CN106055888A - Predication method and device for top-oil temperature of transformer based on error predicting amendment - Google Patents

Abstract

The invention discloses a predication method and device for top-oil temperature of a transformer based on error predicting amendment, pertaining to the on-line monitoring field of the transformer. The method comprises following steps: obtaining a prediction dataset comprising load current and ambient environment of the transformer; using load current of the transformer at the current moment as input amount, utilizing a Susa thermal circuit model to predict top-oil temperature of the transformer in order to obtain top-oil temperature prediction value of the Susa thermal circuit model; normalizing top-oil temperature prediction value of the Susa thermal circuit model and the prediction dataset; using the normalized top-oil temperature prediction value of the Susa thermal circuit model and the prediction dataset as input amount and utilizing a GA-KEDLM model to make a regression forecast to obtain prediction errors; carry out anti-normalization operation on regression prediction errors; using prediction value of top-oil temperature of the Susa thermal circuit model before normalization to delete the regression prediction errors after anti-normalization operation in order to obtain top-oil temperature prediction value corrected by the transformer. The predication method and device for top-oil temperature of the transformer based on error predicting amendment are capable of accurately predicting top-oil temperature of the transformer.

Description

Transformator top-oil temperature Forecasting Methodology based on predicted error amendment and device

Technical field

The present invention relates to transformer online monitoring field, particularly relate to a kind of transformator top layer based on predicted error amendment Oil temperature Forecasting Methodology and device.

Background technology

Power transformer dynamic load ability, insulation ag(e)ing rate depend primarily on its thermal characteristics.Top-oil temperature is to weigh The important indicator of transformator thermal characteristics, is also one of important monitoring variable in transformator running, accurately and reliably predicts top Layer oil temperature arranges transformator dynamic load, prevention transformator hot stall important in inhibiting to reasonably instructing.

Forecasting Methodology currently for transformator top-oil temperature is more, including two quasi-representative models, based on thermal conduction study Half physical model (such as Susa thermal circuit model etc.) and based on nonlinear fitting return mathematical model.There is model excessively in the former Simplify, parameter calculate inaccurate, by the big problem of such environmental effects, there is systematic error in model；There is physics in the latter's model Equivocal problem, generalization waits research.

Susa thermal circuit model is a kind of typical based on the half physical model of thermoelectricity analogy principle in thermal conduction study, it is contemplated that oil Viscosity, on thermal resistance and the impact of oil time constant, has clear and definite physical significance, but specific aim is general.

Summary of the invention

The present invention provides a kind of transformator top-oil temperature Forecasting Methodology based on predicted error amendment and device, energy of the present invention Enough prediction transformator top-oil temperatures accurately.

For solving above-mentioned technical problem, the present invention provides technical scheme as follows:

On the one hand, it is provided that a kind of transformator top-oil temperature Forecasting Methodology based on predicted error amendment, including:

Step 101: obtaining predictive data set, described predictive data set includes load current and the ring of current time transformator Border temperature, and load current, ambient temperature and the top-oil temperature of several moment transformators before current time；

Step 102: with the load current of current time transformator as input quantity, utilize Susa thermal circuit model to transformator Top-oil temperature is predicted, and obtains Susa thermal circuit model top-oil temperature predictive value

Step 103: by described Susa thermal circuit model top-oil temperature predictive valueIt is normalized place with predictive data set Reason；

Step 104: with the Susa thermal circuit model top-oil temperature predictive value after normalizationIt is input with predictive data set Amount, utilizes GA-KELM model to carry out regression forecasting, obtains regression forecasting error

Step 105: to described regression forecasting errorCarry out renormalization process；

Step 106: with the Susa thermal circuit model top-oil temperature predictive value before normalizationDeduct the recurrence after renormalization Forecast errorObtain transformator revised top-oil temperature predictive value.

Further, described GA-KELM model obtains by the following method:

Step 201: obtain the training sample set of described transformator, described training sample set includes multiple training sample, its In, i-th training sample includes T_iTime inscribe the load current of transformator, ambient temperature and top-oil temperature, i ∈ [1, N], N are Training sample number；

Step 202: to i-th training sample, with T_iTime inscribe the load current of transformator and top-oil temperature is input quantity, Utilize Susa thermal circuit model that top-oil temperature is predicted, obtain Susa thermal circuit model top-oil temperature predictive valueWith Susa heat Road model predictive error

Step 203: by described Susa thermal circuit model top-oil temperature predictive valueWith Susa thermal circuit model forecast error It is normalized；

Step 204: set up KELM model；

Step 205: use the Susa thermal circuit model top-oil temperature predictive value of each training sampleSusa thermal circuit model Forecast errorIt is trained with KELM model described in training sample set pair, obtains GA-KELM model.

Further, in described step 102 and step 202, described Susa thermal circuit model top-oil temperature predictive valuePass through Equation below is calculated:

$\left\{\begin{array}{c}\frac{K^2\mathrm{\α}+1}{\mathrm{\α}+1}{\left({\mathrm{\μ}}_{pu}\right)}^{\frac{1-n}{n}}{\mathrm{\Δ\θ}}_{oil,R}={\left({\mathrm{\μ}}_{pu}\right)}^{\frac{1-n}{n}}{\mathrm{\τ}}_{oil,R}\frac{{\mathrm{d\θ}}_{oil}^{e,b}}{dt}+\frac{{({\mathrm{\θ}}_{oil}^{e,b}-{\mathrm{\θ}}_{amb})}^{\frac{1}{n}}}{{\left({\mathrm{\Δ\θ}}_{oil,R}\right)}^{\frac{1-n}{n}}}\\ {\mathrm{\μ}}_{pu}=\frac{\mathrm{\μ}}{{\mathrm{\μ}}_R}=e^{\frac{2797.3}{{\mathrm{\θ}}_{oil}^{e,b}+273}-\frac{2797.3}{{\mathrm{\θ}}_{oil,R}+273}}\end{array}\right.$

Wherein, K is load factor, for the ratio of load current with rated current；α be nominal load loss with open circuit loss it Ratio；Δθ_oil,RFor under nominal load, the stable state temperature rise of top-oil temperature versus environmental；N is the experience of reflection transformator radiating mode Index；τ_oil,RFor specified top layer oil time constant, for thermal capacitance and the product of thermal resistance under nominal load；μ_puFor oil viscosity change because of Son, for the oil viscosity μ under arbitrary temp and the oil viscosity μ under specified top-oil temperature_RRatio；

In described step 202, described Susa thermal circuit model forecast errorIt is calculated by equation below:

${\mathrm{\Δ\θ}}_{oil}^T={\mathrm{\θ}}_{oil}^{e,b}-{\mathrm{\θ}}_{oil}^m$

Wherein,For T in training sample_iThe top-oil temperature of moment transformator.

Further, described step 205 includes:

Step 2051: by input vector x_iInput KELM model, wherein:

$x_i=\[I\left(i\right),I(i-1),I(i-2),{\mathrm{\θ}}_{amb}\left(i\right),{\mathrm{\θ}}_{amb}(i-1),{\mathrm{\θ}}_{amb}(i-2),{\mathrm{\θ}}_{oil}^{e,b}\left(i\right),{\mathrm{\θ}}_{oil}^m(i-1),{\mathrm{\θ}}_{oil}^m(i-2)\]$

I (i) and θ_ambI () is respectively i-th training sample T_iThe load current in moment and ambient temperature, I (i-1), θ_amb (i-1) andIt is respectively the i-th-1 training sample T_i-1The load current in moment, ambient temperature and top-oil temperature, I (i- 2)、θ_amb(i-2) andIt is respectively the i-th-2 training sample T_i-2The load current in moment, ambient temperature and top layer oil Temperature；

Step 2052: obtain x_iCorresponding regression forecasting errorWith output weight beta, wherein:

$\left\{\begin{array}{c}\mathrm{\Δ}{\mathrm{\θ}}_{oil}^e\left(x\right)={\left[\begin{array}{c}K(x,x_1)\\ \·\\ \·\\ \·\\ K(x,x_N)\end{array}\right]}^T\mathrm{\β}\\ \mathrm{\β}={(}^I\left[\begin{array}{c}{\mathrm{\Δ\θ}}_{oil}^T\left(x_1\right)\\ \·\\ \·\\ \·\\ {\mathrm{\Δ\θ}}_{oil}^T\left(x_N\right)\end{array}\right]\end{array}\right.$

K(x_i,x_j) it is kernel function；

Step 2053: use GA that nuclear parameter γ and the penalty coefficient C of KELM model are optimized so thatMean square Error MSE_ΔθMinimum, obtains GA-KELM model, wherein:

${\mathrm{MSE}}_{\mathrm{\Δ}\mathrm{\θ}}=\frac{1}{N}\underset{i=1}{\overset{N}{\Σ}}{({\mathrm{\Δ\θ}}_{oil}^T-{\mathrm{\Δ\θ}}_{oil}^e)}^2.$

Further, described step 201 includes:

Step 2011: obtain T_iTime inscribe the load current of transformator, ambient temperature and top-oil temperature；

Step 2012: reject the data substantially lost efficacy, and replace fail data by interpolated value, obtain training sample；

Step 2013: multiple training samples are formed training sample set.

On the other hand, it is provided that a kind of transformator top-oil temperature prediction means based on predicted error amendment, including:

Predictive data set acquisition module, is used for obtaining predictive data set, and described predictive data set includes current time transformation Load current, ambient temperature and the top layer of several moment transformators before the load current of device and ambient temperature, and current time Oil temperature；

Top-oil temperature prediction module, for the load current of current time transformator as input quantity, utilizing the hot road of Susa The top-oil temperature of transformator is predicted by model, obtains Susa thermal circuit model top-oil temperature predictive value

First normalization module, for by described Susa thermal circuit model top-oil temperature predictive valueEnter with predictive data set Row normalized；

Regression forecasting module, for the Susa thermal circuit model top-oil temperature predictive value after normalizationAnd prediction data Integrate as input quantity, utilize GA-KELM model to carry out regression forecasting, obtain regression forecasting error

Renormalization module, for described regression forecasting errorCarry out renormalization process；

Correcting module, the Susa thermal circuit model top-oil temperature predictive value before using normalizationAfter deducting renormalization Regression forecasting errorObtain transformator revised top-oil temperature predictive value.

Further, described GA-KELM model is obtained by such as lower module:

Training sample set acquisition module, for obtaining the training sample set of described transformator, described training sample set includes Multiple training samples, wherein, i-th training sample includes T_iTime inscribe the load current of transformator, ambient temperature and top layer oil Temperature, i ∈ [1, N], N are training sample number；

Top-oil temperature and error prediction module, for i-th training sample, with T_iTime inscribe the load current of transformator It is input quantity with top-oil temperature, utilizes Susa thermal circuit model that top-oil temperature is predicted, obtain Susa thermal circuit model top layer oil Temperature predictive valueWith Susa thermal circuit model forecast error

Second normalization module, for by described Susa thermal circuit model top-oil temperature predictive valueWith Susa thermal circuit model Forecast errorIt is normalized；

KELM model building module, is used for setting up KELM model；

Training module, for using the Susa thermal circuit model top-oil temperature predictive value of each training sampleThe hot road of Susa Model predictive errorIt is trained with KELM model described in training sample set pair, obtains GA-KELM model.

Further, in described top-oil temperature prediction module and top-oil temperature and error prediction module, described Susa heat Road model top-oil temperature predictive valueIt is calculated by equation below:

$\left\{\begin{array}{c}\frac{K^2\mathrm{\α}+1}{\mathrm{\α}+1}{\left({\mathrm{\μ}}_{pu}\right)}^{\frac{1-n}{n}}{\mathrm{\Δ\θ}}_{oil,R}={\left({\mathrm{\μ}}_{pu}\right)}^{\frac{1-n}{n}}{\mathrm{\τ}}_{oil,R}\frac{{\mathrm{d\θ}}_{oil}^{e,b}}{dt}+\frac{{({\mathrm{\θ}}_{oil}^{e,b}-{\mathrm{\θ}}_{amb})}^{\frac{1}{n}}}{{\left({\mathrm{\Δ\θ}}_{oil,R}\right)}^{\frac{1-n}{n}}}\\ {\mathrm{\μ}}_{pu}=\frac{\mathrm{\μ}}{{\mathrm{\μ}}_R}=e^{\frac{2797.3}{{\mathrm{\θ}}_{oil}^{e,b}+273}-\frac{2797.3}{{\mathrm{\θ}}_{oil,R}+273}}\end{array}\right.$

Wherein, K is load factor, for the ratio of load current with rated current；α be nominal load loss with open circuit loss it Ratio；Δθ_oil,RFor under nominal load, the stable state temperature rise of top-oil temperature versus environmental；N is the experience of reflection transformator radiating mode Index；τ_oil,RFor specified top layer oil time constant, for thermal capacitance and the product of thermal resistance under nominal load；μ_puFor oil viscosity change because of Son, for the oil viscosity μ under arbitrary temp and the oil viscosity μ under specified top-oil temperature_RRatio；

In described top-oil temperature and error prediction module, described Susa thermal circuit model forecast errorPass through equation below It is calculated:

${\mathrm{\Δ\θ}}_{oil}^T={\mathrm{\θ}}_{oil}^{e,b}-{\mathrm{\θ}}_{oil}^m$

Wherein,For T in training sample_iThe top-oil temperature of moment transformator.

Further, described training module includes:

Input block, for by input vector x_iInput KELM model, wherein:

$x_i=\[I\left(i\right),I(i-1),I(i-2),{\mathrm{\θ}}_{amb}\left(i\right),{\mathrm{\θ}}_{amb}(i-1),{\mathrm{\θ}}_{amb}(i-2),{\mathrm{\θ}}_{oil}^{e,b}\left(i\right),{\mathrm{\θ}}_{oil}^m(i-1),{\mathrm{\θ}}_{oil}^m(i-2)\]$

I (i) and θ_ambI () is respectively i-th training sample T_iThe load current in moment and ambient temperature, I (i-1), θ_amb (i-1) andIt is respectively the i-th-1 training sample T_i-1The load current in moment, ambient temperature and top-oil temperature, I (i- 2)、θ_amb(i-2) andIt is respectively the i-th-2 training sample T_i-2The load current in moment, ambient temperature and top layer oil Temperature；

Output unit, is used for obtaining x_iCorresponding regression forecasting errorWith output weight beta, wherein:

$\left\{\begin{array}{c}\mathrm{\Δ}{\mathrm{\θ}}_{oil}^e\left(x\right)={\left[\begin{array}{c}K(x,x_1)\\ \·\\ \·\\ \·\\ K(x,x_N)\end{array}\right]}^T\mathrm{\β}\\ \mathrm{\β}={(}^I\left[\begin{array}{c}{\mathrm{\Δ\θ}}_{oil}^T\left(x_1\right)\\ \·\\ \·\\ \·\\ {\mathrm{\Δ\θ}}_{oil}^T\left(x_N\right)\end{array}\right]\end{array}\right.$

K(x_i,x_j) it is kernel function；

Optimize unit, for using GA that nuclear parameter γ and the penalty coefficient C of KELM model are optimized so that's Mean square error MSE_ΔθMinimum, obtains GA-KELM model, wherein:

${\mathrm{MSE}}_{\mathrm{\Δ}\mathrm{\θ}}=\frac{1}{N}\underset{i=1}{\overset{N}{\Σ}}{({\mathrm{\Δ\θ}}_{oil}^T-{\mathrm{\Δ\θ}}_{oil}^e)}^2.$

Further, described training sample set acquisition module includes:

Data capture unit, is used for obtaining T_iTime inscribe the load current of transformator, ambient temperature and top-oil temperature；

Data-optimized unit, for rejecting the data of substantially inefficacy, and replaces fail data by interpolated value, obtains training sample This；

Training sample set acquiring unit, for forming training sample set by multiple training samples.

The method have the advantages that

The transformator top-oil temperature Forecasting Methodology based on predicted error amendment of the present invention, based on top-oil temperature error prediction Correction model, can obtain the top-oil temperature predictive value that precision is higher, it is possible to reliably analyze the dynamic change trend of top-oil temperature, And then improve transformator accuracy in terms of thermal characteristics measurement, be conducive to preferably instructing the load running of transformator, promote Enter transformator at the popularization of on-Line Monitor Device and the good application of on-line monitoring information.

Accompanying drawing explanation

Fig. 1 is the transformator top-oil temperature Forecasting Methodology flow chart based on predicted error amendment of the present invention；

Fig. 2 is the schematic diagram of Susa thermal circuit model；

Fig. 3 is KELM model schematic；

Fig. 4 is the transformator top-oil temperature Forecasting Methodology schematic diagram based on predicted error amendment of the present invention；

Fig. 5 is transformator measured data curve chart；

Fig. 6 is Susa thermal circuit model top-oil temperature predictive value and Susa thermal circuit model forecast error curve chart；

Fig. 7 be Susa thermal circuit model top-oil temperature actual prediction error and KELM matching prediction obtain forecast error contrast Figure；

Fig. 8 is the transformator top-oil temperature prediction means schematic diagram based on predicted error amendment of the present invention.

Detailed description of the invention

For making the technical problem to be solved in the present invention, technical scheme and advantage clearer, below in conjunction with accompanying drawing and tool Body embodiment is described in detail.

On the one hand, the present invention provides a kind of transformator top-oil temperature Forecasting Methodology based on predicted error amendment, such as Fig. 1 and Shown in Fig. 4, including:

Step 101: obtain predictive data set, it was predicted that data set includes load current and the environment temperature of current time transformator Degree, and load current, ambient temperature and the top-oil temperature of several moment transformators before current time.

Transformator top-oil temperature is closely related with the load current of transformator and ambient temperature, and top-oil temperature also can be subject to To the load current of transformator that (typically takes 15-30min) for the previous period, ambient temperature and the impact of top-oil temperature, we will The load current of transformator, ambient temperature and top-oil temperature are referred to as postponing item for the previous period, the survey top layer of transformator to be predicted Oil temperature, it is to be appreciated that the load current of transformator, ambient temperature and delay item.

Step 102: with the load current of current time transformator as input quantity, utilize Susa thermal circuit model to current time The top-oil temperature of transformator is predicted, and obtains Susa thermal circuit model top-oil temperature predictive value

It is aware of the load current of transformator, it is possible to utilize the top-oil temperature of Susa thermal circuit model prediction transformator, To Susa thermal circuit model top-oil temperature predictive valueSusa thermal circuit model is widely used in the pre-of transformator top-oil temperature Survey, but its model excessively simplifies, parameter calculate inaccurate, by the big problem of such environmental effects, there is systematic error in model, Susa thermal circuit model is as shown in Figure 2.

Step 103: by Susa thermal circuit model top-oil temperature predictive valueIt is normalized with predictive data set.Return One change processes and data can be made to have unified tolerance, convenient calculating.

Step 104: with the Susa thermal circuit model top-oil temperature predictive value after normalizationIt is input with predictive data set Amount, utilizes GA-KELM model (the core extreme learning machine of genetic optimization) to carry out regression forecasting, obtains regression forecasting error

Susa thermal circuit model top-oil temperature predictive value obtained aboveThere is certain error, need to use GA-KELM mould Type calculates regression forecasting errorDuring calculating, after normalizationIt is input quantity with predictive data set, is exported

Step 105: to regression forecasting errorCarry out renormalization process.Due to obtainAfter normalization Value, needs its renormalization.

Step 106: with the Susa thermal circuit model top-oil temperature predictive value before normalizationDeduct the recurrence after renormalization Forecast errorObtain transformator revised top-oil temperature predictive value, the top layer of transformator after the most measurable a period of time Oil temperature, according to top-oil temperature regulating load size.

The transformator top-oil temperature Forecasting Methodology based on predicted error amendment of the present invention, based on top-oil temperature error prediction Correction model, can obtain the top-oil temperature predictive value that precision is higher, it is possible to reliably analyze the dynamic change trend of top-oil temperature, And then improve transformator accuracy in terms of thermal characteristics measurement, be conducive to preferably instructing the load running of transformator, promote Enter transformator at the popularization of on-Line Monitor Device and the good application of on-line monitoring information.

The present invention needs to use GA-KELM model, and GA-KELM model can be trained by the following method and be obtained:

Step 201: obtaining the training sample set of transformator, training sample set includes multiple training sample, wherein, i-th Training sample includes T_iTime inscribe the load current of transformator, ambient temperature and top-oil temperature, i ∈ [1, N], N are training sample Number.

When obtaining training sample set, each a period of time gathers the load current of transformator, ambient temperature and top layer oil Temperature measured data, in this, as a training sample, acquisition time is spaced preferred 15min.

Step 202: to i-th training sample, with T_iTime inscribe the load current of transformator and top-oil temperature is input quantity, Utilize Susa thermal circuit model that top-oil temperature is predicted, obtain Susa thermal circuit model top-oil temperature predictive valueWith Susa heat Road model predictive error

In this step, obtain Susa thermal circuit model top-oil temperature predictive valueMethod identical with abovementioned steps 102, Susa thermal circuit model forecast errorFor top-oil temperature measured value and predictive valueDifference.

Step 203: by Susa thermal circuit model top-oil temperature predictive valueWith Susa thermal circuit model forecast errorCarry out Normalized.With aforementioned, for unified metric, convenient calculating, need to be normalized.

Step 204: setting up KELM model, KELM model is as shown in Figure 3.

Step 205: use the Susa thermal circuit model top-oil temperature predictive value of each training sampleSusa thermal circuit model Forecast errorIt is trained with training sample set pair KELM model, until it reaches pre-provisioning request, obtains GA-KELM model.

Step 201-205 is the method obtaining GA-KELM model, there is no strict precedence relationship with step 101-106, step Rapid 201-205 can be to be carried out complete in advance, had i.e. been previously obtained GA-KELM mould before carrying out top-oil temperature prediction Type, it is also possible to being trained when carrying out top-oil temperature prediction, the GA-KELM model obtained by said method has higher simultaneously Precision.

Further, in step 102 and step 202, Susa thermal circuit model top-oil temperature predictive valueCan be by as follows Formula is calculated:

$\left\{\begin{array}{c}\frac{K^2\mathrm{\α}+1}{\mathrm{\α}+1}{\left({\mathrm{\μ}}_{pu}\right)}^{\frac{1-n}{n}}{\mathrm{\Δ\θ}}_{oil,R}={\left({\mathrm{\μ}}_{pu}\right)}^{\frac{1-n}{n}}{\mathrm{\τ}}_{oil,R}\frac{{\mathrm{d\θ}}_{oil}^{e,b}}{dt}+\frac{{({\mathrm{\θ}}_{oil}^{e,b}-{\mathrm{\θ}}_{amb})}^{\frac{1}{n}}}{{\left({\mathrm{\Δ\θ}}_{oil,R}\right)}^{\frac{1-n}{n}}}\\ {\mathrm{\μ}}_{pu}=\frac{\mathrm{\μ}}{{\mathrm{\μ}}_R}=e^{\frac{2797.3}{{\mathrm{\θ}}_{oil}^{e,b}+273}-\frac{2797.3}{{\mathrm{\θ}}_{oil,R}+273}}\end{array}\right.$

Wherein, K is load factor, for the ratio of load current with rated current；α be nominal load loss with open circuit loss it Ratio；Δθ_oil,RFor under nominal load, the stable state temperature rise of top-oil temperature versus environmental；N is the experience of reflection transformator radiating mode Index；τ_oil,RFor specified top layer oil time constant, for thermal capacitance and the product of thermal resistance under nominal load；μ_puFor oil viscosity change because of Son, for the oil viscosity μ under arbitrary temp and the oil viscosity μ under specified top-oil temperature_RRatio.

And, in step 202, Susa thermal circuit model forecast errorIt is calculated by equation below:

${\mathrm{\Δ\θ}}_{oil}^T={\mathrm{\θ}}_{oil}^{e,b}-{\mathrm{\θ}}_{oil}^m$

Wherein,For T in training sample_iThe top-oil temperature of moment transformator.

As a kind of improvement of the transformator top-oil temperature Forecasting Methodology based on predicted error amendment of the present invention, step 205 include:

Step 2051: by input vector x_iInput KELM model, wherein:

$x_i=\[I\left(i\right),I(i-1),I(i-2),{\mathrm{\θ}}_{amb}\left(i\right),{\mathrm{\θ}}_{amb}(i-1),{\mathrm{\θ}}_{amb}(i-2),{\mathrm{\θ}}_{oil}^{e,b}\left(i\right),{\mathrm{\θ}}_{oil}^m(i-1),{\mathrm{\θ}}_{oil}^m(i-2)\]$

I (i) and θ_ambI () is respectively i-th training sample T_iThe load current in moment and ambient temperature, I (i-1), θ_amb (i-1) andIt is respectively the i-th-1 training sample T_i-1The load current in moment, ambient temperature and top-oil temperature, I (i- 2)、θ_amb(i-2) andIt is respectively the i-th-2 training sample T_i-2The load current in moment, ambient temperature and top layer oil Temperature.

Step 2052: obtain x_iCorresponding regression forecasting errorWith output weight beta, wherein:

$\left\{\begin{array}{c}\mathrm{\Δ}{\mathrm{\θ}}_{oil}^e\left(x\right)={\left[\begin{array}{c}K(x,x_1)\\ \·\\ \·\\ \·\\ K(x,x_N)\end{array}\right]}^T\mathrm{\β}\\ \mathrm{\β}={(}^I\left[\begin{array}{c}{\mathrm{\Δ\θ}}_{oil}^T\left(x_1\right)\\ \·\\ \·\\ \·\\ {\mathrm{\Δ\θ}}_{oil}^T\left(x_N\right)\end{array}\right]\end{array}\right.$

K(x_i,x_j) it is kernel function, it is typically set at RBF core.

Step 2053: use GA (genetic algorithm) that nuclear parameter γ and the penalty coefficient C of KELM model are optimized, obtain GA-KELM model so thatMean square error MSE_ΔθMinimum, wherein:

${\mathrm{MSE}}_{\mathrm{\Δ}\mathrm{\θ}}=\frac{1}{N}\underset{i=1}{\overset{N}{\Σ}}{({\mathrm{\Δ\θ}}_{oil}^T-{\mathrm{\Δ\θ}}_{oil}^e)}^2.$

Step 2051-2053 gives the embodiment being trained KELM model, it is possible to improve further The precision of GA-KELM model.

The precision choosing direct determined GA-KELM model of training sample set, therefore, step 201 includes:

Step 2011: obtain T_iTime inscribe the load current of transformator, ambient temperature and top-oil temperature.With 15min as time Between be spaced, gather the load current of transformator, ambient temperature and top-oil temperature

Step 2012: reject the data (gross error etc.) substantially lost efficacy, and replace fail data by interpolated value, instructed Practice sample.

Step 2013: multiple training samples are formed training sample set.

The training sample set that the present invention obtains enables to train the precision height of the GA-KELM model obtained.

With concrete example beneficial effects of the present invention is described below:

The measured data using certain 10kV/400V three-phase double winding distribution transformer carries out simulating, verifying, transformer parameter As shown in table 1.Gather transformator in November, 2013 load current, ambient temperature and the top-oil temperature of totally 7 days, sampling time interval It it is 15 minutes.Transformator measured data is as shown in Figure 5.

Table 1 transformer parameter

The predictive value correlation curve of Susa thermal circuit model and forecast error curve thereof are as shown in Figure 6.

Use KELM error prediction model that the forecast error of Susa thermal circuit model is fitted, is predicted, then use GA KELM parameter [C, γ] is carried out optimizing, tries to achieve optimized parameter [C, γ]=[0.83,0.092] corresponding to training sample.Ask After obtaining model optimized parameter, the measured data of 4 days first in the past and Susa thermal circuit model forecast error are as training sample pair KELM is trained, and is then predicted the Susa thermal circuit model forecast error of latter 3 days, finally with this forecast error correction The predictive value of Susa thermal circuit model, obtains final predictive value.The GA-KELM prediction to latter 3 days Susa thermal circuit model forecast erroies Value is with actual comparison as shown in Figure 7.

As seen from Figure 7, GA-KELM is basically identical with actual error to the predictive value of Susa thermal circuit model error, GA- The forecast error of Susa thermal circuit model preferably can be fitted and predict by KELM.Between forecast error and actual error Big difference is 1.29 DEG C, i.e. the largest prediction error of forecasting amendment model is 1.29 DEG C.

Be respectively adopted the core extreme learning machine (GA-KELM) of genetic optimization, the support vector machine (GA-SVM) of genetic optimization, Elman neutral net carries out Direct Modeling and prediction, as a comparison method to this paper sample.The precision of prediction of each method is to such as Shown in table 2, MSE is prediction mean square error, e_maxAbsolute value for largest prediction error.

Table 2 each method precision of prediction contrasts

As can be seen from Table 2: MSE and e that GA-KELM model is corresponding_maxMinimum, it was predicted that precision is the highest；Susa thermal circuit model Precision of prediction less than GA-KELM, GA-SVM and Elman neural network model；The precision of prediction of GA-KELM higher than GA-SVM and Elman neural network model.GA-KELM model combines the advantage of half physical model and mathematical model in principle, therefore takes Obtained higher precision of prediction.

On the other hand, the present invention provides a kind of transformator top-oil temperature prediction means based on predicted error amendment, such as Fig. 8 Shown in, including:

Predictive data set acquisition module 11, is used for obtaining predictive data set, it was predicted that data set includes current time transformator Load current and ambient temperature, and current time before the load current of several moment transformators, ambient temperature and top layer oil Temperature；

Top-oil temperature prediction module 12, for the load current of current time transformator as input quantity, utilizes Susa heat The top-oil temperature of transformator is predicted by road model, obtains Susa thermal circuit model top-oil temperature predictive value

First normalization module 13, for by Susa thermal circuit model top-oil temperature predictive valueCarry out with predictive data set Normalized；

Regression forecasting module 14, for the Susa thermal circuit model top-oil temperature predictive value after normalizationWith prediction number According to integrating as input quantity, utilize GA-KELM model to carry out regression forecasting, obtain regression forecasting error

Renormalization module 15, for regression forecasting errorCarry out renormalization process；

Correcting module 16, the Susa thermal circuit model top-oil temperature predictive value before using normalizationDeduct renormalization After regression forecasting errorObtain transformator revised top-oil temperature predictive value.

The transformator top-oil temperature prediction means based on predicted error amendment of the present invention, based on top-oil temperature error prediction Correction model, can obtain the top-oil temperature predictive value that precision is higher, it is possible to reliably analyze the dynamic change trend of top-oil temperature, And then improve transformator accuracy in terms of thermal characteristics measurement, be conducive to preferably instructing the load running of transformator, promote Enter transformator at the popularization of on-Line Monitor Device and the good application of on-line monitoring information.

The present invention needs to use GA-KELM model, and GA-KELM model can be obtained by such as lower module training:

Training sample set acquisition module, for obtaining the training sample set of transformator, training sample set includes multiple training Sample, wherein, i-th training sample includes T_iTime inscribe the load current of transformator, ambient temperature and top-oil temperature, i ∈ [1, N], N is training sample number；

Top-oil temperature and error prediction module, for i-th training sample, with T_iTime inscribe the load current of transformator It is input quantity with top-oil temperature, utilizes Susa thermal circuit model that top-oil temperature is predicted, obtain Susa thermal circuit model top layer oil Temperature predictive valueWith Susa thermal circuit model forecast error

Second normalization module, for by Susa thermal circuit model top-oil temperature predictive valuePredict with Susa thermal circuit model ErrorIt is normalized；

KELM model building module, is used for setting up KELM model；

Training module, for using the Susa thermal circuit model top-oil temperature predictive value of each training sampleThe hot road of Susa Model predictive errorIt is trained with training sample set pair KELM model, obtains GA-KELM model.

The GA-KELM model obtained by above-mentioned modules has higher precision.

Further, in top-oil temperature prediction module and top-oil temperature and error prediction module, Susa thermal circuit model top Layer oil temperature predictive valueIt is calculated by equation below:

$\left\{\begin{array}{c}\frac{K^2\mathrm{\α}+1}{\mathrm{\α}+1}{\left({\mathrm{\μ}}_{pu}\right)}^{\frac{1-n}{n}}{\mathrm{\Δ\θ}}_{oil,R}={\left({\mathrm{\μ}}_{pu}\right)}^{\frac{1-n}{n}}{\mathrm{\τ}}_{oil,R}\frac{{\mathrm{d\θ}}_{oil}^{e,b}}{dt}+\frac{{({\mathrm{\θ}}_{oil}^{e,b}-{\mathrm{\θ}}_{amb})}^{\frac{1}{n}}}{{\left({\mathrm{\Δ\θ}}_{oil,R}\right)}^{\frac{1-n}{n}}}\\ {\mathrm{\μ}}_{pu}=\frac{\mathrm{\μ}}{{\mathrm{\μ}}_R}=e^{\frac{2797.3}{{\mathrm{\θ}}_{oil}^{e,b}+273}-\frac{2797.3}{{\mathrm{\θ}}_{oil,R}+273}}\end{array}\right.$

Wherein, K is load factor, for the ratio of load current with rated current；α be nominal load loss with open circuit loss it Ratio；Δθ_oil,RFor under nominal load, the stable state temperature rise of top-oil temperature versus environmental；N is the experience of reflection transformator radiating mode Index；τ_oil,RFor specified top layer oil time constant, for thermal capacitance and the product of thermal resistance under nominal load；μ_puFor oil viscosity change because of Son, for the oil viscosity μ under arbitrary temp and the oil viscosity μ under specified top-oil temperature_RRatio；

Further, in top-oil temperature and error prediction module, Susa thermal circuit model forecast errorBy equation below meter Obtain:

${\mathrm{\Δ\θ}}_{oil}^T={\mathrm{\θ}}_{oil}^{e,b}-{\mathrm{\θ}}_{oil}^m$

Wherein,For T in training sample_iThe top-oil temperature of moment transformator.

As a kind of improvement of the transformator top-oil temperature prediction means based on predicted error amendment of the present invention, train mould Block includes:

Input block, for by input vector x_iInput KELM model, wherein:

$x_i=\[I\left(i\right),I(i-1),I(i-2),{\mathrm{\θ}}_{amb}\left(i\right),{\mathrm{\θ}}_{amb}(i-1),{\mathrm{\θ}}_{amb}(i-2),{\mathrm{\θ}}_{oil}^{e,b}\left(i\right),{\mathrm{\θ}}_{oil}^m(i-1),{\mathrm{\θ}}_{oil}^m(i-2)\]$

I (i) and θ_ambI () is respectively i-th training sample T_iThe load current in moment and ambient temperature, I (i-1), θ_amb (i-1) andIt is respectively the i-th-1 training sample T_i-1The load current in moment, ambient temperature and top-oil temperature, I (i- 2)、θ_amb(i-2) andIt is respectively the i-th-2 training sample T_i-2The load current in moment, ambient temperature and top layer oil Temperature；

Output unit, is used for obtaining x_iCorresponding regression forecasting errorWith output weight beta, wherein:

$\left\{\begin{array}{c}\mathrm{\Δ}{\mathrm{\θ}}_{oil}^e\left(x\right)={\left[\begin{array}{c}K(x,x_1)\\ \·\\ \·\\ \·\\ K(x,x_N)\end{array}\right]}^T\mathrm{\β}\\ \mathrm{\β}={(}^I\left[\begin{array}{c}{\mathrm{\Δ\θ}}_{oil}^T\left(x_1\right)\\ \·\\ \·\\ \·\\ {\mathrm{\Δ\θ}}_{oil}^T\left(x_N\right)\end{array}\right]\end{array}\right.$

K(x_i,x_j) it is kernel function；

Optimize unit, for using GA that nuclear parameter γ and the penalty coefficient C of KELM model are optimized so that's Mean square error MSE_ΔθMinimum, obtains GA-KELM model, wherein:

${\mathrm{MSE}}_{\mathrm{\Δ}\mathrm{\θ}}=\frac{1}{N}\underset{i=1}{\overset{N}{\Σ}}{({\mathrm{\Δ\θ}}_{oil}^T-{\mathrm{\Δ\θ}}_{oil}^e)}^2.$

Above-mentioned unit gives the embodiment being trained KELM model, it is possible to improve further The precision of GA-KELM model.

The precision choosing direct determined GA-KELM model of training sample set, therefore, training sample set acquisition module Including:

Data capture unit, is used for obtaining T_iTime inscribe the load current of transformator, ambient temperature and top-oil temperature；

Data-optimized unit, for rejecting the data of substantially inefficacy, and replaces fail data by interpolated value, obtains training sample This；

Training sample set acquiring unit, for forming training sample set by multiple training samples.

The training sample set that the present invention obtains enables to train the precision height of the GA-KELM model obtained.

It is above the preferred embodiment of the present invention, it is noted that for those skilled in the art, Under the premise without departing from the principles of the invention, it is also possible to make some improvements and modifications, these improvements and modifications also should be regarded as this The protection domain of invention.

Claims (10)

1. a transformator top-oil temperature Forecasting Methodology based on predicted error amendment, it is characterised in that including:

Step 101: obtaining predictive data set, described predictive data set includes load current and the environment temperature of current time transformator Degree, and load current, ambient temperature and the top-oil temperature of several moment transformators before current time；

Step 102: with the load current of current time transformator as input quantity, utilize the Susa thermal circuit model top layer to transformator Oil temperature is predicted, and obtains Susa thermal circuit model top-oil temperature predictive value

Step 103: by described Susa thermal circuit model top-oil temperature predictive valueIt is normalized with predictive data set；

Step 104: with the Susa thermal circuit model top-oil temperature predictive value after normalizationIt is input quantity with predictive data set, utilizes GA-KELM model carries out regression forecasting, obtains regression forecasting error

Step 105: to described regression forecasting errorCarry out renormalization process；

Step 106: with the Susa thermal circuit model top-oil temperature predictive value before normalizationDeduct the regression forecasting after renormalization ErrorObtain transformator revised top-oil temperature predictive value.

Transformator top-oil temperature Forecasting Methodology based on predicted error amendment the most according to claim 1, it is characterised in that Described GA-KELM model obtains by the following method:

Step 201: obtain the training sample set of described transformator, described training sample set includes multiple training sample, wherein, I training sample includes T_iTime inscribe the load current of transformator, ambient temperature and top-oil temperature, i ∈ [1, N], N for training sample This number；

Step 202: to i-th training sample, with T_iTime inscribe the load current of transformator and top-oil temperature is input quantity, utilize Top-oil temperature is predicted by Susa thermal circuit model, obtains Susa thermal circuit model top-oil temperature predictive valueWith Susa Re Lumo Type forecast error

Step 203: by described Susa thermal circuit model top-oil temperature predictive valueWith Susa thermal circuit model forecast errorCarry out Normalized；

Step 204: set up KELM model；

Step 205: use the Susa thermal circuit model top-oil temperature predictive value of each training sampleSusa thermal circuit model is predicted ErrorIt is trained with KELM model described in training sample set pair, obtains GA-KELM model.

Transformator top-oil temperature Forecasting Methodology based on predicted error amendment the most according to claim 1 and 2, its feature exists In:

In described step 102 and step 202, described Susa thermal circuit model top-oil temperature predictive valueCalculated by equation below Arrive:

$\left\{\begin{array}{c}\frac{K^2\mathrm{\α}+1}{\mathrm{\α}+1}{\left({\mathrm{\μ}}_{pu}\right)}^{\frac{1-n}{n}}{\mathrm{\Δ\θ}}_{oil,R}={\left({\mathrm{\μ}}_{pu}\right)}^{\frac{1-n}{n}}{\mathrm{\τ}}_{oil,R}\frac{{\mathrm{d\θ}}_{oil}^{e,b}}{dt}+\frac{{({\mathrm{\θ}}_{oil}^{e,b}-{\mathrm{\θ}}_{amb})}^{\frac{1}{n}}}{{\left({\mathrm{\Δ\θ}}_{oil,R}\right)}^{\frac{1-n}{n}}}\\ {\mathrm{\μ}}_{pu}=\frac{\mathrm{\μ}}{{\mathrm{\μ}}_R}{e}^{\frac{2797.3}{{\mathrm{\θ}}_{oil}^{e,b}+273}-\frac{2797.3}{{\mathrm{\θ}}_{oil,R}+273}}\end{array}\right.$

Wherein, K is load factor, for the ratio of load current with rated current；α is the ratio of nominal load loss and open circuit loss； Δθ_oil,RFor under nominal load, the stable state temperature rise of top-oil temperature versus environmental；N is that the experience of reflection transformator radiating mode refers to Number；τ_oil,RFor specified top layer oil time constant, for thermal capacitance and the product of thermal resistance under nominal load；μ_puFor oil viscosity changed factor, For the oil viscosity μ under arbitrary temp and the oil viscosity μ under specified top-oil temperature_RRatio；

In described step 202, described Susa thermal circuit model forecast errorIt is calculated by equation below:

${\mathrm{\Δ\θ}}_{oil}^T={\mathrm{\θ}}_{oil}^{e,b}-{\mathrm{\θ}}_{oil}^m$

Wherein,For T in training sample_iThe top-oil temperature of moment transformator.

Transformator top-oil temperature Forecasting Methodology based on predicted error amendment the most according to claim 3, it is characterised in that Described step 205 includes:

Step 2051: by input vector x_iInput KELM model, wherein:

$x_i\[I\left(i\right),I(i-1),I(i-2),{\mathrm{\θ}}_{amb}\left(i\right),{\mathrm{\θ}}_{amb}(i-1),{\mathrm{\θ}}_{amb}(i-2),{\mathrm{\θ}}_{oil}^{e,b}\left(i\right),{\mathrm{\θ}}_{oil}^m(i-1),{\mathrm{\θ}}_{oil}^m(i-2)\]$

I (i) and θ_ambI () is respectively i-th training sample T_iThe load current in moment and ambient temperature, I (i-1), θ_amb(i-1) WithIt is respectively the i-th-1 training sample T_i-1The load current in moment, ambient temperature and top-oil temperature, I (i-2), θ_amb(i-2) andIt is respectively the i-th-2 training sample T_i-2The load current in moment, ambient temperature and top-oil temperature；

Step 2052: obtain x_iCorresponding regression forecasting errorWith output weight beta, wherein:

$\left\{\begin{array}{c}{\mathrm{\Δ\θ}}_{oil}^e\left(x\right)={\left[\begin{array}{c}K(x,x_1)\\ .\\ .\\ .\\ K(x,x_N)\end{array}\right]}^T\mathrm{\β}\\ \mathrm{\β}={(I/C+{\mathrm{\Ω}}_{ELM})}^{-1}\left[\begin{array}{c}{\mathrm{\Δ\θ}}_{oil}^T\left(x_1\right)\\ .\\ .\\ .\\ {\mathrm{\Δ\θ}}_{oil}^T\left(x_N\right)\end{array}\right]\end{array}\right.$

K(x_i,x_j) it is kernel function；

Step 2053: use GA that nuclear parameter γ and the penalty coefficient C of KELM model are optimized so thatMean square error MSE_ΔθMinimum, obtains GA-KELM model, wherein:

${\mathrm{MSE}}_{\mathrm{\Δ}\mathrm{\θ}}=\frac{1}{N}\underset{i=1}{\overset{N}{\Σ}}{({\mathrm{\Δ\θ}}_{oil}^T-{\mathrm{\Δ\θ}}_{oil}^e)}^2.$

Transformator top-oil temperature Forecasting Methodology based on predicted error amendment the most according to claim 3, it is characterised in that Described step 201 includes:

Step 2011: obtain T_iTime inscribe the load current of transformator, ambient temperature and top-oil temperature；

Step 2012: reject the data substantially lost efficacy, and replace fail data by interpolated value, obtain training sample；

Step 2013: multiple training samples are formed training sample set.

6. a transformator top-oil temperature prediction means based on predicted error amendment, it is characterised in that including:

Predictive data set acquisition module, is used for obtaining predictive data set, and described predictive data set includes current time transformator Load current, ambient temperature and the top-oil temperature of several moment transformators before load current and ambient temperature, and current time；

Top-oil temperature prediction module, for the load current of current time transformator as input quantity, utilizing Susa thermal circuit model The top-oil temperature of transformator is predicted, obtains Susa thermal circuit model top-oil temperature predictive value

First normalization module, for by described Susa thermal circuit model top-oil temperature predictive valueNormalizing is carried out with predictive data set Change processes；

Regression forecasting module, for the Susa thermal circuit model top-oil temperature predictive value after normalizationIt is defeated with predictive data set Enter amount, utilize GA-KELM model to carry out regression forecasting, obtain regression forecasting error

Renormalization module, for described regression forecasting errorCarry out renormalization process；

Correcting module, the Susa thermal circuit model top-oil temperature predictive value before using normalizationDeduct returning after renormalization Return forecast errorObtain transformator revised top-oil temperature predictive value.

Transformator top-oil temperature prediction means based on predicted error amendment the most according to claim 6, it is characterised in that Described GA-KELM model is obtained by such as lower module:

Training sample set acquisition module, for obtaining the training sample set of described transformator, described training sample set includes multiple Training sample, wherein, i-th training sample includes T_iTime inscribe the load current of transformator, ambient temperature and top-oil temperature, i ∈ [1, N], N are training sample number；

Top-oil temperature and error prediction module, for i-th training sample, with T_iTime inscribe load current and the top of transformator Layer oil temperature is input quantity, utilizes Susa thermal circuit model to be predicted top-oil temperature, obtains Susa thermal circuit model top-oil temperature pre- Measured valueWith Susa thermal circuit model forecast error

Second normalization module, for by described Susa thermal circuit model top-oil temperature predictive valueWith the prediction of Susa thermal circuit model by mistake DifferenceIt is normalized；

KELM model building module, is used for setting up KELM model；

Training module, for using the Susa thermal circuit model top-oil temperature predictive value of each training sampleSusa thermal circuit model Forecast errorIt is trained with KELM model described in training sample set pair, obtains GA-KELM model.

8., according to the transformator top-oil temperature prediction means based on predicted error amendment described in claim 6 or 7, its feature exists In:

In described top-oil temperature prediction module and top-oil temperature and error prediction module, described Susa thermal circuit model top-oil temperature Predictive valueIt is calculated by equation below:

$\left\{\begin{array}{c}\frac{K^2\mathrm{\α}+1}{\mathrm{\α}+1}{\left({\mathrm{\μ}}_{pu}\right)}^{\frac{1-n}{n}}{\mathrm{\Δ\θ}}_{oil,R}={\left({\mathrm{\μ}}_{pu}\right)}^{\frac{1-n}{n}}{\mathrm{\τ}}_{oil,R}\frac{{\mathrm{d\θ}}_{oil}^{e,b}}{dt}+\frac{{({\mathrm{\θ}}_{oil}^{e,b}-{\mathrm{\θ}}_{amb})}^{\frac{1}{n}}}{{\left({\mathrm{\Δ\θ}}_{oil,R}\right)}^{\frac{1-n}{n}}}\\ {\mathrm{\μ}}_{pu}=\frac{\mathrm{\μ}}{{\mathrm{\μ}}_R}{e}^{\frac{2797.3}{{\mathrm{\θ}}_{oil}^{e,b}+273}-\frac{2797.3}{{\mathrm{\θ}}_{oil,R}+273}}\end{array}\right.$

Wherein, K is load factor, for the ratio of load current with rated current；α is the ratio of nominal load loss and open circuit loss； Δθ_oil,RFor under nominal load, the stable state temperature rise of top-oil temperature versus environmental；N is that the experience of reflection transformator radiating mode refers to Number；τ_oil,RFor specified top layer oil time constant, for thermal capacitance and the product of thermal resistance under nominal load；μ_puFor oil viscosity changed factor, For the oil viscosity μ under arbitrary temp and the oil viscosity μ under specified top-oil temperature_RRatio；

In described top-oil temperature and error prediction module, described Susa thermal circuit model forecast errorCalculated by equation below Obtain:

${\mathrm{\Δ\θ}}_{oil}^T={\mathrm{\θ}}_{oil}^{e,b}-{\mathrm{\θ}}_{oil}^m$

Wherein,For T in training sample_iThe top-oil temperature of moment transformator.

Transformator top-oil temperature prediction means based on predicted error amendment the most according to claim 8, it is characterised in that Described training module includes:

Input block, for by input vector x_iInput KELM model, wherein:

$x_i\[I\left(i\right),I(i-1),I(i-2),{\mathrm{\θ}}_{amb}\left(i\right),{\mathrm{\θ}}_{amb}(i-1),{\mathrm{\θ}}_{amb}(i-2),{\mathrm{\θ}}_{oil}^{e,b}\left(i\right),{\mathrm{\θ}}_{oil}^m(i-1),{\mathrm{\θ}}_{oil}^m(i-2)\]$

I (i) and θ_ambI () is respectively i-th training sample T_iThe load current in moment and ambient temperature, I (i-1), θ_amb(i-1) WithIt is respectively the i-th-1 training sample T_i-1The load current in moment, ambient temperature and top-oil temperature, I (i-2), θ_amb(i-2) andIt is respectively the i-th-2 training sample T_i-2The load current in moment, ambient temperature and top-oil temperature；

Output unit, is used for obtaining x_iCorresponding regression forecasting errorWith output weight beta, wherein:

$\left\{\begin{array}{c}{\mathrm{\Δ\θ}}_{oil}^e\left(x\right)={\left[\begin{array}{c}K(x,x_1)\\ .\\ .\\ .\\ K(x,x_N)\end{array}\right]}^T\mathrm{\β}\\ \mathrm{\β}={(I/C+{\mathrm{\Ω}}_{ELM})}^{-1}\left[\begin{array}{c}{\mathrm{\Δ\θ}}_{oil}^T\left(x_1\right)\\ .\\ .\\ .\\ {\mathrm{\Δ\θ}}_{oil}^T\left(x_N\right)\end{array}\right]\end{array}\right.$

K(x_i,x_j) it is kernel function；

Optimize unit, for using GA that nuclear parameter γ and the penalty coefficient C of KELM model are optimized so thatMean square Error MSE_ΔθMinimum, obtains GA-KELM model, wherein:

${\mathrm{MSE}}_{\mathrm{\Δ}\mathrm{\θ}}=\frac{1}{N}\underset{i=1}{\overset{N}{\Σ}}{({\mathrm{\Δ\θ}}_{oil}^T-{\mathrm{\Δ\θ}}_{oil}^e)}^2.$

Transformator top-oil temperature prediction means based on predicted error amendment the most according to claim 8, its feature exists In, described training sample set acquisition module includes:

Data capture unit, is used for obtaining T_iTime inscribe the load current of transformator, ambient temperature and top-oil temperature；

Data-optimized unit, for rejecting the data of substantially inefficacy, and replaces fail data by interpolated value, obtains training sample；

Training sample set acquiring unit, for forming training sample set by multiple training samples.