CN109871660A - A kind of method for early warning and Fault Locating Method of main transformer heating accident - Google Patents
A kind of method for early warning and Fault Locating Method of main transformer heating accident Download PDFInfo
- Publication number
- CN109871660A CN109871660A CN201910236008.0A CN201910236008A CN109871660A CN 109871660 A CN109871660 A CN 109871660A CN 201910236008 A CN201910236008 A CN 201910236008A CN 109871660 A CN109871660 A CN 109871660A
- Authority
- CN
- China
- Prior art keywords
- formula
- oil
- temperature
- por
- ceramic wall
- 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.)
- Granted
Links
Abstract
A kind of method for early warning and Fault Locating Method of main transformer heating accident, the present invention relates to main transformer heating accident processing method fields.Propose a kind of monitoring of transformer safety operation characteristic and active forewarning technical research based on multisource data fusion, and superheating diagnosis and the active forewarning of bushing shell for transformer are mainly studied, to effectively solve the problems, such as the method for early warning and Fault Locating Method of the intelligent diagnostics of sleeve of main transformer thermal fault and the main transformer heating accident to give warning in advance in current substation.The present invention effectively solves the problems, such as in current substation the intelligent diagnostics of sleeve of main transformer thermal fault and gives warning in advance, the planned arrangement operation of energy and maintenance progress, so that electric system is reached intelligent, automation, brings biggish economic and social benefit for the safe and stable operation of entire electric system.
Description
Technical field
The present invention relates to main transformer heating accident processing method fields.
Background technique
With the rapid development of economy, the demand of electric power is also being continuously increased.The production and living of modern society are all the time
It is all be unable to do without the demand to electricity, operation of power networks failure can all upset the production and life of normal society, institute each time at the same time
There is quite high importance with the safe operation of power grid.Main transformer is as electric equipments, its safety in substation
It is particularly important, and in transformer device structure, casing is even more a critical elements, and inside transformer high and low pressure lead is guided to
Outside fuel tank, play insulation against ground and anchor leg.Generally require casing except electrical strength and enough as defined in meeting
Outside mechanical strength, it is necessary to there is good thermal stability and leakproofness in operation, and the overheat of moment when short circuit can be born
Deng.Bushing shell for transformer would generally lead to failure for various reasons, and fault mode mainly has shelf depreciation, leakage, inside absolutely
Edge problem etc..Once casing breaks down, it is likely that lead to large-area power-cuts, the safety fortune of entire power grid is influenced when serious
Row.Therefore, it is necessary to be monitored to bushing shell for transformer operation conditions and diagnose active forewarning.
It is that superheating or the omen of other failures, according to statistics, mistake occur for main transformer that superheating phenomenon, which occurs, in transformer equipment
The ratio that thermal fault accounts for transforming plant main transformer equipment fault quantity is maximum.By taking power transformer as an example, in whole failures of transformer
Superheated steam drier accounts for 63%；High-energy discharge failure accounts for 18.1%；It overheats and high-energy discharge failure accounts for 10%；Spark discharge accounts for
7%；Dampness or shelf depreciation account for 1.9%, integrate, maximum with ratio shared by superheating, and about 73%.Group simultaneously
The internal or external insulation protection of forming apparatus is susceptible to the influence of temperature, with temperature continuous raising when, insulation protection
The speed of useful life reduction is continuously increased.It is insulated from that the service life is closely bound up the service life of most electrical equipments, insulation protection
Aging is just easy to that equipment is made to break down.Therefore, the overheating fault for studying power transformation device casing has great practical value.
The intelligent diagnosing method for most intelligent information theories that the country uses transformer state fault diagnosis at present
Including multi-intelligence algorithms such as artificial neural network, support vector machines, fuzzy theory, expert system, thick structure collection theories.With intelligence
Can power grid propulsion, the intelligence degree of equipment is also gradually increased in intelligent substation, and transformer diagnosis method is also from traditional
Method makes the transition to artificial intelligence direction, has been achieved for preferable achievement so far.Become using intelligent algorithm
Depressor state on_line monitoring, fault diagnosis and active forewarning are the major progress of intelligent transformer.It applies and is becoming so far
Intelligent algorithm in depressor fault diagnosis mainly has expert system, support vector machines, fuzzy theory, artificial immunity and nerve
Network scheduling algorithm.
The development of expert system is earliest, its main feature is that can the experience and knowledge of expert enough be received and be located with computer capacity
The mode of reason shows, with expert reasoning method and corresponding control strategy, with asking for the expert grade a certain field of horizontal processing
Topic.Expert system is extremely proficient in process symbol information and reasoning from logic, and research object is knowledge, the quantity and matter of knowledge
Amount is higher, and problem-solving ability is stronger.The main problem that expert system faces in transformer fault diagnosis application exists
In: one, its inference method simple, and reasoning can be relatively weak；Two, it is difficult to obtain more perfect knowledge base；Three, it is difficult to knowing
Knowledge is safeguarded；Four, knowledge step is narrow.
Support vector machines (SVM) is a kind of machine learning method based on statistical theory W and structural risk minimization, extensive
The problems such as ability is stronger, and theoretical basis is good, can solve to small sample, local minimum point and over-fitting.However it is this
Algorithm does not have corresponding standard to choose the parameter of kernel function appropriate and regularization.In addition to this, this method there is also
Classification problem, deviation accumulation are also an insoluble problem.
Fuzzy theory solves very by the processing to that cannot carry out accurate Theory with the fuzzy things that classical theory describes
More insurmountable problems of classical theory, for example, not knowing semantic and fuzzy concept etc. in human brain very much, but its shortcoming exists
In fuzzy theory mainly passes through experience by the processing of picture number " cognition uncertain problem " is subordinate to, and the object studied just has one
A feature, i.e. " intension is clear and extension is uncertain ", however so far, we do not develop what transformer may occur
Phenomenon of the failure and reason etc. " intension ", thus fuzzy theory just has certain limitation.
Artificial immunity is that the new research hotspot of one kind can be complete by imitating the different function of different natural immune systems
At being inspired and then being completed to the learning art of external substance difference defense mechanism by Immune System difference, merged it is many its
The advantages of his algorithm, such as neural network, machine inference, are a new investigative techniques, work as in algorithm construction and engineer application
In, technology is not mature enough.
Artificial neural network (ANN) has following features: one, self-organizing and study by the nervous system in simulation human brain
Ability；Two, parallel processing and distributed storage information；Three, fault-tolerant ability is stronger；Four, there is a degree of generalization ability.
ANN technology is relatively mature, and applying in predicting in transformer online monitoring and fault diagnosis system failure has one
Fixed effect, but problems faced is exactly, and how to determine network structure, network parameter to be solved is more, needs great amount of samples
Network training is carried out, and there may be convergence problems.Although these intelligent algorithms are theoretically all gradually matured, this
A little intelligent diagnosing methods are different because algorithm designs, and the accuracy of diagnosis also has very big difference, and most of researchs also only exist
In the laboratory research stage, there are also to be developed for concrete practice.Therefore the accuracy of algorithm how is improved, and intelligent algorithm is answered
Critical issue in engineering practice, being transformer online monitoring and fault diagnosis system is faced.
Summary of the invention
The present invention is in view of the above problems, propose a kind of transformer safety operation characteristic monitoring based on multisource data fusion
With active forewarning technical research, and mainly study bushing shell for transformer superheating diagnosis and active forewarning, to effectively solve
The intelligent diagnostics of sleeve of main transformer thermal fault and main transformer heating accident the problem of giving warning in advance is pre- in certainly current substation
Alarm method and Fault Locating Method.
The technical solution of the present invention is as follows: carrying out early warning according to the following steps:
A1), casing Temperature Rise Model is established by SVR method:
A1.1), SVR algorithm implementation method is analyzed；
Support vector regression (Support Vector Regression, SVR) method passes through Nonlinear MappingIt will be defeated
Enter sample to be mapped in high-dimensional feature space H, so as to utilize structural risk minimization to establish linear return in the space H
Return function
In formula, y' is predicted value, and w is weight vector, and b is bias.
The realization of SVR is to introduce error threshold ε on the basis of SVM and introduce relaxation factor ξiWithMost using structure
Smallization principle, fitting problems are converted into following optimization problem, i.e. formula (2)
C > 0 is penalty factor in formula.It can be exchanged into following optimization problem using the principle of duality for solving formula (3), see formula
(4)。
Thus the nonlinear solshing of SVR can be acquired
K (x in formulai,xj) meeting the kernel function of Mercer condition, common kernel function includes Radial basis kernel function, i.e.,
k(xi, y) and=exp (- γ | | xi-y||2) (6)
Polynomial kernel function, i.e.,
It is with Sigmoid kernel function
k(xi, y) and=tan (xi Ty+v) (8)
A1.2), the bushing shell for transformer Temperature Rise Model based on SVR is established；
Because casing input/output argument difference is big, during model training and prediction, the normalization of data is to improvement mould
The convergence and generalization ability of type are most important, need to pre-process data, and normalization formula is such as shown in (9):
In formula, x*For the value after normalization, x is sample data, xminFor the minimum value of sample data, xmaxFor sample data
Maximum value；
The input parameter of SVR is the time, is exported as θpor, it is filtered using the SVR of Radial basis kernel function, it is thus necessary to determine that
γ in ε in formula (3), C and formula (5).From (1), formula can be seen that θporBy qcuIt is affected, and qcuWith load current
It is square directly proportional.For the above-mentioned SVR filtering parameter of determination, used herein by load current ILAnd θporSVR filtering is carried out simultaneously, is had
Body is as follows: first to square of load currentSVR filtering is carried out, is guaranteed filteredWithCurve is almost overlapped, and determines ε,
C and γ；Above-mentioned parameter is used for θporSVR filtering, θ filtered in this wayporIt is able to reflect the influence of load current；
A2), the differential equation group of Equivalent heat path model is calculated；
A2.1), draw Equivalent heat path model: using qcuThe dielectric loss for indicating current-carrying conductor in casing, uses qdiIndicate capacitor
The dielectric loss of core, uses RcuIndicate that current-carrying conductor to the non-linear thermal resistance of oil, uses RoilIndicate the oily non-linear thermal resistance to ceramics,
Use RporIndicate that ceramic wall to the non-linear thermal resistance of outside air, uses CcuThe thermal capacitance for indicating current-carrying conductor, uses CoilIndicate the heat of oil
Hold, uses CporThe thermal capacitance for indicating ceramic wall, uses θoilIt indicates oil temperature, uses θporIt indicates ceramic wall temperature, uses θambIndicate environment temperature；
A2.1), simplified differential equation group is obtained according to Equivalent heat path model:
A3), the raised influence factor of bushing temperature, the output variable as model are determined；
By means of infrared measurement of temperature means, with the ceramic wall temperature θ measuredporIt is available by (6) formula for output variable, with
θporThe system of variable is second-order system, θporIt is lost with current-carrying conductor and the factors such as capacitance core dielectric loss, environmental condition is related,
The loss of current-carrying conductor is related with casing current, and dielectric loss is related to voltage；
A4), heating accident early warning；The load current of casing, voltage, environment temperature, ambient humidity and the first two are adopted
The ceramic wall temperature at sample moment inputs casing Temperature Rise Model, thus the ceramic wall temperature at the current time predicted；If exceeding
Defined temperature range, then alarm；It finishes.
2, a kind of Fault Locating Method of main transformer heating accident, which is characterized in that due to above-mentioned casing Temperature Rise Model
The implied terms of input value be that all components are kept normally in casing, therefore, once step A4 conversely speaking) obtain
There is biggish difference in the actually detected ceramic wall temperature arrived of the ceramic wall temperature and infrared measurement of temperature of prediction, then can determine that in casing
There is failure in some component；Therefore, this case additionally provides the localization method to failure；Fault location is carried out according to the following steps:
B1), overall calculation；
Due to containing oil temperature in casing in formula (1), which is unable to measure, and therefore, it is necessary to be converted, enables q=qcu+
qdi(1) formula is converted into standard state equation, sees (10) formula:
It enablesx1=θpor, x2=θoil, x0=θamb, then (10) formula can
Become:
It deforms:
To become controllable type canonical form, w is enabled1=x1, w2=-(c1+c2)x1+c2x2+c1x0, then (12) formula becomes
Enable d1=c1+c2, d2=c1+2c2,Then (13) formula simplifies are as follows:
The first formula derivation of (14) formula is obtained:Arrange (15) formula:
By above-mentioned abbreviation, can obtain containing measurable w1The differential equation, if w can be calculated1Single order and second order
Derivative then can use least squares identification and go out d1、d2And duParameter；
Classical differential uses formulaRealize, if but input signal by noise pollution, derivative is defeated
Noise in out can be amplified, or even flood derivative signal therein.To solve the above problems, practical technology is to use
Differential tracker (TD) technology, it may be assumed that
Wherein fhan is steepest comprehensively control function fhan (x1(k),x2(k), r, h), formula is as follows:
Work as x1(k) when-w is intended to 0,
It is acquired using TDAfterwards, parameter identification is carried out using formula (15) and least square method.Least square method is bent
Common method in line fitting, there are three unknown number d for equation in (15) formula1,d2,du, therefore measure points n and should be greater than being equal to 3.
If n measurement point is w1,Formula (15) should be met, be expressed in matrix as:
Zn=Hnλ (18)
Wherein,λ=[d1,d2,du]T,
λ can be found out by Classical Least-Squares formula:
For dynamically track time-varying elliptic parameter and reduce real-time amount of calculation, the present invention use passing with forgetting factor
Push away least square method.Definition
(17) α is forgetting factor in formula,Then identified parameters recurrence formula are as follows:
B2), d is calculated1、d2And du: according to w1Single order and second dervative, calculate d using least square method1、d2And du；
B3), c is calculated1、c2And x2: due to d1=c1+c2, d2=c1+2c2,x1=θpor, x2
=θoil, x0=θamb, wherein θoilIndicate oil temperature, θporIndicate ceramic wall temperature, θambIndicate environment temperature, therefore known d1、d2、
C can be extrapolated1、c2, it is known that duAnd x0And x0First derivative, can calculate and u；
B4), judge ceramic wall whether failure: due toWherein RporIndicate ceramic wall to outside air
Non-linear thermal resistance, CporThe thermal capacitance for indicating ceramic wall, therefore, if c1There is exception, then can determine that and break down for ceramic wall, by
Operator is with changing ceramic wall；If c1It is without exception, then ceramic wall fault-free, into next step；
B5), judge oil whether failure: due toWherein RoilIndicate oil to the non-linear thermal resistance of ceramics, Cpor
The thermal capacitance for indicating ceramic wall, therefore, if step B5) determine ceramic wall fault-free and c2There is exception, then can determine that and occur for oil
Failure, by operator with changing oil；If step B5) determine ceramic wall fault-free and c2It is without exception, into next step；
B6), judge whether oil temperature exception occurs: according to B3) in the u that extrapolates, by formulaObtain u and fever
Measure q and CoilIt is related, if c1、c2It is without exception, using square directly proportional relationship of q and load current, then determine current-carrying conductor
Or capacitance core breaks down, and carries out integral replacing to casing by operator；It finishes.
The present invention effectively solves the problems, such as in current substation the intelligent diagnostics of sleeve of main transformer thermal fault and gives warning in advance, energy
Planned arrangement operation and maintenance progress, make electric system reach intelligent, automation, are the safety and stability of entire electric system
Operation brings biggish economic and social benefit.
Detailed description of the invention
Fig. 1 is the structural schematic diagram of this case middle sleeve,
Fig. 2 is this case Equivalent heat path model schematic,
Fig. 3 is this middle sleeve Temperature Rise Model schematic diagram,
Fig. 4 isThe current curve schematic diagram of sqrt again after SVR filtering,
Fig. 5 is the temperature curve schematic diagram that the casing ceramic wall of SVR filtering is carried out using the parameter in Fig. 4,
Fig. 6 is that the initial data and training under three kinds of models export result schematic diagram,
Fig. 7 is the training output error schematic diagram of three kinds of models,
Fig. 8 is that the initial data and prediction under three kinds of models export result schematic diagram,
Fig. 9 is the prediction output error schematic diagram of three kinds of models.
Specific embodiment
The present invention as shown in figs 1-9, carries out early warning according to the following steps:
A1), casing Temperature Rise Model is established by SVR method:
A1.1), SVR algorithm implementation method is analyzed；
Support vector regression (Support Vector Regression, SVR) method passes through Nonlinear MappingIt will be defeated
Enter sample to be mapped in high-dimensional feature space H, so as to utilize structural risk minimization to establish linear return in the space H
Return function
In formula, y' is predicted value, and w is weight vector, and b is bias.
The realization of SVR is to introduce error threshold ε on the basis of SVM and introduce relaxation factor ξiWithMost using structure
Smallization principle, fitting problems are converted into following optimization problem, i.e. formula (2)
C > 0 is penalty factor in formula.It can be exchanged into following optimization problem using the principle of duality for solving formula (3), see formula
(4)。
Thus the nonlinear solshing of SVR can be acquired
K (x in formulai,xj) meeting the kernel function of Mercer condition, common kernel function includes Radial basis kernel function, i.e.,
k(xi, y) and=exp (- γ | | xi-y||2) (6)
Polynomial kernel function, i.e.,
It is with Sigmoid kernel function
k(xi, y) and=tan (xi Ty+v) (8)
A1.2), the bushing shell for transformer Temperature Rise Model based on SVR is established；
Because casing input/output argument difference is big, during model training and prediction, the normalization of data is to improvement mould
The convergence and generalization ability of type are most important, need to pre-process data, and normalization formula is such as shown in (9):
In formula, x*For the value after normalization, x is sample data, xminFor the minimum value of sample data, xmaxFor sample data
Maximum value；
The result of infrared measurement of temperature is affected by extraneous factor, and measurement result interference is larger, containing high frequency component noise,
This noise is affected for training and the precision predicted, generallys use low-pass filter for noise filtering, but low-pass filter
Output data generate delay so that filtered output is asynchronous with other data, simultaneously for the less occasion of data volume,
Because filtering transient data is not suitable for the study of SVR, data volume is further reduced, to affect training and study essence
Degree.
This case is filtered using the SVR of Radial basis kernel function, can eliminate above-mentioned unfavorable factor, i.e. the input parameter of SVR
For the time, export as θpor, it is filtered using the SVR of Radial basis kernel function, it is thus necessary to determine that the ε in formula (3), C and formula (5)
In γ.From (1), formula can be seen that θporBy qcuIt is affected, and qcuIt is square directly proportional to load current.It is above-mentioned for determination
SVR filtering parameter is used herein by load current ILAnd θporSVR filtering is carried out simultaneously, it is specific as follows: first to load current
SquareSVR filtering is carried out, is guaranteed filteredWithCurve is almost overlapped, and determines ε, C and γ；Above-mentioned parameter is used for
θporSVR filtering, θ filtered in this wayporIt is able to reflect the influence of load current；Fig. 4 be by test of many times,SVR filter
The current curve of sqrt again after wave, it can be seen that the load current for filtering front and back is almost overlapped, and Fig. 5 is using in Fig. 4
Parameter carries out the temperature curve of the casing ceramic wall of SVR filtering, it is seen that filtered curve ratio raw data plot is smooth；
A2), the differential equation group of Equivalent heat path model is calculated；
A2.1), casing (casing described in this case is) heating accident is generally divided into two kinds of situations: current mode fever event
Barrier, voltage-type heating accident.The current mode heating accident main cause of casing is that mounting process is undesirable or conductor for processing
Contact resistance caused by the factors such as oxidation becomes larger；Casing voltage-type heating accident reason is mainly the conducting rod insulation that casing is drawn
Bad, conducting rod can generate a uneven magnetic field to casing insulator, and voltage's distribiuting exception and Leakage Current increase is caused to be drawn
The heating accident risen；Temperature is excessively high when bushing shell for transformer permeability failure also cashes out casing operation or apparent temperature gradient.
The structure of casing is as shown in Figure 1, internal heat is shed by three kinds of heat transfer, convection current and radiation modes, no
Same radiating mode has the rule and feature of its own.According to analog theory, if the differential equation form one of two kinds of physical phenomenons
It causes, and the boundary condition of two carriers and geometry are close, then both non trivial solution analysis solution and experiment solve number having the same
Form.Thermoelectricity analogy method be exactly based on analog theory, by the similar thermal parameter of mathematical definition expression formula and electrical parameter into
Row analogy analogizes hot Lu Dinglv by circuital law according to the corresponding relationship of parameter, so that the thermal circuit model of inside transformer be turned
It is changed to circuit model.
When transformer station high-voltage side bus, the source for covering tubular heat is mainly the medium damage of capacitance core in current-carrying conductor loss and casing
Consumption, the heat that current-carrying conductor loss generates first pass through conductive form and are transmitted to conductive surface, lead in the oil that surface heat is dispersed into
The realization of convection current form is crossed, heat passes through convection current form again and is transmitted to casing ceramic inner walls in oil；The dielectric loss of capacitance core generates
Heat conducted to its surface, also the form through convection current reaches ceramic inner walls.Since capacitance core is thin and thermal capacitance is small, can ignore
Conductive process, inner wall heat are transmitted to its outer wall through heat transfer, finally all heats in the form of radiating with convection current from fuel tank and
The outer wall of ceramics radiates to surrounding air.Therefore, the flowing of heat is similar to electric current flowing in circuit during being somebody's turn to do, and can draw
Thermal circuit model is imitated, as shown in Figure 2.It draws Equivalent heat path model: using qcuThe dielectric loss for indicating current-carrying conductor in casing, uses qdi
The dielectric loss for indicating capacitance core, uses RcuIndicate that current-carrying conductor to the non-linear thermal resistance of oil, uses RoilIndicate oil to the non-thread of ceramics
Property thermal resistance, uses RporIndicate that ceramic wall to the non-linear thermal resistance of outside air, uses CcuThe thermal capacitance for indicating current-carrying conductor, uses CoilIt indicates
The thermal capacitance of oil, uses CporThe thermal capacitance for indicating ceramic wall, uses θoilIt indicates oil temperature, uses θporIt indicates ceramic wall temperature, uses θambIndicate ring
Border temperature；In addition, it is fast for the temperature rise of the temperature rise ratio oil and ceramic wall of current-carrying conductor, its transient process can be ignored；
A2.1), simplified differential equation group is obtained according to Equivalent heat path model:
A3), the raised influence factor of bushing temperature, the output variable as model are determined；
By means of infrared measurement of temperature means, with the ceramic wall temperature θ measuredporIt is available by (6) formula for output variable, with
θporThe system of variable is second-order system, θporIt is lost with current-carrying conductor and the factors such as capacitance core dielectric loss, environmental condition is related,
The loss of current-carrying conductor is related with casing current, and dielectric loss is related to voltage；
Therefore, load current, voltage, environment temperature, ambient humidity and the pottery of the first two sampling instant of casing are selected
Porcelain wall temperature is the ceramic wall temperature at current time as input variable, output variable, and structure is as shown in Figure 3；
A4), heating accident early warning；The load current of casing, voltage, environment temperature, ambient humidity and the first two are adopted
The ceramic wall temperature (i.e. K-1 moment ceramics wall temperature and K-2 moment ceramics wall temperature) at sample moment inputs casing Temperature Rise Model, from
And the ceramic wall temperature (i.e. K moment ceramics wall temperature) at the current time predicted；If exceeding defined temperature range, into
Row alarm；It finishes.
Choose the voltage, electric current, environment temperature, humidity progress real-time monitoring of a 110kV substation, sampling interval 1
Minute, and monitoring is synchronized to bushing temperature using thermal infrared imager, 236 sample datas are acquired altogether, wherein first half
For model training, latter half is for predicting.With load current, voltage, environment temperature, ambient humidity and the first two moment
Input of six variables of ceramic wall temperature as SVR model, ceramic wall temperature have carried out SVR filtering.More originally to research and propose
SVR temperature model performance, be respectively adopted feed-forward type BP neural network and RBF neural modeling, to identical sample set
It is trained and predicts.SVR model selection RBF kernel function, through parameter optimization repeatedly, the C and γ met the requirements is respectively
4.0 and 0.15.The input layer of BP network, hidden layer, output layer number of nodes be respectively 6,15,1, hidden layer transmission function is selected
Tansig function, output layer transmission function select logsig function, and training function selects Levenberg-Marquardt algorithm,
Learning rate Lr=0.01, target error Eg=0.001.Fig. 6 is that the initial data and training under three kinds of models export as a result, figure
7 be the training output error of three kinds of models, and Fig. 8 is that the initial data and prediction under three kinds of models export as a result, Fig. 9 is three kinds of moulds
The prediction output error of type.The RBF network training error very little it can be seen from upper figure, but predict that error is big, illustrate its extensive energy
Power is poor；BP network and SVR model have a preferable generalization ability, and SVR has smaller error, illustrate this paper model more suitable for
Bushing temperature modeling and temperature prediction.
Due to the implied terms of the input value of above-mentioned casing Temperature Rise Model be in casing all components keep normally, because
This, once step A4 conversely speaking) the obtained ceramic wall temperature of prediction and the actually detected ceramic wall temperature arrived of infrared measurement of temperature
There is biggish difference, then can determine that failure occurs in some component in casing；And the solution that those skilled in the art are readily apparent that
Certainly measure is integrally replaced to casing, not only time-consuming and laborious, and is also had equipment cost high and need to integrally be shut down dimension
The disadvantages of shield；Therefore, this case additionally provides the localization method to failure；Fault location is carried out according to the following steps:
B1), overall calculation；
Due to containing oil temperature in casing in formula (1), which is unable to measure, and therefore, it is necessary to be converted, enables q=qcu+
qdi(1) formula is converted into standard state equation, sees (10) formula:
It enablesx1=θpor, x2=θoil, x0=θamb, then (10) formula can
Become:
It deforms:
To become controllable type canonical form, w is enabled1=x1, w2=-(c1+c2)x1+c2x2+c1x0, then (12) formula becomes
Enable d1=c1+c2, d2=c1+2c2,Then (13) formula simplifies are as follows:
The first formula derivation of (14) formula is obtained:Arrange (15) formula:
By above-mentioned abbreviation, can obtain containing measurable w1The differential equation, if w can be calculated1Single order and second order
Derivative then can use least squares identification and go out d1、d2And duParameter；
Classical differential uses formulaRealize, if but input signal by noise pollution, derivative is defeated
Noise in out can be amplified, or even flood derivative signal therein.To solve the above problems, practical technology is to use
Differential tracker (TD) technology, it may be assumed that
Wherein fhan is steepest comprehensively control function fhan (x1(k),x2(k), r, h), formula is as follows:
Work as x1(k) when-w is intended to 0,
It is acquired using TDAfterwards, parameter identification is carried out using formula (15) and least square method.Least square method is bent
Common method in line fitting, there are three unknown number d for equation in (15) formula1,d2,du, therefore measure points n and should be greater than being equal to 3.
If n measurement point is w1,Formula (15) should be met, be expressed in matrix as:
Zn=Hnλ (18)
Wherein,λ=[d1,d2,du]T,
λ can be found out by Classical Least-Squares formula:
For dynamically track time-varying elliptic parameter and reduce real-time amount of calculation, the present invention use passing with forgetting factor
Push away least square method.Definition
(17) α is forgetting factor in formula,Then identified parameters recurrence formula are as follows:
B2), d is calculated1、d2And du: according to w1Single order and second dervative, calculate d using least square method1、d2And du；
B3), c is calculated1、c2And x2: due to d1=c1+c2, d2=c1+2c2,x1=θpor, x2
=θoil, x0=θamb, wherein θoilIndicate oil temperature, θporIndicate ceramic wall temperature, θambIndicate environment temperature, therefore known d1、d2、
C can be extrapolated1、c2, it is known that duAnd x0And x0First derivative, can calculate and u；
B4), judge ceramic wall whether failure: due toWherein RporIndicate ceramic wall to outside air
Non-linear thermal resistance, CporThe thermal capacitance for indicating ceramic wall, therefore, if c1There is exception, then can determine that and break down for ceramic wall, by
Operator is with changing ceramic wall；If c1It is without exception, then ceramic wall fault-free, into next step；
B5), judge oil whether failure: due toWherein RoilIndicate oil to the non-linear thermal resistance of ceramics, Cpor
The thermal capacitance for indicating ceramic wall, therefore, if step B5) determine ceramic wall fault-free and c2There is exception, then can determine that and occur for oil
Failure, by operator with changing oil；If step B5) determine ceramic wall fault-free and c2It is without exception, into next step；
B6), judge whether oil temperature exception occurs: according to B3) in the u that extrapolates, by formulaObtain u and fever
Measure q and CoilIt is related, if c1、c2It is without exception, using square directly proportional relationship of q and load current, then determine current-carrying conductor
Or capacitance core breaks down, and carries out integral replacing to casing by operator；It finishes.
This case is only being capable of measuring ceramic wall temperature, air themperature, can not but obtained by calculating, derivation and exclusion step by step
Under the premise of taking inside pipe casing details, final effective position goes out the abort situation of casing, so that operator can be convenient
Casing is repaired or integral replacing, to effectively solve the intelligent diagnostics of sleeve of main transformer thermal fault in current substation
Problem has many advantages, such as that step is clear, fault location is accurate.
Key technology of the invention
1) data processing and filtering technique of infrared temperature-test technology
The result of infrared measurement of temperature is affected by extraneous factor, and measurement result interference is larger, containing high frequency component noise,
This noise is affected for training and the precision predicted, generallys use low-pass filter for noise filtering, but low-pass filter
Output data generate delay so that filtered output is asynchronous with other data, simultaneously for the less occasion of data volume,
Because filtering transient data is not suitable for the study of SVR, data volume is further reduced, to affect training and study essence
Degree.This project is filtered using the SVR of Radial basis kernel function, can eliminate above-mentioned unfavorable factor, for the above-mentioned SVR filtering ginseng of determination
Number, this project, which is used, carries out SVR filtering simultaneously for load current and sleeve surface temperature, sleeve surface temperature filtered in this way
It is able to reflect the influence of load current.
2) the SVR modeling of transformer temperature rise
Using load current, voltage, environment temperature, six variables of ambient humidity and the ceramic wall temperature at the first two moment as
The input of SVR model, ceramic wall temperature have carried out SVR filtering.And optimizing is carried out to the parameter in SVR using genetic algorithm.
3) it is suitable for the transformer Temperature Rise Model modeling of parameter identification
It is not suitable for parameter identification according to the obtained mathematical model of heat-transfer mechanism, and the parameter in Temperature Rise Model is for becoming
The fault diagnosis of depressor casing has important directive significance, and this project, which is deduced, is suitable for the mathematical modulo based on infrared measurement of temperature
Type is obtained containing measurable differential equation, recognizable Temperature Rise Model parameter out.
4) the derivative calculations technology of the infrared temperature measured
Using the derivative of classical differential calculation, the noise in derivative output can be amplified, or even flood derivative letter therein
Number.It is to solve the above problem using differential tracker (TD) technology that this project, which uses,.
Key difficulties of the invention
1) parameter selection of SVR filtering and temperature rise modeling
The more difficult determination of allowable error and weight parameter in SVR technology, this project carry out parameter optimization using genetic algorithm,
It is filtered constantly in SVR, filtering parameter is associated with load current, preferably solves the above problem；
2) it is suitable for the foundation of the model of parameter identification
It is not suitable for parameter identification according to the obtained mathematical model of heat-transfer mechanism, derivation is suitable for based on infrared measurement of temperature
Mathematical model is obtained containing measurable differential equation, recognizable Temperature Rise Model parameter out.
3) calculating of derivative when parameter identification
The calculating of derivative is easy affected by noise, and the infrared temperature data measured itself contains biggish noise, conventional
Derivative technique be unable to reach practicability.
Claims (2)
1. a kind of method for early warning of main transformer heating accident, which is characterized in that carry out early warning according to the following steps:
A1), casing Temperature Rise Model is established by SVR method:
A1.1), SVR algorithm implementation method is analyzed；
Support vector regression (Support Vector Regression, SVR) method passes through Nonlinear MappingSample will be inputted
Originally it is mapped in high-dimensional feature space H, so as to establish linear regression letter using structural risk minimization in the space H
Number
In formula, y' is predicted value, and w is weight vector, and b is bias.
The realization of SVR is to introduce error threshold ε on the basis of SVM and introduce relaxation factor ξiWithIt is minimized using structure former
Then, fitting problems are converted into following optimization problem, i.e. formula (2)
C > 0 is penalty factor in formula.It can be exchanged into following optimization problem using the principle of duality to solve formula (3), see formula (4).
Thus the nonlinear solshing of SVR can be acquired
K (x in formulai,xj) meeting the kernel function of Mercer condition, common kernel function includes Radial basis kernel function, i.e.,
k(xi, y) and=exp (- γ | | xi-y||2) (6)
Polynomial kernel function, i.e.,
It is with Sigmoid kernel function
k(xi, y) and=tan (xi Ty+v) (8)
A1.2), the bushing shell for transformer Temperature Rise Model based on SVR is established；
Because casing input/output argument difference is big, during model training and prediction, the normalization of data is to improved model
Convergence and generalization ability are most important, need to pre-process data, and normalization formula is such as shown in (9):
In formula, x*For the value after normalization, x is sample data, xminFor the minimum value of sample data, xmaxMost for sample data
Big value；
The input parameter of SVR is the time, is exported as θpor, it is filtered using the SVR of Radial basis kernel function, it is thus necessary to determine that formula (3)
In ε, the γ in C and formula (5).From (1), formula can be seen that θporBy qcuIt is affected, and qcuWith square of load current
It is directly proportional.For the above-mentioned SVR filtering parameter of determination, used herein by load current ILAnd θporSVR filtering is carried out simultaneously, specifically such as
Under: first to square of load currentSVR filtering is carried out, is guaranteed filteredWithCurve is almost overlapped, and determines ε, C and
γ；Above-mentioned parameter is used for θporSVR filtering, θ filtered in this wayporIt is able to reflect the influence of load current；
A2), the differential equation group of Equivalent heat path model is calculated；
A2.1), draw Equivalent heat path model: using qcuThe dielectric loss for indicating current-carrying conductor in casing, uses qdiIndicate capacitance core
Dielectric loss uses RcuIndicate that current-carrying conductor to the non-linear thermal resistance of oil, uses RoilIndicate that oil to the non-linear thermal resistance of ceramics, uses Rpor
Indicate that ceramic wall to the non-linear thermal resistance of outside air, uses CcuThe thermal capacitance for indicating current-carrying conductor, uses CoilIt indicates the thermal capacitance of oil, uses
CporThe thermal capacitance for indicating ceramic wall, uses θoilIt indicates oil temperature, uses θporIt indicates ceramic wall temperature, uses θambIndicate environment temperature；
A2.1), simplified differential equation group is obtained according to Equivalent heat path model:
A3), the raised influence factor of bushing temperature, the output variable as model are determined；
By means of infrared measurement of temperature means, with the ceramic wall temperature θ measuredporIt is available by (6) formula for output variable, with θporBecome
The system of amount is second-order system, θporIt is lost with current-carrying conductor and the factors such as capacitance core dielectric loss, environmental condition is related, current-carrying
The loss of conductor is related with casing current, and dielectric loss is related to voltage；
A4), heating accident early warning；When by the sampling of the load current of casing, voltage, environment temperature, ambient humidity and the first two
The ceramic wall temperature at quarter inputs casing Temperature Rise Model, thus the ceramic wall temperature at the current time predicted；If beyond regulation
Temperature range, then alarm；It finishes.
2. a kind of Fault Locating Method of main transformer heating accident, which is characterized in that defeated due to above-mentioned casing Temperature Rise Model
The implied terms for entering value is that all components are kept normally in casing, therefore, once step A4 conversely speaking) obtained prediction
Ceramic wall temperature and the actually detected ceramic wall temperature that arrives of infrared measurement of temperature there is biggish difference, then can determine that some in casing
There is failure in component；Therefore, this case additionally provides the localization method to failure；Fault location is carried out according to the following steps:
B1), overall calculation；
Due to containing oil temperature in casing in formula (1), which is unable to measure, and therefore, it is necessary to be converted, enables q=qcu+qdiIt will
(1) formula is converted to standard state equation, sees (10) formula:
It enablesx1=θpor, x2=θoil, x0=θamb, then (10) formula is variable are as follows:
It deforms:
To become controllable type canonical form, w is enabled1=x1, w2=-(c1+c2)x1+c2x2+c1x0, then (12) formula becomes
Enable d1=c1+c2, d2=c1+2c2,Then (13) formula simplifies are as follows:
The first formula derivation of (14) formula is obtained:Arrange (15) formula:
By above-mentioned abbreviation, can obtain containing measurable w1The differential equation, if w can be calculated1Single order and second order lead
Number, then can use least squares identification and go out d1、d2And duParameter；
Classical differential uses formulaRealize, if but input signal by noise pollution, in derivative output
Noise can be amplified, or even flood derivative signal therein.To solve the above problems, practical technology is using differential
Tracker (TD) technology, it may be assumed that
Wherein fhan is steepest comprehensively control function fhan (x1(k),x2(k), r, h), formula is as follows:
Work as x1(k) when-w is intended to 0,
It is acquired using TDAfterwards, parameter identification is carried out using formula (15) and least square method.Least square method is that curve is quasi-
Common method in conjunction, there are three unknown number d for equation in (15) formula1,d2,du, therefore measure points n and should be greater than being equal to 3.If n
Measurement point is w1,Formula (15) should be met, be expressed in matrix as:
Zn=Hnλ (18)
Wherein,λ=[d1,d2,du]T,
λ can be found out by Classical Least-Squares formula:
For dynamically track time-varying elliptic parameter and reduce real-time amount of calculation, the present invention using the recursion with forgetting factor most
Small square law.Definition
(17) α is forgetting factor in formula,Then identified parameters recurrence formula are as follows:
B2), d is calculated1、d2And du: according to w1Single order and second dervative, calculate d using least square method1、d2And du；
B3), c is calculated1、c2And x2: due to d1=c1+c2, d2=c1+2c2,x1=θpor, x2=θoil,
x0=θamb, wherein θoilIndicate oil temperature, θporIndicate ceramic wall temperature, θambIndicate environment temperature, therefore known d1、d2, can calculate
C out1、c2, it is known that duAnd x0And x0First derivative, can calculate and u；
B4), judge ceramic wall whether failure: due toWherein RporIndicate ceramic wall to the non-linear of outside air
Thermal resistance, CporThe thermal capacitance for indicating ceramic wall, therefore, if c1There is exception, then can determine that and break down for ceramic wall, by operator
Member is with changing ceramic wall；If c1It is without exception, then ceramic wall fault-free, into next step；
B5), judge oil whether failure: due toWherein RoilIndicate oil to the non-linear thermal resistance of ceramics, CporIt indicates
The thermal capacitance of ceramic wall, therefore, if step B5) determine ceramic wall fault-free and c2There is exception, then can determine that and break down for oil,
By operator with changing oil；If step B5) determine ceramic wall fault-free and c2It is without exception, into next step；
B6), judge whether oil temperature exception occurs: according to B3) in the u that extrapolates, by formulaObtain u and calorific value q and
CoilIt is related, if c1、c2It is without exception, using square directly proportional relationship of q and load current, then determine current-carrying conductor or capacitor
Core breaks down, and carries out integral replacing to casing by operator；It finishes.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910236008.0A CN109871660B (en) | 2019-03-26 | 2019-03-26 | Early warning method and fault positioning method for heating fault of main transformer |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910236008.0A CN109871660B (en) | 2019-03-26 | 2019-03-26 | Early warning method and fault positioning method for heating fault of main transformer |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109871660A true CN109871660A (en) | 2019-06-11 |
| CN109871660B CN109871660B (en) | 2020-09-18 |
Family
ID=66921366
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910236008.0A Active CN109871660B (en) | 2019-03-26 | 2019-03-26 | Early warning method and fault positioning method for heating fault of main transformer |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109871660B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110930075A (en) * | 2019-12-16 | 2020-03-27 | 云南电网有限责任公司信息中心 | Power equipment fault positioning method and device, computer equipment and storage medium |
| CN112002524A (en) * | 2020-08-28 | 2020-11-27 | 华北电力大学（保定） | Oil-immersed transformer bushing |
| CN112036436A (en) * | 2020-07-23 | 2020-12-04 | 国网江苏省电力有限公司检修分公司 | Data noise processing method and processing system of phase modulator oil temperature prediction system |
| CN112231982A (en) * | 2020-10-13 | 2021-01-15 | 宁波大学 | Photovoltaic panel fault detection method based on distributed soft measurement model |
| CN113237506A (en) * | 2021-03-31 | 2021-08-10 | 国网四川省电力公司电力科学研究院 | Early warning method and system for high-voltage bushing tap device and storage medium |
| CN116108760A (en) * | 2023-04-12 | 2023-05-12 | 广东电网有限责任公司佛山供电局 | Main heating characteristic monitoring method and system |
| CN117407679A (en) * | 2023-12-15 | 2024-01-16 | 南京中鑫智电科技有限公司 | Data acquisition method and system of intelligent end screen sensor |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2007317836A (en) * | 2006-05-25 | 2007-12-06 | Tokyo Electric Power Co Inc:The | Diagnosis method of oil-filled transformer |
| CN105203210A (en) * | 2015-10-23 | 2015-12-30 | 国网山西省电力公司大同供电公司 | Fault detection device and detection method of ultra-high voltage (UHV) substation transformer based on 360-degree infrared panoramic view and support vector machine (SVM) |
| CN106055888A (en) * | 2016-05-27 | 2016-10-26 | 国网上海市电力公司 | Predication method and device for top-oil temperature of transformer based on error predicting amendment |
| CN108197396A (en) * | 2018-01-09 | 2018-06-22 | 国网福建省电力有限公司 | A kind of high voltage isolator superheat state Forecasting Methodology based on PSO-SVM |
-
2019
- 2019-03-26 CN CN201910236008.0A patent/CN109871660B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2007317836A (en) * | 2006-05-25 | 2007-12-06 | Tokyo Electric Power Co Inc:The | Diagnosis method of oil-filled transformer |
| CN105203210A (en) * | 2015-10-23 | 2015-12-30 | 国网山西省电力公司大同供电公司 | Fault detection device and detection method of ultra-high voltage (UHV) substation transformer based on 360-degree infrared panoramic view and support vector machine (SVM) |
| CN106055888A (en) * | 2016-05-27 | 2016-10-26 | 国网上海市电力公司 | Predication method and device for top-oil temperature of transformer based on error predicting amendment |
| CN108197396A (en) * | 2018-01-09 | 2018-06-22 | 国网福建省电力有限公司 | A kind of high voltage isolator superheat state Forecasting Methodology based on PSO-SVM |
Non-Patent Citations (1)
| Title |
|---|
| 滕黎: "油浸式变压器热路计算及热点温度遗传支持", 《万方论文数据库》 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110930075A (en) * | 2019-12-16 | 2020-03-27 | 云南电网有限责任公司信息中心 | Power equipment fault positioning method and device, computer equipment and storage medium |
| CN110930075B (en) * | 2019-12-16 | 2023-05-26 | 云南电网有限责任公司信息中心 | Power equipment fault positioning method, device, computer equipment and storage medium |
| CN112036436A (en) * | 2020-07-23 | 2020-12-04 | 国网江苏省电力有限公司检修分公司 | Data noise processing method and processing system of phase modulator oil temperature prediction system |
| CN112036436B (en) * | 2020-07-23 | 2024-02-13 | 国网江苏省电力有限公司检修分公司 | Data noise processing method and processing system of camera oil temperature prediction system |
| CN112002524A (en) * | 2020-08-28 | 2020-11-27 | 华北电力大学（保定） | Oil-immersed transformer bushing |
| CN112231982A (en) * | 2020-10-13 | 2021-01-15 | 宁波大学 | Photovoltaic panel fault detection method based on distributed soft measurement model |
| CN112231982B (en) * | 2020-10-13 | 2024-02-02 | 广东光美能源科技有限公司 | Photovoltaic panel fault detection method based on distributed soft measurement model |
| CN113237506A (en) * | 2021-03-31 | 2021-08-10 | 国网四川省电力公司电力科学研究院 | Early warning method and system for high-voltage bushing tap device and storage medium |
| CN116108760A (en) * | 2023-04-12 | 2023-05-12 | 广东电网有限责任公司佛山供电局 | Main heating characteristic monitoring method and system |
| CN117407679A (en) * | 2023-12-15 | 2024-01-16 | 南京中鑫智电科技有限公司 | Data acquisition method and system of intelligent end screen sensor |
| CN117407679B (en) * | 2023-12-15 | 2024-03-12 | 南京中鑫智电科技有限公司 | Data acquisition method and system of intelligent end screen sensor |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109871660B (en) | 2020-09-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109871660A (en) | A kind of method for early warning and Fault Locating Method of main transformer heating accident | |
| CN106228033B (en) | Three-core cable conductor temperature real-time computing technique based on RBF neural | |
| Ruan et al. | Inversion detection of transformer transient hot spot temperature | |
| Xu et al. | An unknown fault identification method based on PSO-SVDD in the IoT environment | |
| Ke et al. | Compound fault diagnosis method of modular multilevel converter based on improved capsule network | |
| Hell et al. | Participatory learning in power transformers thermal modeling | |
| CN106599384A (en) | Cable transient temperature rise acquisition method adapting to two boundary temperatures | |
| Liu et al. | A comparative research on power transformer fault diagnosis based on several artificial neural networks | |
| Huang et al. | Research on the online measuring technology of inter‐turn insulation of dry air‐core reactor | |
| Vakil-Baghmisheh et al. | Dynamic voltage stability assessment of power transmission systems using neural networks | |
| Shi et al. | A fault location method for distribution system based on one-dimensional convolutional neural network | |
| Zhan et al. | A robustness temperature inversion method for cable straight joints based on improved sparrow search algorithm optimized BPNN | |
| Yahui et al. | The gas regulator fault diagnosis based on PCA-RBF neural network | |
| Wang et al. | Deep domain-invariant long short-term memory network for partial discharge localization in gas-insulated switchgear | |
| Huang et al. | The system of temperature rise monitoring and temperature prediction for power equipment | |
| Liu | Intelligent detection and diagnosis of power failure relying on BP neural network algorithm | |
| Wu et al. | Identification and correction of abnormal measurement data in power system based on graph convolutional network and gated recurrent unit | |
| Laib et al. | Localisation of faults in wiring networks using time domain reflectometry and adaptive neuro‐fuzzy inference system | |
| Fan | Electrical control online monitoring system based on internet of things | |
| Zhang et al. | Neural Network Based on Health Monitoring Electrical Equipment Fault and Biomedical Diagnosis | |
| Yang et al. | A method for calculating hot-spot temperature of traction transformer based on multi-physical model and combined neural network | |
| Luo et al. | A Method for Hot Spot Temperature Monitoring of Oil-Immersed Transformers Combining Physical Simulation and Intelligent Neural Network | |
| Dan et al. | Influence analysis of calculated horizontally layered soil parameters on grounding parameters | |
| Hu et al. | An online deep reinforcement learning based parameter identification method for HVDC system | |
| Li et al. | Application of multi-sensor information fusion technology in the power transformer fault diagnosis |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |