Specific embodiment
The present invention will be described in detail with reference to the accompanying drawings and examples, but not as a limitation of the invention.
In one embodiment, as shown in Figure 1, the disclosure discloses a kind of insulator sudden strain of a muscle based on Support vector regression
Network voltage-prediction method, comprising:
S1, according to the geometric dimension of known insulator sample, calculate insulator sample along face field distribution, obtain along face
The characteristic parameter of field distribution.
S2, the definition according to characteristic parameter calculate the characteristic ginseng value along face field distribution, are denoted as input quantity.
S3, the flashover voltage for calculating test insulator sample and its corresponding breakdown probability, are denoted as output quantity.
S4, pretreatment input quantity and output quantity data.
S5, pretreated data are divided, is divided into training set data and test set data.
S6, training set data typing Support vector regression model is trained, processing is optimized in training process,
Obtain the optimal Support vector regression model of training result.
It is carried out in the optimal Support vector regression model of training result described in S7, the input quantity data inputting by test set
Verifying, with the model prediction the flashover voltage of the insulator after verifying.
Using the technical solution of the embodiment of the present disclosure, at least have the following beneficial effects:
1, the Support vector regression method that the present invention uses is the machine learning method suitable for small sample, model parameter
Simply, parameter optimization is high-efficient.
2, prediction process of the present invention is simple, and prediction accuracy is high, facilitates the guidance optimization gas-solid insulation system of insulator,
The economy and reliability of air insulating device are improved, while experimental amount needed for decreasing gas-solid dielectric(al) test, drop
Low experimentation cost.
In another embodiment, in step S1, according to the geometric dimension of known insulator sample, insulator sample is calculated
Along face field distribution, obtain the characteristic parameter along face field distribution, comprising:
S101, according to the geometric dimension of known insulator sample, establish Electric Field Simulation model.
S102, calculated by limited element analysis technique or Analogue charge method insulator sample along face field distribution, obtain edge
The characteristic parameter of face field distribution.
In another embodiment, in step S1, along face, field distribution includes: (to synthesize field along face along face total electric field
By force), the tangential electric field (tangential field component) for being parallel to flashover path and the normal electric field (normal direction perpendicular to flashover path
Field strength component).
In the present embodiment, the starting of influence insulator arc-over has been comprehensively considered along the characteristic parameter of face field distribution, had been developed
The gas-solid interface of journey along face total electric field strength and its cut, normal component along EDS maps feature, insulator can be significantly improved
The precision of prediction of flashover voltage.
In another embodiment, in step S1, along face, field distribution further includes the external shield electrode of flashover path range
Structure.
In another embodiment, in step S1, the characteristic parameter along face field distribution include along face formate field intensity and its
It cuts, the maximum value of normal direction field strength component, average value, the nonuniformity coefficient of electric field, the maximum value of electric-force gradient and average value, electric field
Square and more than threshold field domain integral.
In another embodiment, in step S2, according to the definition of characteristic parameter, the feature calculated along face field distribution is joined
Numerical value.Insulator geometry is divided into grid by limited element analysis technique by the present embodiment, and calculates the ginseng on grid node
Numerical value.Wherein, i is number (i=1,2 ..., n) of the insulator along surface grids node;XiIndicate insulator along i-th of face grid
Characteristic parameter X on node;E, Et, EnInsulator is respectively indicated along the formate field intensity in face, the tangential component of field strength and normal direction point
Amount.Specific features parameter is defined as follows:
1) it formate field intensity and its cuts, the maximum value of normal direction field strength component
Emax=max (Ei)=Ei1;EmaxIndicate that insulator along the maximum value of face formate field intensity, corresponds to i-th1A grid node
Formate field intensity Ei1。
l_Emax=li1;l_EmaxIndicate EmaxCorresponding position, corresponding i-th1The arc length l of a grid nodei1。
Et_Emax=Eti1;Et_EmaxIndicate EmaxThe tangential field component at place, corresponding i-th1The tangential field of a grid node
Eti1。
En_Emax=Eni1;En_EmaxIndicate EmaxThe normal direction field strength component at place, corresponding i-th1The normal direction field strength of a grid node
Eni1。
Etmax=max (Eti1)=Eti2;EtmaxIndicate that insulator along the maximum value of face tangential field component, corresponds to i-th2It is a
The tangential field E of grid nodeti2。
l_Etmax=li2;l_EtmaxIndicate EtmaxCorresponding position, corresponding i-th2The arc length l of a grid nodei2。
E_Etmax=Eti2;E_EmaxIndicate EtmaxThe formate field intensity at place, corresponding i-th2The formate field intensity E of a grid nodeti2。
En_Etmax=Eni2;En_EmaxIndicate EtmaxThe normal direction field strength component at place, corresponding i-th2The normal direction field strength of a grid node
Eni2。
Enmax=max (Eni)=Eni3;EnmaxIndicate that insulator along the maximum value of face normal direction field strength component, corresponds to i-th3A net
The normal direction field strength E of lattice nodeni3。
l_Enmax=li3;l_EnmaxIndicate EnmaxCorresponding position, corresponding i-th3The arc length l of a grid nodei3。
E_Enmax=Ei3;E_EnmaxIndicate EnmaxThe formate field intensity at place, corresponding i-th3The formate field intensity E of a grid nodei3。
Et_Enmax=Eti3;Et_EnmaxIndicate EnmaxThe tangential field component at place, corresponding i-th3The tangential field of a grid node
Strong Eti3。
2) average value
E_avIndicate insulator along the average value of face formate field intensity, corresponding insulator edge
The average value of all mesh nodes in face (i=1,2 ..., n) formate field intensity.
Et_avIndicate insulator along the average value of face tangential field, corresponding insulator
Along the average value of all mesh nodes in face (i=1,2 ..., n) tangential field.
En_avIndicate that insulator along the average value of face normal direction field strength, corresponds to exhausted
Average value of edge along all mesh nodes in face (i=1,2 ..., n) normal direction field strength.
3) nonuniformity coefficient of electric field
F=Emax/E_av;F indicates the nonuniformity coefficient along face total electric field.
ft=Etmax/Et_av;ftIndicate the nonuniformity coefficient along the tangential electric field in face.
fn=Enmax/En_av;fnIndicate the nonuniformity coefficient along face normal electric field.
4) maximum value and average value of electric-force gradient
E_gmIndicate insulator along the absolute value of face formate field intensity maximum of gradients;Indicate electric-force gradient of the insulator on surface grids node.
Et_gmIndicate insulator along the absolute of face tangential field maximum of gradients
Value;Indicate tangential electric-force gradient of the insulator on surface grids node.
En_gmIndicate insulator along the absolute of face normal direction magnetic field gradient maximum value
Value;Indicate normal electric field gradient of the insulator on surface grids node.
E_gaIndicate insulator being averaged along face formate field intensity gradient
Value.
Et_gaIndicate insulator along the flat of face tangential field gradient
Mean value.
En_gaIndicate insulator along face normal direction magnetic field gradient
Average value.
5) electric field square, the domain integral more than threshold field
Wherein, WeIndicate the quadratic sum along face electric field, EiIndicate that insulator synthesizes field along face
By force.
Wherein, WeaIndicate the average value of the quadratic sum along face electric field.
Wherein, VxIt indicates to be greater than maximum value x%E along face formate field intensitymaxThe product in region
Point, EiIndicate i-th of grid along face formate field intensity;ExIndicate x%Emax;The percentage of x expression field strength maximum value, x=10~
90。
VxIndicate the x% (x% for being greater than its maximum value along face tangential field component
Etmax) region integral, EtiIndicate i-th of grid along face tangential field component;EtxX% (the x%E of expressiontmax);X is indicated
The percentage of tangential field component maximum value, x=10~90.
VxIndicate the x% (x%E for being greater than its maximum value along face normal direction field strengthnmax)
The integral in region, EniIndicate i-th of grid along face normal direction field strength component;ExIndicate x%Enmax;X indicates normal direction field strength component
The percentage of maximum value, x=10~90.
In another embodiment, in step S3, the flashover voltage for testing insulator sample and its corresponding breakdown are calculated
Probability, comprising:
S301, the flashover voltage experimental test data that flashover number is greater than 30 times are obtained.
S302, it is distributed using normal distribution or Weibull, the flashover voltage of insulator sample and its corresponding is calculated
Breakdown probability.
In the present embodiment, flashover voltage experimental test data are to obtain under outer construction frequency 50Hz, direct current or surge voltage
Data.
In another embodiment, in step S4, input quantity and output quantity data are pre-processed, comprising:
S401, input quantity and output quantity data are normalized.
S402, dimension-reduction treatment is carried out to the data after normalized, obtains preprocessed data.
In the present embodiment, pass through Min-max standardized method (maximum value-minimum value method, as shown in formula (1)) or z-
Input quantity and output quantity data is normalized in score standardized method.
Wherein, xiIndicate input quantity and/or output quantity data, xmaxAnd xminRespectively indicate xiMaximum value and minimum value,
Value after indicating normalization.
In the present embodiment, dimension-reduction treatment is carried out by Principal Component Analysis Method or Pearson correlation coefficient method.Pearson phase
Shown in relationship number r calculation method such as formula (2).
Wherein Xi, yiFor i-th of value of two variable arrays,For the average value of two groups of variables.
In another embodiment, in step S6, training set data typing Support vector regression model is trained,
Processing is optimized in training process, obtains the optimal Support vector regression model of training result, comprising:
S601, the suitable Support vector regression kernel function of selection.
S602, training set data typing Support vector regression model is trained.
Processing is optimized to kernel functional parameter and penalty factor using grid data service in S603, training process, is obtained
The optimal Support vector regression model of training result.
In another embodiment, method further include:
The average deviation that S8, computation model are predicted, and judge average deviation whether within the scope of allowable error;If allowing
In error range, then decision model is effective;If not within the scope of allowable error, return step S603, again to kernel function
Parameter and penalty factor optimize processing, until obtaining the optimal Support vector regression model of training result.
In the present embodiment, the allowable error range of the average deviation of model prediction is no more than 20%.
In the present embodiment, pass through the average deviation of following formula computation model prediction:
Wherein, δ indicates the average deviation of model prediction, UpIndicate the flashover voltage of the insulator of model prediction, UfIndicate practical
The flashover voltage of the insulator of test.
In the present embodiment, the average deviation predicted by computation model, using the average deviation of prediction as model prediction
Evaluation criterion, the validity and accuracy of verifying this method prediction.
In another embodiment, in step S5, the ratios of training set data and test set data be 1: 1,2: 1,3: 1 or
4∶1。
In another embodiment, in step S601, the Support vector regression kernel function (kernel of selection
It function) is Radial basis kernel function, shown in expression formula such as formula (4).
Wherein Xi, XjIt is the width parameter of function, the radial effect range of control function for the point in space.
The objective function and constraint condition of support vector regression (SVR) are as shown in formula (5) and formula (6).
Wherein, ε indicates patient deviation from regression band, ξi,For relaxation factor, C is penalty factor, for controlling
Found in objective function and be spaced the smallest hyperplane and guarantee the weight between data point deviation minimum, ω and b be it needs to be determined that
Parameter.
A kind of insulator based on Support vector regression that the disclosure discloses is dodged below by some specific embodiments
Network voltage-prediction method describes in detail, but not as the restriction to the disclosure.
As shown in fig. 6, a kind of the flashover voltage of the insulator prediction technique based on Support vector regression, including such as lower section
Method:
1, Electric Field Simulation model is established according to the geometric dimension (as shown in Figure 2) of known insulator sample.Using finite element
Analytic approach, the insulator being calculated on flashover path is along face field distribution characteristic parameter (flashover path such as Fig. 3 shows).
2, it according to the definition of characteristic parameter, calculates along face field distribution characteristic ginseng value, is denoted as input quantity.
3, the flashover voltage (flashover number is greater than 30 times) for obtaining test insulator sample, is punctured using normal distribution
The flashover voltage that probability is 50%, is denoted as output quantity.
4, to avoid influence of the unit to calculated result, using maximum value-minimum value method to input quantity data and output quantity
Data carry out parameter normalization processing, related to carrying out along face field distribution characteristic ginseng value using Pearson correlation coefficient method
Property analysis (dimension-reduction treatment).In this example, by calculating the correlation between characteristic ginseng value and flashover voltage, rejecting and flashover
The unrelated characteristic ginseng value of voltage (correlation coefficient threshold is | r1| < 0.3), then calculate the phase between residue character parameter value
Guan Xing, reject the higher characteristic ginseng value of cross correlation measure (correlation coefficient threshold is | r2| < 0.4), guarantee input characteristic parameter value
Between orthogonality, obtain pretreated data (input characteristic parameter determine, process as shown in Figure 4).
5, pretreated data are classified (training data setting and test data are arranged), is divided into 75% training
Collect the test set data of data and 25%.
6, the Support vector regression model based on Radial basis kernel function is selected, is then supported using training set data training
Vector machine regression model.Using grid data service to the parameter (γ) and penalty factor (C) of Radial basis kernel function in training process
Processing is optimized, the optimal Support vector regression model of training result is obtained.
7, it will be verified in the optimal Support vector regression model of the input quantity data inputting training result of test set,
With the model prediction the flashover voltage of the insulator after verifying.
8, the average deviation of computation model prediction obtains the average deviation of flashover voltage prediction no more than 20%.
In the present embodiment, prediction model parameters are calculated using training set data, and with test set data to model
It is evaluated, prediction result and prediction average deviation such as Fig. 5 (a), Fig. 5 (b) are shown.
In other example, it may be necessary to manually adjust between input parameter, between input parameter and output parameter
Correlation coefficient threshold (r1、r2)。
It applies specific embodiment in the disclosure principle and embodiment of the disclosure is described in detail, the above reality
The application for applying example is only used for the application method and its thinking for helping to understand the disclosure, does not constitute the limit to disclosure application scenarios
System.There will be changes according to the actual situation in specific embodiments and applications for the disclosure, is not departing from the disclosure
In the case where technical characteristic given by technical solution, to increase made by technical characteristic, deform or with the same content in this field
Replacement, the protection scope of the disclosure should all be belonged to.