CN110045237A - Transformer state parametric data prediction technique and system based on drosophila algorithm optimization - Google Patents
Transformer state parametric data prediction technique and system based on drosophila algorithm optimization Download PDFInfo
- Publication number
- CN110045237A CN110045237A CN201910276198.9A CN201910276198A CN110045237A CN 110045237 A CN110045237 A CN 110045237A CN 201910276198 A CN201910276198 A CN 201910276198A CN 110045237 A CN110045237 A CN 110045237A
- Authority
- CN
- China
- Prior art keywords
- transformer state
- transformer
- parametric data
- data
- parameter
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01D—MEASURING NOT SPECIALLY ADAPTED FOR A SPECIFIC VARIABLE; ARRANGEMENTS FOR MEASURING TWO OR MORE VARIABLES NOT COVERED IN A SINGLE OTHER SUBCLASS; TARIFF METERING APPARATUS; MEASURING OR TESTING NOT OTHERWISE PROVIDED FOR
- G01D21/00—Measuring or testing not otherwise provided for
- G01D21/02—Measuring two or more variables by means not covered by a single other subclass
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/12—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing
- G01R31/1227—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing of components, parts or materials
- G01R31/1263—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing of components, parts or materials of solid or fluid materials, e.g. insulation films, bulk material; of semiconductors or LV electronic components or parts; of cable, line or wire insulation
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/12—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing
- G01R31/1227—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing of components, parts or materials
- G01R31/1263—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing of components, parts or materials of solid or fluid materials, e.g. insulation films, bulk material; of semiconductors or LV electronic components or parts; of cable, line or wire insulation
- G01R31/1281—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing of components, parts or materials of solid or fluid materials, e.g. insulation films, bulk material; of semiconductors or LV electronic components or parts; of cable, line or wire insulation of liquids or gases
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The transformer state parametric data prediction technique based on drosophila algorithm optimization that the invention discloses a kind of, it is the following steps are included: S100: obtaining the transformer state amount data in a period of time, and it is converted into the transformer state moment matrix of matrix form, the transformer state amount includes the related data of transformer state parameter;S200: building transformer state parametric data prediction model is acquired the hyper parameter of the prediction model based on drosophila algorithm, is trained based on the transformer state moment matrix to the prediction model;S300: transformer state parametric data is predicted based on the transformer state parametric data prediction model through step S200 training.This method is avoided that hyper parameter selection falls into local convergence, to promote prediction model training effectiveness, guarantees higher transformer state parametric data predictablity rate and reliability.In addition, the invention also discloses the transformer state parametric data forecasting systems accordingly based on drosophila algorithm optimization.
Description
Technical field
The present invention relates in electric system power transmission and transformation equipment operation maintenance area transformer state parameter prediction method and
System more particularly to a kind of transformer state parameter prediction method and system based on drosophila algorithm optimization.
Background technique
Essential core equipment of the power transformer as power transmission and transformation system guarantees that it can be run healthy and stablely with weight
Want meaning.Usually pass through monitoring running state of transformer and predict its variation tendency, to effectively prevent transformer fault, carry out
Prediction scheme guarantees the stable operation of transformer.In order to effectively monitor running state of transformer and predict its variation tendency, it usually needs
Each transformer state parameter of monitoring reflection running state of transformer simultaneously predicts it.
The state parameter information dimension of transformer is very high, and data volume is huge, and different state parameter data are from different perspectives
The operating status of transformer is reflected to a certain extent.The state parameter data of transformer be mainly included in line monitoring data,
Test experiment data, operation inspection data and transformer self-technique supplemental characteristic etc., these data can be in all its bearings
The operating status for reflecting transformer, carrying out prediction to running state of transformer using these data has important research significance.
Transformer online monitoring state parameter data are a complicated data sequences, not to transformer online monitoring state parameter data
The variation tendency come is predicted, the operating status variation of transformer can be better anticipated.
Traditional transformer state Parameter Prediction Models only only account for unitary variant or the data of a few variable
The variation tendency in sequence future is judged.And when being fitted to historical data, more remote go through can not be retained
Effect of the history information to current time, to cannot achieve the prediction to the following long period scale.
Generalized regression nerve networks (generalized regression neural network, GRNN) are by the U.S.
The novel radial base neural net of one kind that scholar Donld F.Specht is proposed.It is based on nonlinear regression analysis
Radial base neural net.The network has the ability to predict any form of function from historical data, and finally converges on sample
Amount accumulates most optimized regression faces.It is similar with multi-layered perception neural networks in structure, including input layer, mode layer, summation
Layer and output layer.Generalized regression nerve networks are not needing to adjust the connection between each neuron in the training process of network
Weight, the learning process of network depend only on data sample itself, and what model uniquely needed to set is smoothing factor parameter, very
Artificial Subjective Factors are reduced in big degree, are therefore particularly suitable for the transformer state parameter number with nonlinear characteristic
According to prediction.
Prediction model based on neural network building can be used for the prediction of transformer state parametric data, but there is super ginseng
Number chooses the problem of being easily trapped into local convergence, causing prediction model training effectiveness low, influence predictablity rate and reliability.
Summary of the invention
An object of the present invention is to provide a kind of transformer state parametric data prediction side based on drosophila algorithm optimization
Method, this method are avoided that hyper parameter selection falls into local convergence, to promote prediction model training effectiveness, guarantee higher transformation
Device state parameter data predictablity rate and reliability.
According to foregoing invention purpose, the invention proposes a kind of transformer state parametric datas based on drosophila algorithm optimization
Prediction technique is predicted transformer state parametric data, be the described method comprises the following steps:
S100: the transformer state amount data in a period of time are obtained, and are converted into the transformer shape of matrix form
State moment matrix, the transformer state amount include the related data of transformer state parameter;
S200: building transformer state parametric data prediction model acquires the super of the prediction model based on drosophila algorithm
Parameter is trained the prediction model based on the transformer state moment matrix;
S300: transformer state parameter is predicted based on the transformer state parametric data prediction model through step S200 training
Data.
Transformer state parametric data prediction technique proposed by the present invention based on drosophila algorithm optimization, uses drosophila to calculate
Method acquires the hyper parameter of prediction model, falls into local convergence to avoid hyper parameter selection, so that prediction model training effectiveness is promoted,
Guarantee higher transformer state parametric data predictablity rate and reliability.The parameter of setting value before hyper parameter is training,
Rather than the parameter value obtained by training.Under normal conditions, it needs to optimize hyper parameter, selects one group of optimal super ginseng
Number, to improve the efficiency and effect of training.Drosophila algorithm (Fruit Fly Optimization Algorithm, FOA) be by
A kind of optimization algorithm that TaiWan, China scholar Wen-Tsao Pan is proposed, the thinking of the algorithm are the behaviors simulating drosophila and looking for food:
First with the general orientation of smell discovery things, the specific location of things then is found simultaneously with vision in the case where short distance again
Close to the process looked for food.Drosophila algorithm has good global optimization performance, therefore the prediction model acquired based on drosophila algorithm
Hyper parameter, be avoided that hyper parameter selection fall into local convergence, to promote prediction model training effectiveness, guarantee higher transformation
Device state parameter data predictablity rate and reliability.
The usual prediction model is constructed based on neural network.
Further, in the transformer state parametric data prediction technique of the present invention based on drosophila algorithm optimization,
Transformer state parametric data prediction model described in step S200 is constructed based on generalized regression nerve networks, and the hyper parameter is
Smoothing factor in the generalized regression nerve networks.
In above scheme, generalized regression nerve networks are the radial base neural nets based on nonlinear regression analysis.
The network has the ability to predict any form of function from historical data, and finally converges on sample size and accumulate most optimization
Regression aspect;Generalized regression nerve networks are in the power for not needing to adjust the connection between each neuron in the training process of network
Weight, the learning process of network depend only on data sample itself, and what model uniquely needed to set is smoothing factor parameter, very greatly
Artificial Subjective Factors are reduced in degree, are therefore particularly suitable for the transformer state parametric data with nonlinear characteristic
Prediction.In addition, generalized regression nerve networks can retain the effect of historical information more remote to current time, so as to
To realize the prediction to the following long period scale.The smoothing factor to be arranged only for the generalized regression nerve networks
One hyper parameter.
Further, in the transformer state parametric data prediction technique of the present invention based on drosophila algorithm optimization,
Drosophila algorithm described in step S200 initializes the position of individual in population using dynamic step length.
Tradition side's drosophila algorithm initializes the position of individual in population using fixed step size.Fixed step size is asked there are following
Topic: if the step-length of setting is excessive, the search capability that will lead to algorithm dies down, and the search time of cost is too long, leads to algorithm
Efficiency it is relatively low;If the step-length being arranged is too small, algorithm is easily trapped into local optimum.Therefore above scheme considers dynamically to walk
It is long to replace traditional fixed step size, to avoid the problem that above-mentioned fixed step size exists.
Further, in the transformer state parametric data prediction technique of the present invention based on drosophila algorithm optimization,
Drosophila algorithm described in step S200 uses cross-validation method, and population is divided into multiple equal sub- populations, is then carried out respectively
Optimization analysis, final reselection optimal solution.
Above scheme ensure that algorithm can make full use of data, prevent algorithm from falling into locally optimal solution.
Further, in the transformer state parametric data prediction technique of the present invention based on drosophila algorithm optimization,
The prediction model is trained using error backpropagation algorithm in step S200.
In above scheme, the structural parameters of the prediction model generally include the hidden layer number of plies of generalized regression network, mind
Through first number of nodes, the error backpropagation algorithm is determined for the structural parameters of the prediction model.
Further, in the transformer state parametric data prediction technique of the present invention based on drosophila algorithm optimization,
The related data of transformer state parameter described in step S100 includes the ratio number between the data and/or parameter of parameter itself
According to.
In above scheme, in order to further ensure prediction effect, it is often desirable that Donna enters the transformer state of some dimensions
Amount, i.e. quantity of state range of choice are not limited to state parameter, can also include the related datas such as the ratio between state parameter.
Further, in the transformer state parametric data prediction technique of the present invention based on drosophila algorithm optimization,
Transformer state parameter described in step S100 includes transformer body parameter and substation parameter.
In above scheme, since running state of transformer is not only influenced by transformer body parameter, also by substation's ring
The influence of border parameter, therefore transformer body parameter and substation parameter are included in the selection model of transformer state parameter
It encloses.
Further, in the above-mentioned transformer state parametric data prediction technique based on drosophila algorithm optimization, the change
Depressor ontology parameter includes the gas content and/or temperature of oil in transformer dissolved in transformer oil.
In above scheme, the gas content and temperature of oil in transformer dissolved in transformer oil can reflect change to a certain extent
The degree of depressor insulation ag(e)ing or failure, therefore the gas content dissolved in transformer oil and temperature of oil in transformer are included in transformer
The range of choice of ontology parameter.
Further, in the above-mentioned transformer state parametric data prediction technique based on drosophila algorithm optimization, the change
Power station environment parameter includes one of temperature, surface humidity, relative humidity, mean wind speed or a variety of.
In above scheme, the factors such as temperature, surface humidity, relative humidity, mean wind speed can cause the performance of transformer
It influences, it is considered to be the correlative factor of transformer state, therefore by factors such as temperature, surface humidity, relative humidity, mean wind speeds
It is included in the range of choice of transformer body parameter.
Further, in the above-mentioned transformer state parametric data prediction technique based on drosophila algorithm optimization, the change
The gas dissolved in depressor oil includes H2、CO、CH4、C2H2、C2H4、C2H6One of or it is a variety of.
In above scheme, associated arguments can be obtained based on dissolved gas analysis (DGA) corresponding chromatographic data
Initial data.It includes hydrogen (H that above scheme, which picks up the gaseous species selected,2), carbon monoxide (CO), methane (CH4), acetylene
(C2H2), ethylene (C2H4) and ethane (C2H6)。
Further, in the above-mentioned transformer state parametric data prediction technique based on drosophila algorithm optimization, it is based on institute
The gas dissolved in transformer oil is stated, the ratio data between the parameter includes CH4/H2、 C2H2/C2H4、C2H4/C2H6、C2H6/
CH4、C2H2/CH4、C2H6/C2H2One of or it is a variety of.
In above scheme, in order to expand the dimension of input vector, uses for reference and utilized dissolved gas analysis (DGA) ratio
The thinking that value diagnoses transformer fault, by common International Electrotechnical Commission (IEC) ratio, Rogers ratio and
Dornenburg ratio is also included in the range of choice of input vector, i.e., by CH4/H2、C2H2/C2H4、C2H4/C2H6、C2H6/CH4、
C2H2/CH4、C2H6/C2H26 groups of DGA ratios are also included in the ratio data range of choice between the parameter.
It is a further object of the present invention to provide a kind of, and the transformer state parametric data based on drosophila algorithm optimization predicts system
System, the system can be avoided that hyper parameter selection falls into local convergence, to promote prediction model training effectiveness, guarantee higher change
Depressor state parameter data predictablity rate and reliability.
According to foregoing invention purpose, the invention proposes a kind of transformer state parametric datas based on drosophila algorithm optimization
Forecasting system, the system include the data acquisition module and data processing module of data connection, are joined using above-mentioned transformer state
Any one method measured in data predication method predicts transformer state parametric data.
Transformer state parametric data forecasting system proposed by the present invention based on drosophila algorithm optimization, by using upper
It states any transformer state parametric data prediction technique to predict transformer state parametric data, therefore, according to aforementioned original
Reason, the system are avoided that hyper parameter selection falls into local convergence, to promote prediction model training effectiveness, guarantee higher transformation
Device state parameter data predictablity rate and reliability.
Transformer state parametric data prediction technique of the present invention based on drosophila algorithm optimization can be avoided that super ginseng
Number is chosen and falls into local convergence, to promote prediction model training effectiveness, guarantees higher transformer state parametric data prediction
Accuracy rate and reliability.The method of the present invention use drosophila algorithm optimization and prediction model combination, have considerable flexibility and
Expansion, can be with the suitable prediction model of unrestricted choice.The method of the present invention is experiments verify that with preferable capability of fitting and in advance
Survey ability.
More than transformer state parametric data forecasting system of the present invention based on drosophila algorithm optimization equally has
Advantages and beneficial effects.
Detailed description of the invention
Fig. 1 is that the process of the transformer state parametric data prediction technique of the present invention based on drosophila algorithm optimization is shown
It is intended to.
Fig. 2 be it is of the present invention based on the transformer state parametric data prediction technique of drosophila algorithm optimization in a kind of reality
Apply the flow diagram under mode.
Fig. 3 is the root-mean-square error R for verifying smoothing factor optimization process in examplemseChange curve schematic diagram.
Fig. 4 is concentration of methane gas prediction result percentage error schematic diagram in verifying example.
Specific embodiment
Below in conjunction with Figure of description and specific embodiment to the change of the present invention based on drosophila algorithm optimization
Depressor state parameter data predication method and system are described in further detail.
Fig. 1 illustrates the process of the transformer state parametric data prediction technique based on drosophila algorithm optimization.
As shown in Figure 1, the process of the transformer state parametric data prediction technique of the invention based on drosophila algorithm optimization
Include:
S100: the transformer state amount data in a period of time are obtained, and are converted into the transformer shape of matrix form
State moment matrix, the transformer state amount include the related data of transformer state parameter;
S200: building transformer state parametric data prediction model acquires the super of the prediction model based on drosophila algorithm
Parameter is trained the prediction model based on the transformer state moment matrix;
S300: transformer state parameter is predicted based on the transformer state parametric data prediction model through step S200 training
Data.
In some embodiments, transformer state parametric data prediction model is based on general regression neural in step S300
Network struction, the hyper parameter are the smoothing factor in the generalized regression nerve networks.
In some embodiments, drosophila algorithm described in step S200 is using dynamic step length initialization individual in population
Position.
In some embodiments, drosophila algorithm described in step S200 uses cross-validation method, population is divided into multiple
Then equal sub- population optimizes analysis, final reselection optimal solution respectively.
In some embodiments, prediction model is trained using error backpropagation algorithm in step S200.
In some embodiments, in step S100 the related data of transformer state parameter include parameter itself data
And/or the ratio data between parameter.
In some embodiments, transformer state parameter includes transformer body parameter and substation's ring in step S100
Border parameter.
Wherein, in some embodiments, transformer body parameter include the gas content dissolved in transformer oil and/or
Temperature of oil in transformer;In some embodiments, substation parameter includes temperature, surface humidity, relative humidity, mean wind speed
One of or it is a variety of.
Wherein, in some embodiments, the gas dissolved in transformer oil includes H2、CO、CH4、C2H2、 C2H4、C2H6
One of or it is a variety of.
Wherein, in some embodiments, based on the gas dissolved in transformer oil, the ratio data between parameter includes
CH4/H2、C2H2/C2H4、C2H4/C2H6、C2H6/CH4、C2H2/CH4、C2H6/C2H2One of or it is a variety of.
Fig. 2 illustrates the transformer state parametric data prediction technique based on drosophila algorithm optimization in one embodiment
Process.
As shown in Fig. 2, the transformer state parametric data prediction technique of the present invention based on drosophila algorithm optimization exists
A kind of process under embodiment includes the following steps 1- step 6:
Step 1: the transformer online monitoring state quantity data in a period of time is obtained by data acquisition module.
The step transformer online monitoring state quantity data interior for a period of time by data collecting module collected, wherein
Transformer online monitoring quantity of state includes ratio between transformer state parameter and parameter.Wherein, transformer state parameter specifically wraps
Include the isallobaric device ontology parameter of the gas content dissolved in transformer oil, temperature of oil in transformer and temperature, surface temperature, opposite
The substations parameter such as humidity and mean wind speed.Wherein, the gas dissolved in transformer oil includes hydrogen (H2), carbon monoxide
(CO), methane (CH4), acetylene (C2H2), ethylene (C2H4) and ethane (C2H6), ratio is using in 6 groups of transformer oil between parameter
The gas content ratio of dissolution indicates respectively CH in the form of the ratio between molecular formula4/H2, C2H2/C2H4, C2H4/C2H6, C2H6/
CH4, C2H2/CH4, C2H6/C2H2。
Step 2: transformer online monitoring state quantity data is converted to by transformer online monitoring by data processing module
State moment matrix X.
The step by data processing module using deviation standardized method to transformer online monitoring state quantity data into
Row normalized obtains transformer online monitoring state moment matrix X:
Wherein, X1、X2And XrIndicate that each transformer online monitoring quantity of state, subscript 1,2 ... n indicate time series.
Step 3: transformer state parametric data prediction model is constructed by data processing module.
In the step, transformer state parametric data prediction model is constructed based on generalized regression nerve networks, and by more
Secondary repetition test chooses optimal result to determine the structure of prediction model.Wherein, the structural parameters of prediction model include that broad sense is returned
Return the hidden layer number of plies, the neuron node number of network.
Generalized regression nerve networks are made of input layer, mode layer, summation layer and output layer.Generalized regression nerve networks
The characteristics of basic thought is calculated using nonlinear regression, probability theory of the Inspiration Sources in Statistics.For
One network, when being approached with function, network output is exactly regression result of the network for input vector.If f (x, y)
It is the probability density function of stochastic variable x and y, wherein x is the input vector of p dimension, and y is corresponding output vector.Known X be with
One group of measured value of machine variable x, then regressand value of the y relative to X are as follows:
It is also the prediction output of corresponding y under conditions of input is X.
When probability density function f (X, y) is unknown, need to just the method using Parzen non-parametric estmation be used, utilize acquisition
The sample data arrivedDensity function estimation
In formulaxi, yiIt is stochastic variable
The sample observations of x and y;N is the capacity of sample;σ is the spread factor of Gaussian function, also referred to as smoothing factor or extension ginseng
Number.
By formula (2)Instead of the f (X, y) of (1), while the sequence of exchange integral and adduction, it can obtain:
BecauseSo formula (3) is with further abbreviation are as follows:
By formula (4) it is found that when constructing GRNN network, network needs only one parameter of hyper parameter σ of training to need to set.
When smoothing factor σ is intended to infinity, d (X, xi) it is intended to zero.ToRepresent the mean value of all sample dependent variables.
When smoothing factor σ is intended to zero,It is then very close with training sample.At this time if the point of prediction is in training sample set
When in conjunction, predicted value can and sample desired value it is very close, but be once new data, be fail in sample comprising into
The point gone, then prediction effect may be excessively poor, is unable to satisfy prediction and requires.So only when smoothing factor σ value is suitable,
The dependent variable of all training samples could be fully considered into, with future position dependent variable corresponding to the close sample point
Also it has been attached bigger weight.Traditional mode is that smoothing factor σ is obtained by cross-validation method, is mentioned to a certain extent
The high approximation capability and classification capacity of network.The present invention then comes in step 4 below in conjunction with improved drosophila optimization algorithm
Smoothing factor σ is chosen, numerical results show that this method has better global convergence, improve prediction during prediction
Precision and reliability.
Step 4: acquiring the smoothing factor of prediction model based on drosophila algorithm by data processing module.
Drosophila algorithm steps are as follows:
Step 4-1: in each iteration, smoothing factor and model judging quota root-mean-square error value are as a pair of of ginseng
Number, carries out random initializtion Init_X, Init_Y to it.
Step 4-2: the position of individual in population is initialized:
xi=Init_X+l*r (5)
yi=Init_Y+l*r (6)
In formula, l indicates step value, and r is the random number in [0,1] section, xi,yiFor individual position coordinates.
Step 4-3: individual and origin distance d are calculatedi, and seek decision content pi.。
pi=1/di (8)
Step 4-4: optimum individual in group is found out.
[bestS, bestIndex]=max (p) (9)
Step 4-5: decision content and new coordinate are updated.
BestSmell=bestS (10)
X=x (bestIndex) (11)
Y=y (bestIndex) (12)
BestSmell is the optimal smoothing factor parameter of model in formula.
Since in step 4-2, the step-length of conventional method is fixed.If the step-length of setting is excessive, calculation will lead to
The search capability of method dies down, and the search time of cost is too long, causes the efficiency of algorithm relatively low;If the step-length being arranged is too small,
Algorithm is easily trapped into local optimum.Secondly traditional drosophila algorithm has that convergence precision is low.The present invention is to step value
Selection improves, and replaces fixed step size using dynamic step length.The stepsize formula of optimization is as follows:
In formula, kiIt is current the number of iterations, k is maximum number of iterations, l0It is initial step length, liIt is current step.
And the problem low for convergence precision, present invention employs the methods of cross validation, population are divided into multiple equal
Sub- population, then optimize analysis, the best solution of final reselection respectively.The program ensure that algorithm can make full use of
Data prevent algorithm from falling into locally optimal solution.
Step 5: transformer online monitoring state moment matrix X being based on by data processing module, prediction model is instructed
Practice.
In the step, by before the transformer online monitoring quantity of state in transformer online monitoring state moment matrix X 80%
Data are trained prediction model using error backpropagation algorithm, determine the structural parameters of the prediction model as input.
Step 6: housebroken transformer state parametric data prediction model being based on by data processing module and predicts transformation
Device state parameter data.
In the step, by remaining 20% transformer online monitoring shape in transformer online monitoring state moment matrix X
State amount inputs trained prediction model as input data, inputs to GRNN neural net layer, and neural net layer output is pre-
Survey result.
Above-mentioned data acquisition module and the mutual data connection of data processing module, constitute the present embodiment based on drosophila algorithm
The transformer state parametric data forecasting system of optimization.The system is using above-mentioned transformer state parametric data prediction technique to change
Depressor state parameter data are predicted.
Test verifying is carried out to above-described embodiment below.
Fig. 3 illustrates the root-mean-square error R of smoothing factor optimization process in this verifying examplemseChange curve,
Fig. 4 illustrates concentration of methane gas prediction result percentage error in this verifying example.
This verifying example is using the transformer state parametric data prediction technique and system of above-described embodiment to transformer shape
State property data are predicted.Existed based on certain 220kV transformer oil chromatographic on-Line Monitor Device oil colours Data acquisition transformer
Line monitoring state amount data, the sampling interval of data are 1 day.Using 400 groups of monitoring data as training sample, by 30 groups of monitoring numbers
According to as test sample.
In order to evaluate the accuracy and validity of combination forecasting proposed by the present invention, using following interpretational criteria
To be analyzed:
The true value of test set and the root-mean-square error value R of predicted valuemse, expression formula are as follows:
The average percentage error of true value and predicted value, expression formula are as follows:
Maximum percentage error, expression formula are as follows:
In formula, N is the number of test set data, xiIt is true value,It is predicted value.
This verifying example is with methane CH4Illustrate the process entirely predicted for forecasting of Gas Concentration result.It inputs first
Transformer online monitoring quantity of state vector is successively are as follows: CH4(only history concentration), H2、CO、C2H2、C2H4、C2H6, it is total hydrocarbon, total flammable
Gas concentration, C2H2/C2H4、C2H4/C2H6, environment temperature and oil temperature totally 11 groups of monitoring data.These vectors are standardized place
Reason, is mapped between [0,1], transfer function are as follows:
In formula, xminFor the minimum value of vector sample data, xmaxFor the maximum value of vector sample data.
Transformer state parametric data prediction model based on generalized regression nerve networks building.
The smoothing factor σ of prediction model is determined based on improved drosophila algorithm.The root mean square of smoothing factor optimization process misses
Poor RmseChange curve is finally restrained in the 44th iteration, R at this time as shown in figure 3, by 100 iterationmseIt is worth minimum
The value of 0.0105, corresponding smoothing factor σ are 0.0867.
The transformer state parametric data prediction model constructed based on generalized regression nerve networks is trained based on above-mentioned vector,
Determine the hidden layer number of plies, the neuron node number of model.
Concentration of methane gas is predicted based on housebroken transformer state parametric data prediction model, true value and
The percentage error of predicted value is as shown in figure 4, abscissa is time series, and unit is day, and ordinate is percentage error,
Unit is %, and average percentage error is 0.55% in figure, maximum percentage error 1.32%.
By prediction technique of the invention with traditional support vector machines (SVM), Feedback Neural Network (BPNN) prediction technique
Result be compared.The transformer state parametric data of prediction is concentration of methane gas data, and the number of test set is 30, in advance
It is as shown in table 1 to survey effect.
Influence of the different prediction techniques of table 1. to prediction effect
Prediction technique | Mean error % | Worst error % |
The method of the present invention | 0.55 | 1.32 |
SVM | 3.15 | 4.31 |
BPNN | 1.86. | 2.36 |
As it can be seen from table 1 in the identical situation of sample number of test set, it is proposed by the present invention excellent based on drosophila algorithm
The transformer state parametric data prediction technique and system prediction result of change are more accurate, preferable to the fitting of data.
Different test set numbers will also result in certain influence to the accuracy of prediction.Respectively to 10,30,60,90
Data point is predicted that prediction result is as shown in table 2.
Influence of the different test set numbers of samples of table 2. to prediction effect
As can be drawn from Table 2: when test set number of samples is smaller, prediction model proposed by the present invention is more several than other
The accuracy rate of model prediction wants high.With increasing for test set number of samples, the prediction accuracy of several models is all declined,
But prediction technique proposed by the present invention is substantially better than remaining two kinds of model in estimated performance.
It should be noted that the above list is only specific embodiments of the present invention, it is clear that the present invention is not limited to implement above
Example has many similar variations therewith.If those skilled in the art directly exports or joins from present disclosure
All deformations expected, are within the scope of protection of the invention.
Claims (12)
1. a kind of transformer state parametric data prediction technique based on drosophila algorithm optimization, to transformer state parametric data
It is predicted, it is characterised in that the described method comprises the following steps:
S100: the transformer state amount data in a period of time are obtained, and are converted into the transformer state amount of matrix form
Matrix, the transformer state amount include the related data of transformer state parameter;
S200: building transformer state parametric data prediction model acquires the hyper parameter of the prediction model based on drosophila algorithm,
The prediction model is trained based on the transformer state moment matrix;
S300: transformer state parameter number is predicted based on the transformer state parametric data prediction model through step S200 training
According to.
2. the transformer state parametric data prediction technique based on drosophila algorithm optimization, feature exist as described in claim 1
In transformer state parametric data prediction model described in step S200 is constructed based on generalized regression nerve networks, the super ginseng
Number is the smoothing factor in the generalized regression nerve networks.
3. the transformer state parametric data prediction technique based on drosophila algorithm optimization, feature exist as described in claim 1
In drosophila algorithm described in step S200 initializes the position of individual in population using dynamic step length.
4. the transformer state parametric data prediction technique based on drosophila algorithm optimization, feature exist as described in claim 1
Use cross-validation method in, drosophila algorithm described in step S200, population be divided into multiple equal sub- populations, then respectively into
Row optimization analysis, final reselection optimal solution.
5. the transformer state parametric data prediction technique based on drosophila algorithm optimization, feature exist as described in claim 1
In being trained to the prediction model in step S200 using error backpropagation algorithm.
6. the transformer state parametric data prediction technique based on drosophila algorithm optimization, feature exist as described in claim 1
In the related data of transformer state parameter described in step S100 includes the ratio between the data and/or parameter of parameter itself
Data.
7. the transformer state parametric data based on drosophila algorithm optimization as described in any claim in claim 1-6 is pre-
Survey method, which is characterized in that transformer state parameter described in step S100 includes transformer body parameter and substation
Parameter.
8. the transformer state parametric data prediction technique based on drosophila algorithm optimization, feature exist as claimed in claim 7
In the transformer body parameter includes the gas content and/or temperature of oil in transformer dissolved in transformer oil.
9. the transformer state parametric data prediction technique based on drosophila algorithm optimization, feature exist as claimed in claim 7
In the substation parameter includes one of temperature, surface humidity, relative humidity, mean wind speed or a variety of.
10. the transformer state parametric data prediction technique based on drosophila algorithm optimization, feature exist as claimed in claim 8
In the gas dissolved in the transformer oil includes H2、CO、CH4、C2H2、C2H4、C2H6One of or it is a variety of.
11. the transformer state parametric data prediction technique based on drosophila algorithm optimization as claimed in claim 10, feature
It is, based on the gas dissolved in the transformer oil, the ratio data between the parameter includes CH4/H2、C2H2/C2H4、C2H4/
C2H6、C2H6/CH4、C2H2/CH4、C2H6/C2H2One of or it is a variety of.
12. a kind of transformer state parametric data forecasting system based on drosophila algorithm optimization, which is characterized in that the system includes
The data acquisition module and data processing module of data connection are joined using the transformer state of any one of claim 1-11
It measures data predication method and predicts transformer state parametric data.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910276198.9A CN110045237A (en) | 2019-04-08 | 2019-04-08 | Transformer state parametric data prediction technique and system based on drosophila algorithm optimization |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910276198.9A CN110045237A (en) | 2019-04-08 | 2019-04-08 | Transformer state parametric data prediction technique and system based on drosophila algorithm optimization |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110045237A true CN110045237A (en) | 2019-07-23 |
Family
ID=67276443
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910276198.9A Pending CN110045237A (en) | 2019-04-08 | 2019-04-08 | Transformer state parametric data prediction technique and system based on drosophila algorithm optimization |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110045237A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110674604A (en) * | 2019-09-20 | 2020-01-10 | 武汉大学 | Transformer DGA data prediction method based on multi-dimensional time sequence frame convolution LSTM |
CN111651936A (en) * | 2020-05-27 | 2020-09-11 | 辽宁工程技术大学 | FOA-GRNN-based modeling design method for ultra-wideband antenna with double trap characteristics |
CN113393107A (en) * | 2021-06-07 | 2021-09-14 | 东方电气集团科学技术研究院有限公司 | Incremental calculation method for state parameter reference value of power generation equipment |
CN113435486A (en) * | 2021-06-15 | 2021-09-24 | 西安科技大学 | Coal gangue identification method based on PCA-IFOA-SVM combined with gray level-texture fusion features |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103942625A (en) * | 2014-04-23 | 2014-07-23 | 天津大学 | Hydropower station plant dam structure vibration response prediction method based on FOA-GRNN |
CN107677473A (en) * | 2017-09-23 | 2018-02-09 | 哈尔滨理工大学 | A kind of GRNN rotating machinery fault Forecasting Methodologies based on FOA optimizations |
CN108390380A (en) * | 2018-05-16 | 2018-08-10 | 国网山东省电力公司青岛供电公司 | A kind of transformer state parameter trend forecasting method and system |
CN108665052A (en) * | 2017-03-30 | 2018-10-16 | 华为技术有限公司 | The method and apparatus of drosophila algorithm optimization generalized regression nerve networks |
CN109164248A (en) * | 2018-09-18 | 2019-01-08 | 山东理工大学 | A kind of predicting model for dissolved gas in transformer oil method |
CN109242205A (en) * | 2018-10-09 | 2019-01-18 | 国网上海市电力公司 | A kind of transformer state prediction technique and system |
-
2019
- 2019-04-08 CN CN201910276198.9A patent/CN110045237A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103942625A (en) * | 2014-04-23 | 2014-07-23 | 天津大学 | Hydropower station plant dam structure vibration response prediction method based on FOA-GRNN |
CN108665052A (en) * | 2017-03-30 | 2018-10-16 | 华为技术有限公司 | The method and apparatus of drosophila algorithm optimization generalized regression nerve networks |
CN107677473A (en) * | 2017-09-23 | 2018-02-09 | 哈尔滨理工大学 | A kind of GRNN rotating machinery fault Forecasting Methodologies based on FOA optimizations |
CN108390380A (en) * | 2018-05-16 | 2018-08-10 | 国网山东省电力公司青岛供电公司 | A kind of transformer state parameter trend forecasting method and system |
CN109164248A (en) * | 2018-09-18 | 2019-01-08 | 山东理工大学 | A kind of predicting model for dissolved gas in transformer oil method |
CN109242205A (en) * | 2018-10-09 | 2019-01-18 | 国网上海市电力公司 | A kind of transformer state prediction technique and system |
Non-Patent Citations (1)
Title |
---|
朱沛恒: "基于果蝇算法优化的概率神经网络在变压器故障诊断中的应用", 《电力大数据》 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110674604A (en) * | 2019-09-20 | 2020-01-10 | 武汉大学 | Transformer DGA data prediction method based on multi-dimensional time sequence frame convolution LSTM |
CN110674604B (en) * | 2019-09-20 | 2022-07-08 | 武汉大学 | Transformer DGA data prediction method based on multi-dimensional time sequence frame convolution LSTM |
CN111651936A (en) * | 2020-05-27 | 2020-09-11 | 辽宁工程技术大学 | FOA-GRNN-based modeling design method for ultra-wideband antenna with double trap characteristics |
CN113393107A (en) * | 2021-06-07 | 2021-09-14 | 东方电气集团科学技术研究院有限公司 | Incremental calculation method for state parameter reference value of power generation equipment |
CN113393107B (en) * | 2021-06-07 | 2022-08-12 | 东方电气集团科学技术研究院有限公司 | Incremental calculation method for state parameter reference value of power generation equipment |
CN113435486A (en) * | 2021-06-15 | 2021-09-24 | 西安科技大学 | Coal gangue identification method based on PCA-IFOA-SVM combined with gray level-texture fusion features |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110045237A (en) | Transformer state parametric data prediction technique and system based on drosophila algorithm optimization | |
Zhang et al. | Sound quality prediction of vehicle interior noise and mathematical modeling using a back propagation neural network (BPNN) based on particle swarm optimization (PSO) | |
CN107274067B (en) | Distribution transformer overload risk assessment method | |
CN110133393A (en) | A kind of electricity consumption monitoring system and method based on non-intruding monitor technology | |
CN103177288A (en) | Transformer fault diagnosis method based on genetic algorithm optimization neural network | |
CN104408562A (en) | Photovoltaic system generating efficiency comprehensive evaluation method based on BP (back propagation) neural network | |
CN110542819A (en) | transformer fault type diagnosis method based on semi-supervised DBNC | |
CN110363334B (en) | Grid line loss prediction method of photovoltaic grid connection based on gray neural network model | |
CN107103184A (en) | A kind of high-voltage cable joint temperature predicting method | |
CN112163371A (en) | Transformer bushing state evaluation method | |
CN110705887A (en) | Low-voltage transformer area operation state comprehensive evaluation method based on neural network model | |
CN112307677A (en) | Power grid oscillation mode evaluation and safety active early warning method based on deep learning | |
CN115758151A (en) | Combined diagnosis model establishing method and photovoltaic module fault diagnosis method | |
CN110045236A (en) | Transformer state parametric data prediction technique and system based on core pivot element analysis optimization | |
CN112446175B (en) | Distributed power supply site selection optimization method based on power distribution network probability load flow calculation | |
CN113379116A (en) | Cluster and convolutional neural network-based line loss prediction method for transformer area | |
CN109378834A (en) | Large scale electric network voltage stability margin assessment system based on information maximal correlation | |
Khalyasmaa et al. | Defining membership functions in power equipment state assessment problem | |
Sun et al. | Prediction-based manufacturing center self-adaptive demand side energy optimization in cyber physical systems | |
Yahui et al. | The gas regulator fault diagnosis based on PCA-RBF neural network | |
Zhou et al. | Transient stability assessment of large-scale AC/DC hybrid power grid based on separation feature and deep belief networks | |
Li et al. | Residential Photovoltaic Power Forecasting Considering Division of Weather Type Index Interval | |
CN110851792A (en) | Staged and layered sampling method for operating intelligent electric energy meter | |
CN111008779A (en) | Transformer health state assessment method based on multi-factor assessment model | |
Zhang et al. | [Retracted] Neural Network Based on Health Monitoring Electrical Equipment Fault and Biomedical Diagnosis |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
WD01 | Invention patent application deemed withdrawn after publication | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190723 |