CN106126944B - A kind of power transformer top-oil temperature interval prediction method and system - Google Patents
A kind of power transformer top-oil temperature interval prediction method and system Download PDFInfo
- Publication number
- CN106126944B CN106126944B CN201610489216.8A CN201610489216A CN106126944B CN 106126944 B CN106126944 B CN 106126944B CN 201610489216 A CN201610489216 A CN 201610489216A CN 106126944 B CN106126944 B CN 106126944B
- Authority
- CN
- China
- Prior art keywords
- oil temperature
- prediction
- learning machine
- data
- kernel
- 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.)
- Expired - Fee Related
Links
- 238000000034 method Methods 0.000 title claims abstract description 64
- 238000012549 training Methods 0.000 claims abstract description 68
- 238000012795 verification Methods 0.000 claims abstract description 39
- 238000005070 sampling Methods 0.000 claims description 23
- 238000010606 normalization Methods 0.000 claims description 20
- 238000012545 processing Methods 0.000 claims description 17
- 239000013598 vector Substances 0.000 claims description 11
- 230000006870 function Effects 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 5
- 239000011159 matrix material Substances 0.000 claims description 3
- 238000004422 calculation algorithm Methods 0.000 description 7
- 238000012706 support-vector machine Methods 0.000 description 7
- 230000009286 beneficial effect Effects 0.000 description 4
- 238000013528 artificial neural network Methods 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 108090000672 Annexin A5 Proteins 0.000 description 1
- 230000032683 aging Effects 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000009413 insulation Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000004804 winding Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16Z—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS, NOT OTHERWISE PROVIDED FOR
- G16Z99/00—Subject matter not provided for in other main groups of this subclass
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J2203/00—Indexing scheme relating to details of circuit arrangements for AC mains or AC distribution networks
- H02J2203/10—Power transmission or distribution systems management focussing at grid-level, e.g. load flow analysis, node profile computation, meshed network optimisation, active network management or spinning reserve management
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Business, Economics & Management (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Tourism & Hospitality (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- General Business, Economics & Management (AREA)
- Quality & Reliability (AREA)
- Operations Research (AREA)
- Marketing (AREA)
- Entrepreneurship & Innovation (AREA)
- Housings And Mounting Of Transformers (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of power transformer top-oil temperature interval prediction method and system, based on core extreme learning machine and Bootstrap methods, obtain original training set data, and sub- training set is generated by Bootstrap methods；Multiple core extreme learning machine top-oil temperature prediction models are trained using sub- training set data；Original training set is predicted with multiple core limit learning models, the training sample of noise prediction core extreme learning machine, and training noise prediction core extreme learning machine are generated according to prediction result；Verification collection is predicted using multiple core extreme learning machine top-oil temperature prediction models, and using the observation noise variance of noise prediction core extreme learning machine prediction top-oil temperature；According to the observation noise variance that is obtained to the variance of top-oil temperature prediction result and prediction of multiple core limit study, the forecast interval of top-oil temperature is calculated.The present invention can obtain the clear reliable transformer top-oil temperature forecast interval in certain confidence level.
Description
Technical Field
The invention relates to a method and a system for predicting top layer oil temperature intervals of a power transformer.
Background
The safety and aging rate of the oil-paper insulation system of the oil-immersed power transformer are mainly influenced by the internal temperature rise, and the top oil temperature is one of important thermal variables for describing the internal temperature rise state of the transformer. During the operation of the transformer load, it must be ensured that the limit value is not exceeded. According to the load curve and the ambient temperature condition of the transformer, the top-layer oil temperature at the future moment is accurately and reliably predicted, on one hand, the load capacity of the transformer can be fully utilized on the premise of ensuring the safety and the reliability of the transformer, and on the other hand, the method has important significance for preventing the transformer from generating overheat faults.
At present, all prediction methods for the top-layer oil temperature of the transformer are point prediction methods, namely, a specific prediction value at a certain moment is given, only the total error level is taken as a performance index of prediction precision, and uncertainty existing in a prediction model and positive and negative characteristics of a prediction error are not considered.
Disclosure of Invention
The invention provides a method and a system for predicting the top oil temperature interval of a power transformer in order to solve the problems.
In order to achieve the purpose, the invention adopts the following technical scheme:
a top-layer oil temperature interval prediction method for a power transformer comprises the following steps:
(1) Collecting load current, ambient temperature and top oil temperature of a power transformer to form an original measured data set;
(2) Randomly sampling an original measured data set with multiple replacement by using a Bootstrap method to form multiple groups of random data sets;
(3) Carrying out normalization processing on the random data sets, taking each group of random data sets as a training set of a kernel extreme learning machine model, and training the kernel extreme learning machine;
(4) Predicting the original measured data set by using each trained kernel limit learning machine model to form a plurality of prediction outputs, and solving the average value and the variance of the prediction outputs;
(5) Forming a new kernel limit learning machine model according to the predicted output, and approximating the observation noise variance of the top oil temperature;
(6) Acquiring input data of a verification kernel extreme learning machine model to form a verification set, carrying out normalization processing on the verification set, predicting the verification set by each trained kernel extreme learning machine model, and solving the average value and the variance of the verification set;
(7) And predicting the observation noise variance of the top oil temperature of the verification set by using a new kernel limit learning machine model, calculating a prediction interval on a set confidence level, and performing inverse normalization processing on a prediction result.
In the step (1), the original measured data set comprises a plurality of data pairs, each data pair comprises input data and target data, and the input data is a vector consisting of the load current and the ambient temperature of the transformer at the current moment, and the load current, the ambient temperature and the top oil temperature of the transformer at the previous two sampling moments; the target data is the measured value of the top oil temperature of the transformer at the current moment.
In the step (2), the Bootstrap method is adopted to carry out random sampling on the original data set with the original data set in a place-back manner for L times, the number of samples collected each time is N, and is the same as the number of sample data pairs of the original data set, and finally L groups of data sets are obtained.
In the step (3), the training method of the kernel limit learning machine model specifically comprises the following steps: training samples are D = { (x) n ,y n ) N =1,2 \8230n }, input data x n And target data y n And calculating the connection weight beta of the hidden layer and the output layer of the kernel limit learning machine model according to the following formula:
wherein I is a diagonal matrix, C is a penalty coefficient, y outputs a target vector, y = [ y ] 1 ,y 2 ,…,y n ] T ,n＝1,2…N，K(x i ,x j ) For the kernel function, set to RBF kernel:
K(x i ,x j )＝exp(-γ||x i ,x j || 2 )
wherein, γ>, 0, as a kernel parameter, | | x i -x j And | | is a European norm.
In the step (5), a new training set D is constructed new ＝{(x n ,r 2 (x n ) N =1,2 \ 8230n }, where the residual r 2 (x n ) As shown in the following formula:
wherein, when the nth data point { x n ,y n Sampling at Bootstrap to get the l training set D l In the middle, q bn =0; otherwise, q is bn =1, the predicted output of the L +1 th kernel-limit learning model is g (x), which is an estimate of the observed noise variance σ 2 ∈ (x) for the top oil temperature.
And (4) the verification set in the step (6) is a vector consisting of the load current and the ambient temperature of the transformer at the current moment, and the load current, the ambient temperature and the top-layer oil temperature of the transformer at the previous two sampling moments.
In the step (6), the data of the verification set is in the same format as the data of the original measured data set.
In the step (7), the upper and lower limit values of the prediction interval of the model at the (1-alpha)% confidence level are calculated, and the specific method is as follows:
in the formulaAs a function of the t distribution of the degrees of freedom dfQuantile, degree of freedom df is taken as L.
A top-layer oil temperature interval prediction system of a power transformer comprises:
the system comprises an original measured data set acquisition module, a data processing module and a data processing module, wherein the original measured data set acquisition module is used for acquiring load current, ambient temperature and top oil temperature of a power transformer to form an original measured data set;
the method comprises the following steps that a Bootstrap method sub-training set generation module performs repeated replaced random sampling on an original measured data set by using a Bootstrap method to form a plurality of groups of random data sets, and each group of random data sets is used as a kernel limit learning machine model to be trained;
the top-layer oil temperature prediction module is used for predicting the original measured data set by utilizing each trained kernel limit learning machine model to form a plurality of prediction outputs and solving the average value and the variance of the prediction outputs;
the noise variance training set generation module is used for predicting the verification set by each trained kernel limit learning machine model, calculating to obtain a residual value according to a prediction result, and generating a new kernel limit learning machine model training set for predicting the top oil temperature observation noise variance;
the verification set data acquisition module is used for acquiring input data of the verification kernel limit learning machine model to form a verification set;
the normalization module is used for performing normalization processing on the original measured data set and the verification set;
the prediction interval calculation module is used for calculating the average value and the variance of the top oil temperature kernel limit learning machine model according to the prediction value of the top oil temperature kernel limit learning machine model, predicting the noise variance by adopting the noise variance kernel limit learning machine model and finally calculating the upper limit value and the lower limit value of the prediction interval;
and the anti-normalization module is used for carrying out anti-normalization processing on the upper limit value and the lower limit value of the prediction interval.
The beneficial effects of the invention are as follows:
the method for the top-layer oil temperature interval of the power transformer based on the kernel limit learning machine and the Bootstrap method can obtain a reliable and clear top-layer oil temperature prediction interval, can give the upper limit value and the lower limit value of the top-layer oil temperature of the transformer at a certain confidence level, can provide more reliable information compared with the traditional method of giving only a certain accurate top-layer oil temperature prediction value, is beneficial to better guiding the load operation of the transformer, and promotes the deep application of the online monitoring information of the transformer.
Drawings
FIG. 1 is a flow chart of a method for a top-layer oil temperature interval of a power transformer based on a nuclear limit learning machine and a Bootstrap method;
FIG. 2 is a schematic view of a KELM model;
FIG. 3 is a schematic diagram of a method for predicting the top-layer oil temperature interval of the power transformer based on a nuclear limit learning machine and a Bootstrap method;
FIG. 4 is a graph of measured data of a transformer;
FIG. 5 is a top oil temperature prediction interval value and a top oil temperature measured curve chart according to the method of the present invention;
fig. 6 is a schematic diagram of a device for predicting the top-layer oil temperature interval of the power transformer based on a nuclear limit learning machine and a Bootstrap method according to the present invention.
The specific implementation mode is as follows:
the invention is further described with reference to the following figures and examples.
A method and a device for predicting a top layer oil temperature interval of a power transformer based on a nuclear limit learning machine and a Bootstrap method belong to the field of transformer online monitoring, and the method comprises the following steps: acquiring original training set data, and generating a sub-training set by a Bootstrap method; training a plurality of top oil temperature prediction models of the kernel extreme learning machines by adopting sub-training set data; predicting an original training set by using a plurality of kernel extreme learning models, generating a training sample of a noise prediction kernel extreme learning machine according to a prediction result, and training the noise prediction kernel extreme learning machine; predicting the verification set by adopting a plurality of kernel extreme learning machine top oil temperature prediction models, and predicting the observation noise variance of the top oil temperature by adopting a noise prediction kernel extreme learning machine; and calculating a prediction interval of the top oil temperature according to the variance of the top oil temperature prediction results of the multiple kernel limit learning and the observation noise variance obtained by prediction.
Specifically, the invention provides a power transformer top layer oil temperature interval prediction based on a nuclear limit learning machine and a Bootstrap method, as shown in fig. 1 and 3, the prediction method comprises the following steps:
step 101: obtaining the originalStarting measured data set D = { (x) n ,y n ) N =1,2 \ 8230n }, the raw measured data set comprising a plurality of data pairs, each data pair comprising input data x n And target data y n . The input data are vectors consisting of the load current and the ambient temperature of the transformer at the current moment, and the load current, the ambient temperature and the top oil temperature of the transformer at the previous two sampling moments; the target data is the measured value of the top oil temperature of the transformer at the current moment.
Step 102: randomly sampling the original data set with a Bootstrap method in L times (L = 20-50, typically L = 30) with a set back, collecting N samples each time, the number of the samples is the same as that of the sample data pairs of the original data set, and finally obtaining L groups of data sets, such as D in FIG. 3 1 ～D L ；
Step 103: normalizing the L groups of data sets obtained in the step 102;
step 104: respectively training L Kernel Extreme Learning Machine (KELM) models according to the normalized L group data sets;
step 105: predicting an original data set by adopting trained L KELM models, obtaining L prediction outputs for each moment, and solving the average value and variance of the L outputs at each moment;
step 106: constructing a group of training sets according to the prediction result of the step 105, and training a KELM model (the L +1 th KELM model) for approximating the observation noise variance of the top oil temperature;
step 107: acquiring verification set data, wherein the data format is the same as that of the training set data, and performing normalization processing on the verification set data;
step 108: and predicting the verification set by using L KELM models, calculating the output average value and variance of the L models, and predicting the observation noise variance of the top oil temperature of the verification set by using an L +1 th KELM model.
Step 109: from the prediction results of step 108, a prediction interval at the (1- α)% confidence level is calculated, and the prediction results are denormalized.
Further, in step 101, in each data pair, the input data is as follows:
x n ＝[I(k),I(k-1),I(k-2),θ amb (k),θ amb (k-1),θ amb (k-2),θ oil (k-1),θ oil (k-2)]
wherein, the parenthesis inside k, k-1, k-2 respectively represent the current time, the previous sampling time and the previous two sampling times. I is the load current, θ amb Is the ambient temperature, θ oil The measured top oil temperature is obtained. Target data y n Is the measured top oil temperature value at the current moment. Considering that the top-layer oil temperature of the transformer is closely related to the load current and the ambient temperature of the transformer, and the top-layer oil temperature is also affected by the load current, the ambient temperature and the top-layer oil temperature of the transformer in the previous period (generally taking 15-30 min), the two sampling time intervals before and after the model input are preferably 15min.
In steps 103 and 109, the calculation formulas of the normalization processing and the inverse normalization processing are respectively as follows:
s'＝(s-s min )/(s max -s)
s＝(s max s'-s min )/(s'-1)
wherein s and s' are values before and after normalization of each variable, respectively, s min And s max Respectively, the minimum and maximum values of each variable.
In steps 104 and 108, the training method of the KELM model is specifically as follows: training samples are D = { (x) n ,y n ) N =1,2 \8230n }, and the connection weight β of the KELM hidden layer and the output layer is obtained according to the following formula:
wherein I is a diagonal matrix, C is a penalty coefficient, y outputs a target vector, y = [ y ] 1 ,y 2 ,…,y n ] T ,n＝1,2…N。K(x i ,x j ) For kernel functions, set to RBF kernel:
K(x i ,x j )＝exp(-γ||x i ,x j || 2 )
wherein, γ>, 0, as a kernel parameter, | | x i -x j And | | is a Euclidean norm.
The prediction output formula of the trained kernel-limit learning machine model is as follows:
the topological structure of the kernel limit learning is shown in fig. 2, and the kernel limit learning machine has better nonlinear fitting regression capability, so that the prediction of the top layer oil temperature by using the kernel limit learning machine is reasonable.
In the step 105, the average value of the predicted outputs of the L KELM models at each momentAnd the variance σ 2 boot (x) is as follows:
wherein, willAs an estimate of the true value μ (x), σ 2 boot (x) is taken as an estimate of the predicted value variance σ 2 f (x) caused by the model itself.
Suppose there is a set of data D = { (x) n ,y n ) N =1,2 \8230n }, where the target value y and the input vector x have a non-linear mapping relationship μ (x), the target value being affected by noise:
y＝μ(x)+ε(x)
ε (x) is the target value observed noise, generally following a 0-mean normal distribution. Performing fitting regression prediction on the time sequence by using a KELM model, wherein the prediction output isThe prediction error is then:
y-f(x)＝μ(x)-f(x)+ε(x)
the error includes a system error mu (x) -f (x) (the difference between the real value and the predicted value) of the model and a data observation noise error epsilon (x), which are generally considered to be independent of each other, and the variance corresponding to the prediction error y-f (x) is:
σ 2 boot (x) is taken as an estimate of the model self-variance σ 2 f (x).
In said step 106, a new training set D is constructed new ＝{(x n ,r 2 (x n ) N =1,2 8230n }, where the residual r is 2 (x n ) As shown in the following formula:
wherein, when the nth data point { x } n ,y n The l training set D obtained by Bootstrap sampling l In the middle, q bn =0; otherwise, q is bn =1. The predicted output of the L +1 th KELM model is g (x), which is an estimate of the observed noise variance σ 2 ε (x) for the top oil temperature.
In step 109, the upper and lower prediction interval limits at the (1- α)% confidence level for the model are calculated by:
in the formulaAs a function of the t distribution of the degrees of freedom dfQuantile, and the value of the degree of freedom df is L.
The method adopts the coverage (PI coverage, PICP) and the average prediction interval Width (Mean PI Width, MPIW) as evaluation indexes of a prediction interval, and the evaluation indexes are respectively used for measuring the reliability and the definition of the prediction interval, and the calculation formula is as follows:
wherein, N test Verifying the number of samples; c. C i Is a Boolean quantity, if the predicted target value is in the prediction interval c i =1, otherwise, c i ＝0；U i And L i Respectively representing the upper and lower bounds of the prediction interval.
The method for the top-layer oil temperature interval of the power transformer based on the kernel limit learning machine and the Bootstrap method can obtain a reliable and clear top-layer oil temperature prediction interval, can give the upper limit value and the lower limit value of the top-layer oil temperature of the transformer at a certain confidence level, can provide more reliable information compared with the traditional method of giving only a certain accurate top-layer oil temperature prediction value, is beneficial to better guiding the load operation of the transformer, and promotes the deep application of the online monitoring information of the transformer.
Steps 101-106 are the building process of the Bootstrap-KELM top layer oil temperature interval prediction model, which is obtained based on the measured data and by using the measured data as a training set, and is completed before the top layer oil temperature prediction is performed. If the model building and training process is performed before the top oil temperature prediction is performed, the process may be started from step 107. By the method, a prediction interval result with high coverage rate and narrow average prediction interval width can be obtained.
The beneficial effects of the invention are illustrated in the following by specific examples:
the method comprises the steps of carrying out simulation verification by adopting measured data of a certain 10kV/400V three-phase double-winding distribution transformer, collecting load current, ambient temperature and top oil temperature of the transformer for 7 days in 11 months in 2013, wherein the sampling time interval is 15 minutes. The measured data of the transformer is shown in fig. 4.
Generating sample input data, taking data of the first 4 days as an original training set, taking data of the last 3 days as a verification set, and performing L =30 times of random back-placing sampling on the data of the original training set to obtain L =30 sub-training sets. And selecting a KELM model RBF core parameter gamma =1 and a penalty coefficient C =0.5, and training L KELMs through sub-training set data. And predicting the original training set through the trained L KELM models, generating a training set of the L +1 th KELM model according to a prediction result, and training the training set. And predicting the verification set by using L KELMs, predicting the observation variance by using an L +1 th KELM model, and generating prediction intervals under 95%, 90% and 80% confidence levels according to prediction results, wherein curves of the prediction intervals and the actually measured top oil temperature are shown in FIG. 5.
An Extreme Learning Machine (ELM), a BP neural network and a Support Vector Machine (SVM) are respectively adopted as a comparison algorithm of a KELM algorithm, a Bootstrap interval prediction model is established, and the prediction interval index pair ratio of each method is shown in table 1.
TABLE 1 comparison of prediction interval indexes of methods
As can be seen from table 1: on the 95% confidence level, the coverage rate of each algorithm reaches 100%, and the average prediction interval width of the KELM and the SVM algorithms is similar and superior to that of the BP neural network and the SVM algorithm; at the confidence level of 90% and 80%, the coverage rate and average prediction interval width indexes of the BP neural network and the ELM are obviously inferior to those of the KELM and the SVM; from the aspect of algorithm training complexity, the KELM only needs one-step analysis, and the SVM adopts a method for solving convex optimization, so that the KELM has advantages over the SVM in the aspect of algorithm complexity.
In another aspect, the present invention provides an error prediction correction-based transformer top-layer oil temperature prediction apparatus, as shown in fig. 6, including:
an original training data set obtaining module 11, configured to obtain an original training data set, where the original measured data set includes a plurality of data pairs, each data pair includes input data and target data, the input data is a vector formed by a load current and an ambient temperature of a transformer at a current time and load currents, the ambient temperature, and a top-layer oil temperature of the transformers at two previous sampling times, and the target data is a measured value of the top-layer oil temperature of the transformer at the current time;
and the Bootstrap method sub-training set generation module 12 is used for generating L sub-training sets which are respectively used for training L KELM models. Random sampling with replacement is adopted, and the number of samples collected each time is N, which is the same as the number of sample data pairs in the original data set.
And a verification set (prediction set) data obtaining module 13, configured to obtain input data of a verification set KELM model, so as to obtain a predicted value. The verification set data is a vector composed of the load current and the ambient temperature of the transformer at the current moment, and the load current, the ambient temperature and the top oil temperature of the transformer at the previous two sampling moments.
A normalization module 14, configured to normalize the data of the original training set, the sub-training set, and the verification set;
the top oil temperature KELM prediction module 15 comprises L KELM models which are obtained by respectively adopting L sub-training sets for training and used for predicting the top oil temperature value f l (x)。
And a noise variance training set generation module 16, which predicts the original training set by using L KELM models, calculates residual values according to prediction results, and generates an L +1 training set of the KELM model for predicting the observation noise variance of the top oil temperature.
And a noise variance KELM prediction module 17 for adopting the training set of the noise variance training set generation module to train and predicting the noise variance observed by the top oil temperature.
And the prediction interval calculation module 18 is used for calculating the average value and the variance of the L top oil temperature KELM models according to the predicted values of the top oil temperature KELM models, predicting the noise variance by using the noise variance KELM prediction module, and finally calculating to obtain the upper limit value and the lower limit value of the prediction interval.
And the inverse normalization module 19 is configured to perform inverse normalization processing on the upper and lower limit values of the prediction interval.
Although the embodiments of the present invention have been described with reference to the accompanying drawings, it is not intended to limit the scope of the present invention, and it should be understood by those skilled in the art that various modifications and variations can be made without inventive efforts by those skilled in the art based on the technical solution of the present invention.
Claims (8)
1. A method for predicting a top layer oil temperature interval of a power transformer is characterized by comprising the following steps: the method comprises the following steps:
(1) Collecting load current, ambient temperature and top oil temperature of a power transformer to form an original measured data set;
(2) Randomly sampling an original measured data set with a Bootstrap method in a multiple-time replacement mode to form a plurality of groups of random data sets;
(3) Carrying out normalization processing on the random data sets, taking each group of random data sets as a training set of a kernel extreme learning machine model, and training the kernel extreme learning machine;
(4) Predicting the original measured data set by using each trained kernel limit learning machine model to form a plurality of prediction outputs, and solving the average value and the variance of the prediction outputs;
(5) Forming a new kernel limit learning machine model according to the predicted output, and approximating the observation noise variance of the top oil temperature;
(6) Acquiring input data of a verification kernel limit learning machine model to form a verification set, carrying out normalization processing on the verification set, predicting the verification set by each trained kernel limit learning machine model, and solving the average value and the variance of the verification set;
(7) Predicting the observation noise variance of the top oil temperature of the verification set by using a new kernel limit learning machine model, calculating a prediction interval on a set confidence level, and performing inverse normalization processing on a prediction result;
the describedIn the step (3), the training method of the kernel limit learning machine model specifically comprises the following steps: training samples are D = { (x) n ,y n ) N =1,2 \8230n }, and input data x n And target data y n And calculating the connection weight beta of the hidden layer and the output layer of the kernel limit learning machine model according to the following formula:
wherein I is a diagonal matrix, C is a penalty coefficient, y outputs a target vector, y = [ y ] 1 ,y 2 ,…,y n ] T ,n＝1,2…N，K(x i ,x j ) For kernel functions, set to RBF kernel:
K(x i ,x j )＝exp(-γ||x i ,x j || 2 )
wherein, γ>, 0, as a kernel parameter, | | x i -x j | | is the Euclidean norm;
in the step (7), the upper and lower limit values of the prediction interval of the model at the (1-alpha)% confidence level are calculated, and the specific method is as follows:
in the formulaAs a function of the t distribution of the degrees of freedom dfQuantile, degree of freedom df is taken as L.
2. The method for predicting the top-layer oil temperature interval of the power transformer as claimed in claim 1, wherein: in the step (1), the original measured data set comprises a plurality of data pairs, each data pair comprises input data and target data, and the input data is a vector consisting of the load current and the ambient temperature of the transformer at the current moment, and the load current, the ambient temperature and the top oil temperature of the transformer at the previous two sampling moments; the target data is the measured value of the top oil temperature of the transformer at the current moment.
3. The method for predicting the top-layer oil temperature interval of the power transformer as claimed in claim 1, wherein: in the step (2), the Bootstrap method is adopted to carry out random sampling on the original data set for L times with replacement, and finally L groups of data sets are obtained.
4. The method for predicting the oil temperature interval at the top layer of the power transformer as claimed in claim 1, wherein: in the step (2), the number of samples collected each time is the same as the number of sample data pairs in the original data set.
5. The method for predicting the top-layer oil temperature interval of the power transformer as claimed in claim 1, wherein: in the step (5), a new training set D is constructed new ＝{(x n ,r 2 (x n ) N =1,2 \ 8230n }, where the residual r 2 (x n ) As shown in the following formula:
wherein, when the nth data point { x n ,y n Sampling at Bootstrap to get the l training set D l In the middle, q bn =0; otherwise, q bn =1, the predicted output of the L +1 th kernel-limit learning model is g (x), which is an estimate of the observed noise variance σ 2 ∈ (x) for the top oil temperature.
6. The method for predicting the oil temperature interval at the top layer of the power transformer as claimed in claim 1, wherein: and (4) the verification set in the step (6) is a vector consisting of the load current and the ambient temperature of the transformer at the current moment, and the load current, the ambient temperature and the top-layer oil temperature of the transformer at the previous two sampling moments.
7. The method for predicting the oil temperature interval at the top layer of the power transformer as claimed in claim 1, wherein: in the step (6), the data of the verification set is in the same format as the data of the original measured data set.
8. A top layer oil temperature interval prediction system of a power transformer is characterized in that: the method comprises the following steps:
the system comprises an original measured data set acquisition module, a data processing module and a data processing module, wherein the original measured data set acquisition module is used for acquiring load current, ambient temperature and top oil temperature of a power transformer to form an original measured data set;
the method comprises the following steps that a Bootstrap method sub-training set generation module performs repeated replaced random sampling on an original measured data set by using a Bootstrap method to form a plurality of groups of random data sets, and each group of random data sets is used as a kernel limit learning machine model to be trained;
the top-layer oil temperature prediction module is used for predicting the original measured data set by utilizing each trained kernel limit learning machine model to form a plurality of prediction outputs and solving the average value and the variance of the prediction outputs;
the noise variance training set generation module is used for predicting the verification set by each trained kernel limit learning machine model, calculating to obtain a residual value according to a prediction result, and generating a new kernel limit learning machine model training set for predicting the top oil temperature observation noise variance;
the noise variance KELM prediction module is used for training by adopting a training set of the noise variance training set generation module and predicting the observation noise variance of the top oil temperature;
the verification set data acquisition module is used for acquiring input data of the verification kernel limit learning machine model to form a verification set;
the normalization module is used for normalizing the original measured data set and the verification set;
the prediction interval calculation module is used for calculating the average value and the variance of the top oil temperature kernel limit learning machine model according to the prediction value of the top oil temperature kernel limit learning machine model, predicting the noise variance by adopting the noise variance kernel limit learning machine model and finally calculating the upper limit value and the lower limit value of the prediction interval;
and the inverse normalization module is used for carrying out inverse normalization processing on the upper limit value and the lower limit value of the prediction interval.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610489216.8A CN106126944B (en) | 2016-06-28 | 2016-06-28 | A kind of power transformer top-oil temperature interval prediction method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610489216.8A CN106126944B (en) | 2016-06-28 | 2016-06-28 | A kind of power transformer top-oil temperature interval prediction method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106126944A CN106126944A (en) | 2016-11-16 |
| CN106126944B true CN106126944B (en) | 2018-05-25 |
Family
ID=57284955
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610489216.8A Expired - Fee Related CN106126944B (en) | 2016-06-28 | 2016-06-28 | A kind of power transformer top-oil temperature interval prediction method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106126944B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107392352B (en) * | 2017-06-21 | 2020-10-23 | 广州市香港科大霍英东研究院 | Battery future temperature prediction method and system based on fusion extreme learning machine |
| CN108765192A (en) * | 2018-05-25 | 2018-11-06 | 广东电网有限责任公司 | A kind of transformer upper layer oil temperature prediction technique, system and equipment based on big data |
| CN109635468B (en) * | 2018-12-18 | 2023-02-03 | 太原理工大学 | Method for predicting stability of angular contact ball bearing retainer |
| CN111831025B (en) * | 2019-04-19 | 2021-07-06 | 宁波奥克斯高科技有限公司 | Oil temperature control method of transformer and transformer using same |
| CN110232240B (en) * | 2019-06-12 | 2020-03-13 | 贵州电网有限责任公司 | Improved transformer top layer oil temperature prediction method |
| CN111461922B (en) * | 2020-04-02 | 2023-04-21 | 国网冀北电力有限公司唐山供电公司 | Real-time prediction method for hot spot temperature of transformer based on extreme learning machine |
| CN112100574B (en) * | 2020-08-21 | 2024-10-29 | 西安交通大学 | AAKR model uncertainty calculation method and system based on resampling |
| CN113901731B (en) * | 2021-12-10 | 2022-03-15 | 四川瑞康智慧能源有限公司 | Electric quantity prediction method, device, medium and equipment |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102928700A (en) * | 2012-10-24 | 2013-02-13 | 上海市电力公司 | Method for dynamically monitoring state of power transformer in real time |
| CN104919380A (en) * | 2012-11-19 | 2015-09-16 | Abb技术有限公司 | Profiling transformer of power system |
| CN105046374A (en) * | 2015-08-25 | 2015-11-11 | 华北电力大学 | Power interval predication method based on nucleus limit learning machine model |
-
2016
- 2016-06-28 CN CN201610489216.8A patent/CN106126944B/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102928700A (en) * | 2012-10-24 | 2013-02-13 | 上海市电力公司 | Method for dynamically monitoring state of power transformer in real time |
| CN104919380A (en) * | 2012-11-19 | 2015-09-16 | Abb技术有限公司 | Profiling transformer of power system |
| CN105046374A (en) * | 2015-08-25 | 2015-11-11 | 华北电力大学 | Power interval predication method based on nucleus limit learning machine model |
Non-Patent Citations (2)
| Title |
|---|
| 变压器顶层油温预测热模型影响因素分析及其改进;陈伟根;《高电压技术》;20110630;第37卷(第6期);第1329页－1335页 * |
| 基于T-S模型的电力变压器顶层油温预测研究;熊浩等;《中国电机工程学报》;20071031;第27卷(第30期);第15页－第19页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106126944A (en) | 2016-11-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106126944B (en) | A kind of power transformer top-oil temperature interval prediction method and system | |
| CN106055888B (en) | Transformer top-oil temperature prediction technique based on predicted error amendment and device | |
| US20190265768A1 (en) | Method, system and storage medium for predicting power load probability density based on deep learning | |
| CN104392390B (en) | A kind of secondary equipment of intelligent converting station appraisal procedure based on TOPSIS models | |
| CN104569844B (en) | Valve-regulated sealed lead-acid batteries health status monitoring method | |
| CN110222897A (en) | A kind of distribution network reliability analysis method | |
| CN116150897A (en) | Machine tool spindle performance evaluation method and system based on digital twin | |
| CN112115648B (en) | Transformer top layer oil temperature prediction method based on improved deep learning method | |
| CN112069727B (en) | Intelligent transient stability evaluation system and method with high reliability for power system | |
| CN112883497B (en) | Space valve reliability assessment method based on multi-source information fusion | |
| CN110942084A (en) | Loss reduction measure making method based on synchronous line loss abnormity identification | |
| CN108510147B (en) | Electric energy quality comprehensive evaluation method based on residual error fluctuation model | |
| CN106251027A (en) | Electric load probability density Forecasting Methodology based on fuzzy support vector quantile estimate | |
| CN113791351B (en) | Lithium battery life prediction method based on transfer learning and difference probability distribution | |
| CN115564310A (en) | Reliability evaluation method for new energy power system based on convolutional neural network | |
| CN104008433A (en) | Method for predicting medium-and-long-term power loads on basis of Bayes dynamic model | |
| CN115508770A (en) | KL-NB algorithm-based electric energy meter operation state online evaluation method | |
| CN114897331A (en) | Power transformer risk assessment method based on three-parameter interval gray number decision | |
| CN110674984A (en) | Tri-Training-Lasso-BP network-based static voltage stability margin prediction method | |
| CN110413657A (en) | Average response time appraisal procedure towards seasonal form non-stationary concurrency | |
| CN114280490A (en) | Lithium ion battery state of charge estimation method and system | |
| CN116736214A (en) | CVT stability evaluation method and system | |
| CN110414086B (en) | Sensitivity-based comprehensive stress acceleration factor calculation method | |
| CN115840119A (en) | Power cable line degradation diagnostic system and method using database samples | |
| CN103928923B (en) | A kind of network stationary power quality method for early warning based on sensitivity analysis |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20180525 Termination date: 20200628 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |