state of a transformer according to the present invention;
FIG. 2 is a diagram of a comprehensive prediction model of the operation state of the transformer according to the present invention.
Detailed Description
The following detailed description of embodiments of the present invention is provided in connection with the accompanying drawings and examples. The following examples are intended to illustrate the invention but are not intended to limit the scope of the invention.
A comprehensive prediction method for the running state of a transformer based on multi-dimensional data evaluation is disclosed, as shown in FIG. 1, and comprises the following steps:
step 1: acquiring multi-dimensional historical data corresponding to transformers with different operation years to construct a training set and a prediction model;
the transformer multi-dimensional historical data comprises a wire outlet end temperature signal, a winding vibration signal, a short-circuit reactance signal, a grounding current signal, an iron core magnetic field intensity signal, an oil temperature signal, an oil level signal, a micro-water content signal, a partial discharge signal, an oil chromatography signal, a dielectric loss signal, an insulation resistance signal, a box body metal content signal, a tap switch vibration signal and a fan vibration signal; setting a data set Xi (t operation time limit), and setting Yi (t operation time limit corresponding transformer multidimensional historical data), wherein t is 1, 2. Taking n data in front of a training sample set as training samples, and taking the rest data as test samples; performing phase space reconstruction on the training sample and the test sample, converting the operation age sequence into a matrix form, and using the matrix form for LS.SVM learning of a least square support vector machine;
support Vector Machines (SVMs) are the first method proposed by cortex and Vapnik in 1995 to identify patterns, and are a new method and technology in data mining. With the intensive research of a large number of experts and scholars at home and abroad on the SVM, the application field of the SVM is continuously extended, and the importance of the SVM on the aspects of density estimation, regression prediction and the like is more and more obvious. The support vector machine prediction model has better generalization capability on predicted future samples, and the generalization capability is superior to that of a traditional neural network model and a fuzzy algorithm model. The standard support vector machine algorithm is to convert an actual problem into a quadratic convex programming problem with inequality constraint, and the least square support vector machine is to convert the actual problem into a problem of solving a group of linear equations, so that the calculation is simplified, and the convergence speed is improved. Due to the fact that the multi-dimensional historical data of the transformer is limited, the support vector machine algorithm can predict the future data change trend more effectively. Therefore, the theory is adopted as a prediction method of the overall operation state of the transformer, and the samples learned by the least squares support vector machine LS.SVM are as follows:
where m is the embedding dimension of the model and n is the number of training samples.
Establishing a mapping relation between a training sample and a prediction model:
wherein the content of the first and second substances,and predicting the value of the multidimensional historical data of the running years in advance.
Let training sample D { (x)k,yk)|k1,2, N, where x isk∈Rn,yke.R are input and output data, xkIndicates the running age, y, of the transformer corresponding to the test datakRefers to the multidimensional data of the transformer obtained by testing, and yk=f(xk) Where f (x) is a non-linear mapping function expressed as:
where ω is the weight vector of the feature space, b is the offset,predicting a least square fitting function of the model, wherein N is the number of samples;
then the least squares support vector machine regression estimation solves the following optimization problem:
in the formula, ek∈RnIs an error variable and gamma is an adjustment parameter. ω is the weight vector of the feature space, J (ω, e) is the least squares estimation function, and b is the offset.
Introducing Lagrangian functionsαkE, R is a Lagrange multiplier, and the following linear equation set is obtained according to the optimization theory:
wherein y is [ y ═ y1,...,yN]T,1v=[1,...,1]T,α=[α1,...,αN]T,k,l=1,...,N,K(xk,xl) I is a Lagrange function matrix parameter for a kernel function meeting the Mexican condition;
and (3) solving a and b by using a least square method, and then performing regression estimation on the least square support vector machine to obtain a formula expressed as a prediction model:
step 2: carrying out RBF kernel function training and MSE parameter optimization on the model;
the RBF kernel function training is to carry out cross validation method optimization on a penalty factor c and an RBF kernel function parameter gamma, wherein the RBF kernel is as follows:
K(xi,xj)=exp(-||xi-xj||2/2σ2)
sigma is a width parameter of model optimization;
is provided withThe prediction accuracy optimization problem is transformed into the following minimization problem,
the minimum value of the above equation depends on the choice of the parameters (c, γ), and the best parameter is chosen to maximize SVM regression performance.
Training a least square support vector machine by using the determined parameters and the training samples to obtain a prediction model; and (3) normalizing the data by adopting range normalization, wherein the normalized value is in the range of [0, 1 ]:
in the formula, X (i) is a value to be normalized in a certain column of a transformer multi-dimensional data sample, maxx (i) and minx (i) are maximum and minimum values of the column of the sample respectively, Y (i) is an actual value at the moment i, maxy (i) and miny (i) are actual maximum and minimum values respectively, and X '(i) and Y' (i) are normalized values respectively;
and step 3: regression function y for training samples according totTraining and judging the error of a prediction result:
ɑkis a Lagrange multiplier, K is a kernel function satisfying the Mexican condition, and b is an offset;
parameters corresponding to the time to be predicted, specifically including a leading-out terminal temperature signal, a winding vibration signal, a short-circuit reactance signal, a grounding current signal, an iron core magnetic field intensity signal, an oil temperature signal, an oil level signal, a micro-water content signal, a partial discharge signal, an oil chromatography signal, a dielectric loss signal, an insulation resistance signal, a box body metal content signal, a tap switch vibration signal and a fan vibration signal) are input into a trained least square support vector machine model for prediction, and a prediction result is obtained; to test the validity of the model, the prediction results were evaluated using a prediction error, which is calculated as:
in the formula, xiCorresponding to the actual value of the multidimensional data of the transformer for a certain operation age sequence;a calculated value for the prediction data; the prediction error is large, which indicates that the precision of the prediction method is low, and the prediction process is continued until the prediction precision meets the preset threshold;
if the prediction error is larger than the given error range, returning to the step 2 until the prediction result is smaller than the given error range of the model; the result obtained by prediction is used as new input of a prediction model of the least square support vector machine, and the corresponding value of the target operation age can be obtained by prediction, so that multi-step prediction is realized; respectively carrying out multi-dimensional data prediction of the next target operation year sequence by adopting a trained prediction model to obtain a prediction result of the multi-dimensional data of the transformer, namely finally obtaining the predicted value of the (n + l) th step:
and 4, step 4: calculating and predicting Euclidean distance of the multi-dimensional data and data information entropy weight;
and calculating the quantized values of the multidimensional data one by one according to the multidimensional data obtained by prediction. Because the evaluation indexes have different dimensions and magnitude levels, the evaluation indexes cannot be directly compared, the evaluation indexes need to be standardized, each index in the index layer is an intermediate index, and the quantization method is shown as the formula:
wherein P is normalized index value, x is measured index value, a1And a2The minimum value and the maximum value of the index are shown, and the values are determined from the published documents of the industry, namely the preventive test regulations of the power equipment and the operational regulations of the transformer.
And (3) calculating the quantized value of the multi-dimensional data prediction result and the Euclidean distance between the quantized value of the standard multi-dimensional data when the transformer leaves the factory, wherein the calculation formula is as follows:
in the formula ktPredicting a quantized value, k, of a result for multidimensional dataiAnd D is the Euclidean distance of the multidimensional data.
And evaluating the state of the multidimensional data according to the calculated Euclidean distance of the multidimensional data as the state value of the multidimensional data, and calculating the information entropy of each multidimensional data, namely:
Hj=-PlnP
wherein i is 1,2, …, n; j is 1,2, …, m.
Calculating the entropy weight w of each multidimensional data, namely:
and 5: according to the prediction result of the multidimensional data, predicting the state of the transformer component and the overall operation state of the transformer, as shown in FIG. 2;
in the component state prediction layer, transformer component state prediction is carried out, and after quantitative calculation is carried out according to a leading-out terminal temperature signal, a winding vibration signal, a short-circuit reactance signal, a grounding current signal, an iron core magnetic field intensity signal, an oil temperature signal, an oil level signal, a micro-water content signal, a partial discharge signal, an oil chromatography signal, a dielectric loss signal, an insulation resistance signal, a box metal content signal, a tap switch vibration signal and a fan vibration signal which are obtained through prediction in the multi-dimensional data prediction layer, the transformer component state (a winding state, an iron core state, a transformer oil state, an insulation component state, a sleeve state and an accessory state) is predicted by adopting the following formula.
In the formula, DjState values for the transformer component states (winding state, core state, transformer oil state, insulation component state, bushing state, accessory state).The characteristic data is a set of Euclidean distances of multidimensional data, namely a leading-out terminal temperature signal, a winding vibration signal, a short-circuit reactance signal, a grounding current signal, an iron core magnetic field intensity signal, an oil temperature signal, an oil level signal, a micro-water content signal, a partial discharge signal, an oil chromatography signal, a dielectric loss signal, an insulation resistance signal, a box body metal content signal, a tap switch vibration signal and a fan vibration signal. w is aiAnd the weight of the entropy value corresponding to the Euclidean distance of the multidimensional data.
And in the equipment state prediction layer, predicting the overall operation state of the transformer. And predicting the overall operation state of the transformer according to the prediction result of the transformer component states (winding states, iron core states, transformer oil states, insulating component states, sleeve states and accessory states) predicted in the component prediction layer.
In the formula, DmThe state value of the overall running state of the transformer is obtained.Is a collection of state values for the transformer component states (winding state, core state, transformer oil state, insulation component state, bushing state, accessory state). w is ajThe entropy weight is corresponding to the state value of the component state (winding state, iron core state, transformer oil state, insulating component state, sleeve state and accessory state).
The foregoing description is only exemplary of the preferred embodiments of the disclosure and is illustrative of the principles of the technology employed. It will be appreciated by those skilled in the art that the scope of the invention in the embodiments of the present disclosure is not limited to the specific combination of the above-mentioned features, but also encompasses other embodiments in which any combination of the above-mentioned features or their equivalents is made without departing from the inventive concept as defined above. For example, the above features and (but not limited to) technical features with similar functions disclosed in the embodiments of the present disclosure are mutually replaced to form the technical solution.
Claims (3)
1. A comprehensive prediction method for the running state of a transformer based on multi-dimensional data evaluation is characterized by comprising the following steps:
step 1: acquiring multi-dimensional historical data corresponding to transformers with different operation years to construct a training set and a prediction model;
the transformer multi-dimensional historical data comprises a wire outlet end temperature signal, a winding vibration signal, a short-circuit reactance signal, a grounding current signal, an iron core magnetic field intensity signal, an oil temperature signal, an oil level signal, a micro-water content signal, a partial discharge signal, an oil chromatography signal, a dielectric loss signal, an insulation resistance signal, a box body metal content signal, a tap switch vibration signal and a fan vibration signal; setting a data set Xi (t operation time limit), and setting Yi (t operation time limit corresponding transformer multidimensional historical data), wherein t is 1, 2. Taking n data in front of a training sample set as training samples, and taking the rest data as test samples; performing phase space reconstruction on the training sample and the test sample, converting the operation age sequence into a matrix form, and using the matrix form for LS.SVM learning of a least square support vector machine;
step 2: carrying out RBF kernel function training and MSE parameter optimization on the model;
the RBF kernel function training is to carry out cross validation method optimization on a penalty factor c and an RBF kernel function parameter gamma, wherein the RBF kernel is as follows:
K(xi,xj)=exp(-||xi-xj||2/2σ2)
sigma is a width parameter of model optimization;
is provided withThe prediction accuracy optimization problem is transformed into the following minimization problem,
the minimum value of the above formula depends on the selection of parameters (c, gamma), and the optimal parameter is selected to optimize the SVM regression performance;
training a least square support vector machine by using the determined parameters and the training samples to obtain a prediction model; and (3) normalizing the data by adopting range normalization, wherein the normalized value is in the range of [0, 1 ]:
in the formula, X (i) is a value to be normalized in a certain column of a transformer multi-dimensional data sample, maxx (i) and minx (i) are maximum and minimum values of the column of the sample respectively, Y (i) is an actual value at the moment i, maxy (i) and miny (i) are actual maximum and minimum values respectively, and X '(i) and Y' (i) are normalized values respectively;
and step 3: regression function y for training samples according totTraining and judging the error of a prediction result:
ɑkis a Lagrange multiplier, K is a kernel function satisfying the Mexican condition, and b is an offset;
parameters corresponding to the time to be predicted specifically comprise a leading-out terminal temperature signal, a winding vibration signal, a short-circuit reactance signal, a grounding current signal, an iron core magnetic field intensity signal, an oil temperature signal, an oil level signal, a micro-water content signal, a partial discharge signal, an oil chromatography signal, a dielectric loss signal, an insulation resistance signal, a box body metal content signal, a tap switch vibration signal and a fan vibration signal, and are input into a trained least square support vector machine model for prediction to obtain a prediction result; to test the validity of the model, the prediction results were evaluated using a prediction error, which is calculated as:
in the formula, xiCorresponding to the actual value of the multidimensional data of the transformer for a certain operation age sequence;a calculated value for the prediction data; the prediction error is large, which indicates that the precision of the prediction method is low, and the prediction process is continued until the prediction precision meets the preset threshold;
if the prediction error is larger than the given error range, returning to the step 2 until the prediction result is smaller than the given error range of the model; the result obtained by prediction is used as new input of a prediction model of the least square support vector machine, and the corresponding value of the target operation age can be obtained by prediction, so that multi-step prediction is realized; respectively carrying out multi-dimensional data prediction of the next target operation year sequence by adopting a trained prediction model to obtain a prediction result of the multi-dimensional data of the transformer, namely finally obtaining the predicted value of the (n + l) th step:
and 4, step 4: calculating and predicting Euclidean distance of the multi-dimensional data and data information entropy weight;
step 4.1: calculating the quantized values of the multidimensional data one by one aiming at the multidimensional data obtained by prediction; the quantization method is shown as the formula:
wherein P is normalized index value, x is measured index value, a1And a2A minimum value and a maximum value representing the index;
step 4.2: and (3) calculating the quantized value of the multi-dimensional data prediction result and the Euclidean distance between the quantized value of the standard multi-dimensional data when the transformer leaves the factory, wherein the calculation formula is as follows:
in the formula ktPredicting a quantized value, k, of a result for multidimensional dataiD is the Euclidean distance of the multidimensional data;
step 4.3: and evaluating the state of the multidimensional data according to the calculated Euclidean distance of the multidimensional data as the state value of the multidimensional data, and calculating the information entropy of each multidimensional data, namely:
Hj=-PlnP
wherein i is 1,2, …, n; j is 1,2, …, m;
calculating the entropy weight w of each multidimensional data, namely:
and 5: and predicting the state of the transformer component and the overall operation state of the transformer according to the prediction result of the multidimensional data.
2. The comprehensive transformer operating state prediction method based on multi-dimensional data evaluation as claimed in claim 1, wherein the samples learned by the least squares support vector machine ls.svm in step 1 are:
in the formula, m is the embedding dimension of the model, and n is the number of training samples;
establishing a mapping relation between a training sample and a prediction model:
wherein the content of the first and second substances,for one-step predictive value ahead of operating age multi-dimensional historical data,
let training sample D { (x)k,yk) 1, 2., N }, where x isk∈Rn,yke.R are input and output data, xkIndicates the running age, y, of the transformer corresponding to the test datakRefers to the multidimensional data of the transformer obtained by testing, and yk=f(xk) Where f (x) is a non-linear mapping function expressed as:
where ω is the weight vector of the feature space, b is the offset,predicting a least square fitting function of the model, wherein N is the number of samples;
then the least squares support vector machine regression estimation solves the following optimization problem:
in the formula, ek∈RnIs an error variable, gamma is an adjustment parameter, omega is a weight vector of a characteristic space, J (omega, e) is a least square estimation function, and b is an offset;
introducing Lagrangian functionsFor lagrange multipliers, according to the optimization theory, the following linear system of equations is obtained:
wherein y is [ y ═ y1,...,yN]T,1v=[1,...,1]T,α=[α1,...,αN]T,k,l=1,...,N,K(xk,xl) I is a Lagrange function matrix parameter for a kernel function meeting the Mexican condition;
and (3) solving a and b by using a least square method, and then performing regression estimation on the least square support vector machine to obtain a formula expressed as a prediction model:
3. the comprehensive transformer operating state prediction method based on multi-dimensional data evaluation as claimed in claim 1, wherein in step 5:
the transformer component state prediction is carried out in a component state prediction layer, and the transformer component state is predicted by adopting the following formula after quantitative calculation according to a leading-out terminal temperature signal, a winding vibration signal, a short-circuit reactance signal, a grounding current signal, an iron core magnetic field intensity signal, an oil temperature signal, an oil level signal, a micro-water content signal, a partial discharge signal, an oil chromatography signal, a dielectric loss signal, an insulation resistance signal, a box metal content signal, a tap switch vibration signal and a fan vibration signal which are obtained by prediction in a multi-dimensional data prediction layer, and specifically comprises a winding state, an iron core state, a transformer oil state, an insulation component state, a sleeve state and an accessory state:
in the formula, DjState values for transformer component states (winding state, core state, transformer oil state, insulation component state, bushing state, accessory state);the characteristic of the method is that the characteristic is a set of Euclidean distances of multidimensional data, namely a leading-out terminal temperature signal, a winding vibration signal, a short-circuit reactance signal, a grounding current signal, an iron core magnetic field intensity signal, an oil temperature signal, an oil level signal, a micro-water content signal, a partial discharge signal, an oil chromatography signal, a dielectric loss signal, an insulation resistance signal, a box body metal content signal, a tapping switch vibration signal and a fan vibration signal; w is aiEntropy weight corresponding to Euclidean distance of the multidimensional data;
the overall running state of the transformer is predicted in an equipment state prediction layer; and predicting the overall operation state of the transformer according to the prediction result of the transformer component states (winding states, iron core states, transformer oil states, insulating component states, sleeve states and accessory states) predicted in the component prediction layer:
in the formula, DmThe state value of the integral running state of the transformer is obtained;the transformer component state specifically comprises a set of state values of a winding state, an iron core state, a transformer oil state, an insulation component state, a sleeve state and an accessory state; w is ajThe component states specifically include entropy weight corresponding to state values of a winding state, an iron core state, a transformer oil state, an insulating component state, a sleeve state and an accessory state.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110208048.1A CN113011464A (en) | 2021-02-25 | 2021-02-25 | Comprehensive prediction method for running state of transformer based on multi-dimensional data evaluation |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110208048.1A CN113011464A (en) | 2021-02-25 | 2021-02-25 | Comprehensive prediction method for running state of transformer based on multi-dimensional data evaluation |
Publications (1)
Publication Number | Publication Date |
---|---|
CN113011464A true CN113011464A (en) | 2021-06-22 |
Family
ID=76385871
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110208048.1A Pending CN113011464A (en) | 2021-02-25 | 2021-02-25 | Comprehensive prediction method for running state of transformer based on multi-dimensional data evaluation |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113011464A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114167221A (en) * | 2021-12-13 | 2022-03-11 | 华北电力大学(保定) | Epoxy resin insulation aging discrimination and inspection method under different voltage frequencies |
CN114678962A (en) * | 2022-05-24 | 2022-06-28 | 山东科华电力技术有限公司 | Distributed array temperature measurement abnormal data transmission monitoring system based on power internet of things |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108564204A (en) * | 2018-03-23 | 2018-09-21 | 西安理工大学 | Least square method supporting vector machine power predicating method based on maximal correlation entropy criterion |
CN108680814A (en) * | 2018-08-29 | 2018-10-19 | 国网河北省电力有限公司电力科学研究院 | A kind of various dimensions running state of transformer evaluation method |
CN108805217A (en) * | 2018-06-20 | 2018-11-13 | 山东大学 | A kind of health state of lithium ion battery method of estimation and system based on support vector machines |
CN109740859A (en) * | 2018-12-11 | 2019-05-10 | 国网山东省电力公司淄博供电公司 | Transformer condition evaluation and system based on Principal Component Analysis and support vector machines |
CN110940374A (en) * | 2019-10-31 | 2020-03-31 | 沈阳工业大学 | Transformer health grade evaluation system and method based on big data fusion |
-
2021
- 2021-02-25 CN CN202110208048.1A patent/CN113011464A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108564204A (en) * | 2018-03-23 | 2018-09-21 | 西安理工大学 | Least square method supporting vector machine power predicating method based on maximal correlation entropy criterion |
CN108805217A (en) * | 2018-06-20 | 2018-11-13 | 山东大学 | A kind of health state of lithium ion battery method of estimation and system based on support vector machines |
CN108680814A (en) * | 2018-08-29 | 2018-10-19 | 国网河北省电力有限公司电力科学研究院 | A kind of various dimensions running state of transformer evaluation method |
CN109740859A (en) * | 2018-12-11 | 2019-05-10 | 国网山东省电力公司淄博供电公司 | Transformer condition evaluation and system based on Principal Component Analysis and support vector machines |
CN110940374A (en) * | 2019-10-31 | 2020-03-31 | 沈阳工业大学 | Transformer health grade evaluation system and method based on big data fusion |
Non-Patent Citations (2)
Title |
---|
曹辰: "基于机械与电气参量的变压器绕组变形状态综合评估方法", 《中国博士学位论文全文数据库 工程科技Ⅱ辑》 * |
陈彦文: "220kV油浸电力变压器绕组变形在线诊断方法研究", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 * |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114167221A (en) * | 2021-12-13 | 2022-03-11 | 华北电力大学(保定) | Epoxy resin insulation aging discrimination and inspection method under different voltage frequencies |
CN114167221B (en) * | 2021-12-13 | 2023-06-13 | 华北电力大学(保定) | Epoxy resin insulation aging discrimination and test method under different voltage frequencies |
CN114678962A (en) * | 2022-05-24 | 2022-06-28 | 山东科华电力技术有限公司 | Distributed array temperature measurement abnormal data transmission monitoring system based on power internet of things |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104914327B (en) | Transformer fault maintenance Forecasting Methodology based on real-time monitoring information | |
Jahromi et al. | An approach to power transformer asset management using health index | |
CN113011464A (en) | Comprehensive prediction method for running state of transformer based on multi-dimensional data evaluation | |
CN109490726A (en) | Electric power transformer insulated state evaluating method based on Clouds theory | |
CN103605881A (en) | Fault tree and analytic hierarchy process based evaluation method of state of power transformer | |
CN102621421A (en) | Transformer state evaluation method based on correlation analysis and variable weight coefficients | |
CN110940374A (en) | Transformer health grade evaluation system and method based on big data fusion | |
CN104537487A (en) | Assessment method of operating dynamic risk of electric transmission and transformation equipment | |
CN109583520B (en) | State evaluation method of cloud model and genetic algorithm optimization support vector machine | |
CN111080072A (en) | Distribution transformer health index evaluation method, device and system | |
CN104749519A (en) | Correlation analysis based on-load voltage regulating transformer tapping switch operating state judgment method | |
CN110058103A (en) | Intelligent transformer fault diagnosis system based on Vxworks platform | |
CN112257265A (en) | Power transformer service life prediction method based on health index and unequal-interval gray model | |
CN106501641B (en) | A kind of transformer quality state appraisal procedure | |
CN112307963A (en) | Converter transformer running state identification method based on vibration signals | |
CN110618340A (en) | Transformer state evaluation method | |
CN115112978A (en) | Fault detection method for insulating ring main unit of power system | |
CN106777512B (en) | Transformer energy-saving loss-reducing adaptability evaluation method | |
CN103728517A (en) | Correcting algorithm for transformer state assessment by considering poor working conditions | |
CN115864310A (en) | Multi-parameter fusion power transformer defect identification, state classification and active safety protection method | |
CN115564170A (en) | Reliability assessment method for intelligent power distribution terminal | |
CN114462692A (en) | Power grid old and old equipment technical improvement strategy optimization and adjustment method | |
CN103076439A (en) | Method for estimating content of characteristic gas generated by action of conversion selector of converter transformer | |
CN114019365A (en) | On-load tap-changer fault diagnosis method based on gas detection technology in oil | |
CN109711687B (en) | Insulator state fuzzy evaluation method based on improved entropy method |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210622 |
|
RJ01 | Rejection of invention patent application after publication |