CN114021470B - Relay storage life prediction method based on AMFO algorithm and SVM algorithm - Google Patents
Relay storage life prediction method based on AMFO algorithm and SVM algorithm Download PDFInfo
- Publication number
- CN114021470B CN114021470B CN202111353432.7A CN202111353432A CN114021470B CN 114021470 B CN114021470 B CN 114021470B CN 202111353432 A CN202111353432 A CN 202111353432A CN 114021470 B CN114021470 B CN 114021470B
- Authority
- CN
- China
- Prior art keywords
- algorithm
- optimal
- svm
- principal component
- flame
- 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.)
- Active
Links
- 238000004422 calculation algorithm Methods 0.000 title claims abstract description 48
- 238000000034 method Methods 0.000 title claims abstract description 28
- 238000003860 storage Methods 0.000 title claims abstract description 26
- 230000006870 function Effects 0.000 claims abstract description 45
- 238000012549 training Methods 0.000 claims abstract description 31
- 238000005457 optimization Methods 0.000 claims abstract description 28
- 238000012360 testing method Methods 0.000 claims abstract description 21
- 230000015556 catabolic process Effects 0.000 claims abstract description 16
- 238000006731 degradation reaction Methods 0.000 claims abstract description 16
- 238000000513 principal component analysis Methods 0.000 claims abstract description 14
- 239000000463 material Substances 0.000 claims abstract description 4
- 239000013598 vector Substances 0.000 claims description 37
- 239000011159 matrix material Substances 0.000 claims description 20
- 238000004458 analytical method Methods 0.000 claims description 12
- 230000009467 reduction Effects 0.000 claims description 8
- 230000003044 adaptive effect Effects 0.000 claims description 7
- 238000012545 processing Methods 0.000 claims description 7
- 238000004364 calculation method Methods 0.000 claims description 6
- 238000011156 evaluation Methods 0.000 claims description 5
- 230000009977 dual effect Effects 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims description 2
- 238000012804 iterative process Methods 0.000 claims description 2
- 238000012163 sequencing technique Methods 0.000 claims description 2
- 230000009466 transformation Effects 0.000 claims description 2
- 238000005312 nonlinear dynamic Methods 0.000 abstract description 2
- 230000008901 benefit Effects 0.000 description 4
- 238000012706 support-vector machine Methods 0.000 description 4
- 230000007246 mechanism Effects 0.000 description 3
- 230000007774 longterm Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012847 principal component analysis method Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000000630 rising effect Effects 0.000 description 2
- OYPRJOBELJOOCE-UHFFFAOYSA-N Calcium Chemical compound [Ca] OYPRJOBELJOOCE-UHFFFAOYSA-N 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000006698 induction Effects 0.000 description 1
- 238000002955 isolation Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 230000026683 transduction Effects 0.000 description 1
- 238000010361 transduction Methods 0.000 description 1
Abstract
The invention discloses a relay storage life prediction method based on AMFO algorithm and SVM algorithm, which comprises the following steps: step 1, obtaining material performance parameters; step 2, carrying out principal component analysis on the parameter data to obtain principal component variables; step 3, the principal component variables are respectively taken as training set data and test set data, and the training set data are input into an SVM model for training and learning; step 4: optimizing parameters in a kernel function of the SVM model by adopting a self-adaptive moth flame optimization algorithm, constructing the SVM model by utilizing the optimal parameters, and establishing an optimized model for life prediction; step 5, setting a failure threshold value, and predicting a degradation curve reaching the failure threshold value; step 6, obtaining the final predicted life through calculating the probability density function. According to the invention, the data sample is put into the SVM model for training and then is subjected to life prediction, and the nonlinear dynamic self-adaptive step length method is introduced, so that the searching capability of the moth is enhanced, and the global optimizing capability of the moth is improved.
Description
Technical Field
The invention relates to a relay storage life prediction method, in particular to a relay storage life prediction method which integrates a self-adaptive moth flame optimization algorithm and an SVM algorithm.
Background
The electromagnetic relay is used as a basic electric control device, has the advantages of high conversion depth, good physical isolation performance and the like, and is widely applied to the field of electric control of spaceflight, military and the like. And is an indispensable link for the relay on the electronic device in the long-term storage stage. In the long-term storage link, the relay is influenced by environmental factors, so that the relay can fail in use, and the electronic product can fail.
The shelf life of a relay is an important parameter in studying the reliability of the relay. The method for predicting the storage life of the relay is researched, so that the storage life of the relay can be accurately predicted, and the expression form and failure mechanism of the storage failure of the relay are analyzed. In recent years, a relay is applied to various control circuits as an automatic control switch, and at the same time, the relay is also an electronic component with high failure rate. The reliability of the relay can influence the reliability of a system in which the relay is positioned, the reliability of the relay is low, the development of an electrical system is limited to a great extent, and meanwhile, the safety production of various industries is seriously influenced, so that safety accidents are caused.
Support Vector Machines (SVMs) were proposed by Vapnik in 1995 to discuss two classes of data classification but linearly non-timesharable. Therefore, the SVM is mainly used for solving the quadratic programming problem, namely solving a quadratic equation and continuously improving the quadratic programming problem, so that the model kernel function, the model parameters and the model training algorithm are continuously researched and deeply expanded. The SVM has natural advantages in solving the problem of small samples, and can solve the problems of overfitting and dimension disasters in learning, and the SVM converts the problem of linear inseparability in a low-dimensional space into a high-dimensional space by constructing a kernel function, so that the SVM can be linearly separated, and a classification hyperplane is constructed to maximize the interval between samples to realize classification.
The moth flame optimization algorithm (MFO algorithm) has the characteristics of simple model, few parameters, strong local searching capability, strong parallel optimization capability, good global property, difficult falling into the performance characteristics of local extremum and the like. However, in the traditional moth flame optimization algorithm, the updating mechanism of the moth position is realized through a logarithmic spiral function, but the function only defines the moth flying to candela, so that the moth is easy to fall into local optimum, certain defects exist in global optimization, and the problems of low calculation speed and low prediction precision of the traditional relay storage failure life prediction algorithm are caused.
Disclosure of Invention
In order to solve the problems of slow calculation speed and low prediction accuracy of a relay storage failure life prediction algorithm in the prior art, the invention provides a relay storage life prediction method which integrates a self-adaptive moth flame optimization algorithm (AMFO algorithm) and an SVM algorithm.
In order to achieve the above purpose, the invention is realized by the following technical scheme:
The invention relates to a relay storage life prediction method integrating AMFO algorithm and SVM algorithm, which comprises the following steps:
step 1: carrying out an accelerated degradation storage test on the electromagnetic relay to obtain material performance parameters required by building a training sample, such as contact resistance, arcing time, arcing energy, over-travel time, suction time, release time, rebound time and the like;
Step 2: performing principal component analysis on the degradation data, performing dimension reduction processing, removing redundant data, retaining factors with larger contribution to the overall sample, and acquiring principal component variables;
step 3: the principal component variables are respectively taken as training set data and test set data, and the training set data are input into an SVM model for training and learning;
Step 4: optimizing parameters in a kernel function of the SVM model by adopting a self-adaptive moth flame optimization Algorithm (AMFO), determining optimal parameters C and sigma, constructing the SVM model by utilizing the optimal parameters, improving the accuracy of prediction data of the SVM model, and constructing an optimized model for life prediction;
Step 5: setting a failure threshold, namely a service life end point, and selecting test samples in different intervals to enable initial prediction time points to be different, so as to predict a degradation curve reaching the failure threshold;
Step 6: and obtaining the final predicted life through calculating the probability density function, and comparing the final predicted life with the actual relay storage life to perform error analysis.
The performance data in step 1 is complex and has a large number of redundant items, and there is a complex nonlinear relationship between the factors, so that the analysis is difficult when the complex nonlinear relationship is used as an original data sample.
And 2, extracting data characteristics and reducing data dimension by adopting a Principal Component Analysis (PCA), and predicting by taking the reduced data as the input quantity of the SVM, so that the outstanding advantage of the SVM for solving the problem of small sample complexity can be reflected.
The support vector machine algorithm adopted in the step3 is a data mining method based on a statistical learning theory, and the mechanism is to find a superior classification hyperplane meeting classification requirements, so that optimal classification of linear separable data can be realized theoretically.
For a given sample set (x i,yi),i＝1,2,...,l,x∈Rn, y e { ±1}, the hyperplane is denoted as (ω x) +b=0, which is required to satisfy the following constraint in order for the classification to be correctly classified for all samples and to have a classification interval: y i [ (ω x) +b ]. Gtoreq.1, i=1, 2, l. the classification interval can be calculated as 2/||ω| 2, the problem of constructing an optimal hyperplane is thus translated into solving under constraints:
Where ω is the normal vector of the hyperplane and ω' is the transposed vector of ω.
To solve this constraint optimization problem, a Lagrangian function is introduced:
a i > 0 is the Lagrangian multiplier. The solution of the constraint optimization problem is determined by the saddle point of the lagrangian function, and the solution of the optimization problem satisfies the bias guide for w and b at the saddle point as 0, converting the QP problem into a corresponding dual problem, namely:
and alpha i in the formula is the Lagrangian multiplier corresponding to each data sample, and the factors have the following relationship
Obtaining the optimal solutionFor support vectors, calculate optimal weight vector w * and optimal bias b * as:
Thus, an optimal classification hyperplane (ω **x)+b* =0, and an optimal classification function is obtained
Wherein: w * is the optimal weight vector, b * is the optimal bias, a i is the support vector,Is the j-th value in the support vector.
And 4, determining optimal parameters by adopting a self-adaptive moth flame optimization algorithm, and constructing an SVM model according to the optimal parameters. The moths are actually search individuals moving within the search space, each of which surrounds a flame, and once a better solution is found, are updated to the location of the flames in the next generation.
The MFO algorithm is a triplet that approximates the global best in the optimization problem: mfo= (I, P, T), I is a function that generates a random population of moths and corresponding fitness values. The system model is as follows: P is the primary function that moves the moths in the search space P accepts the matrix M and returns updated M, P: M.fwdarw.M. If the termination criterion is met, the T function returns true; if not, the T function returns false, T: m→ { true, false }.
After initializing the function I, the position of the moth in the search space is changed by utilizing the function P, the updated moth matrix is output, and the individual position of the updated moth is judged. AMFO introduces a dynamic self-adaptive step factor alpha and a dynamic self-adaptive weight factor omega, expressed in the form of
Where l is the current iteration number and T is the maximum iteration number.
And 5, setting a failure threshold, namely a service life end point, and selecting test samples in different intervals to enable initial prediction time points to be different, so as to predict a degradation curve reaching the failure threshold.
Step 6, obtaining the final predicted life through calculating the probability density function. When the predicted degradation curve reaches a set failure threshold, an error analysis is performed in order to more accurately derive a final lifetime, in comparison with the actual relay storage lifetime.
The beneficial effects of the invention are as follows:
1. According to the invention, a principal component analysis method (PRINCIPAL COMPONENT ANALYSIS, PCA) is adopted to extract data characteristics and reduce data dimension, the reduced data is used as the input quantity of the SVM to predict, and the outstanding advantage of the SVM for solving the problem of small sample complexity can be represented.
2. Compared with the traditional moth flame algorithm, the self-adaptive moth flame optimization algorithm adopts a nonlinear dynamic self-adaptive step length method, and when the moth is close to the candelabra fire and searches for the optimal solution, the self-adaptive step length value is larger, which indicates that the algorithm can search for a larger range and the searching strength is larger, so that the moth searching capability can be enhanced.
3. The invention adopts a novel small sample learning method with a solid theoretical basis, namely an SVM, which basically does not relate to probability measure, law of large numbers and the like, thus being different from the existing statistical method. Essentially, the method avoids the traditional process from induction to deduction, realizes efficient transduction reasoning from training samples to forecasting samples, and greatly simplifies the problems of common classification, regression and the like. The final decision function of the SVM is determined by only a few support vectors, the computational complexity being dependent on the number of support vectors, rather than the dimension of the sample space, which avoids a "dimension disaster" in a sense. The few support vectors determine the final result, which not only helps us grasp the key samples and "reject" a large number of redundant samples, but also presupposes that the method is not only simple in algorithm, but also has better robustness.
Drawings
Fig. 1 is a flowchart of a relay shelf life prediction method of the present invention.
FIG. 2 is a flow chart of the adaptive moth flame optimization algorithm of the present invention.
Fig. 3 is a schematic diagram of a support vector machine, where black and white points represent two different samples, H 1 and H 2 are straight lines parallel to H, and H 1 and H 2 pass through the sample closest to the sample classification line and maximize the distance between H 1 and H 2.
Detailed Description
Embodiments of the invention are disclosed in the drawings, and for purposes of explanation, numerous practical details are set forth in the following description. However, it should be understood that these practical details are not to be taken as limiting the invention. That is, in some embodiments of the invention, these practical details are unnecessary.
The method is mainly divided into three parts, wherein the first part is preliminary processing of degradation data, a principal component analysis method is carried out on the degradation data, dimension reduction processing is carried out on the degradation data, redundant data is removed, variables with large contribution to sample data are reserved, and principal component variables are obtained. The second part is to measure the principal component variables from a training set and a sample set, put the training set and the sample set into an SVM model for training and learning, and then use a test set for storage life and analysis. And the third part optimizes the parameters of the SVM model by utilizing a self-adaptive moth flame optimization algorithm, so that the prediction accuracy and the convergence rate of the model are improved. Finally, the final predicted life is obtained by calculating the probability density function, and is compared with the actual storage life to carry out error analysis. Provides a new thought for the reliability research of electromagnetic relays, and has important significance for the reliability research of systems.
The JZC-200M electromagnetic relay of a certain factory is selected to carry out an accelerated storage degradation test, and the obtained experimental data are used as examples to give the concrete steps of the implementation example:
step 1: initial conditions for the test were set, and storage temperatures were set at 80 ℃, 106 ℃, 135 ℃ and 170 ℃, respectively. After the experimental parameters are set, the test system is used for starting the accelerated degradation test, and the upper computer is used for collecting data signals. The material performance parameters required by establishing a training sample, such as contact resistance, arcing time, arcing energy, over-travel time, suction time, release time, rebound time and the like, are obtained through experiments, the quantity of degradation data is huge, a large amount of redundant data exists, and all factors are mutually influenced and have complex nonlinear relations, so that the data processing and analysis workload of the sample is large and difficult.
Step 2: and carrying out principal component analysis on the original data, carrying out dimension reduction treatment, retaining the factors with larger overall contribution rate to the sample, removing the factors with lower contribution rate, and obtaining principal component variables, wherein the original data comprises contact resistance, arcing time, arcing energy, over-travel time, suction time, release time and rebound time.
The principle of principal component analysis (PRINCIPAL COMPONENT ANALYSIS, PCA) is to find a set of mutually orthogonal coordinate axes in order from the original feature space, starting from the first coordinate axis, first select the direction of maximum variance in the original feature space, then select the direction of maximum variance in the plane formed orthogonal to the previous coordinate axis, then select the direction of maximum variance in the plane orthogonal to the first two coordinate axes, and so on, to obtain k coordinate axes, and most of the information in the original data is contained in the k coordinate axes, and then the variance of the coordinate axes is almost 0, so that it can be ignored. And finally, forming a new feature space by the k features, and carrying out dimension reduction processing on the original data to a certain extent. The method comprises the following specific steps:
step 2-1: the original data set is vector normalized to form a P random vector x= (X 1,x2,...,xp)T, n total sample size, X i＝(xi1,xi2,...,xip)T), and then standard transformation is performed: Final construction matrix Z implements matrix normalization therein
Step 2-2: calculating the normalized matrix to obtain a correlation coefficient matrix
Step 2-3: solving a characteristic equation |R-gamma p |=0 of R to obtain P characteristic roots, determining a principal component information value to ensure that the accumulated contribution rate of the principal component can exceed 85%, and solving an equation set Rb=gamma p b for each gamma p to obtain a unit characteristic vector b;
step 2-4: converting the standardized index variable into a main component: Wherein b j is the corresponding j-th column single feature vector;
Step 2-5: and carrying out weighted summation on the m principal components through scoring, obtaining a final comprehensive evaluation value, wherein the weighted ratio is defined as the variance contribution rate of each principal component. Before principal component analysis:
Wherein the method comprises the steps of After principal component analysis:
Wherein: (a '1,a′2,a′3,...a′p)' is the unitized feature vector of the feature root, F i is the ith principal component of x and F i is a set of linear independent vectors.
Step 3: the contact resistance data is taken as training set and test set data, the training set data is input into the SVM model for training and learning, and specifically, the training set data is input into the SVM model for training and learning, and the training and learning steps comprise the following steps:
Step 3-1: for a given sample set (x i,yi),i＝1,2,...,l,x∈Rn, y e { ±1}, the hyperplane is denoted as (ω x) +b=0, which is required to satisfy the following constraint in order for the classification to be correctly classified for all samples and to have a classification interval: y i, (ω x) +b- > 1, i=1, 2,..i., the classification interval is calculated to be 2/|ω| 2, the problem of constructing an optimal hyperplane translates into solving under constraints:
where ω is the normal vector of the hyperplane, b is the hyperplane deviation, ω' is the transposed vector of ω.
Step 3-2: solving the constraint optimization problem, and introducing a Lagrangian function:
a i > 0 is a lagrangian multiplier, the solution of the constraint optimization problem is determined by the saddle point of the lagrangian function, and the solution of the optimization problem satisfies the bias of 0 to the hyperplane normal vector w and the hyperplane bias b at the saddle point, and the QP problem is converted into a corresponding dual problem, namely:
and alpha i in the formula is the Lagrangian multiplier corresponding to each data sample, and the factors have the following relationship
Obtaining the optimal solutionThe calculation of the optimal weight vector w * and the optimal bias b * are respectively as follows:
an optimal classification hyperplane (ω **x)+b* =0, and the optimal classification function is
Wherein: w * is the optimal weight vector, b * is the optimal bias, a * is the support vector,Is the j-th value in the support vector.
Step 4: and optimizing parameters in the SVM model by adopting a self-adaptive moth flame optimization algorithm, improving the prediction data precision of the SVM model, and establishing an optimized model for life prediction.
The adaptive moth flame optimization Algorithm (AMFO) adopted here determines optimal parameters C and sigma, and builds an SVM model by using the optimal parameters, and comprises the following steps:
Step 4-1: initializing parameters, setting the population size N of the moths, the searched space dimension d, the maximum iteration number T and the flame number N, and initializing the positions of the moths in the space;
step 4-2: the matrix OM stores fitness values of the moths, and the matrix M represents positions of the moths:
mij＝[bub(i)-blb(i)]rand()+blb(i)
Wherein b ub (i) and b lb (i) are the upper and lower limits, respectively, of the ith moth position;
step 4-3: the matrix OF stores the fitness value OF the flame, and the matrix F represents the position OF the flame:
step 4-4: calculating the fitness value of the individuals, sequencing all the individuals according to the sequence from small to large, finding out the optimal position of the moths and assigning the optimal position to the flame;
step 4-5: using the formula Updating the dynamic adaptive inertial weights and flame position using the formulaReducing the number of flames, wherein ω is a dynamic adaptive weight factor, wherein N is the maximum of the number of flames, l is the current iteration number, T is the maximum iteration number, and the iteration number l=l+1;
Step 4-6: will adapt the step formula Applied to the distance formula of the moth M i and the flame F j Updating the distance D i between the moth and the flame, wherein l is the current iteration number, T is the maximum iteration number,Is a parameter;
Step 4-7: updating the positions of the moths by using a logarithmic spiral function formula S (M i,Fj)＝Di*ept*cos2πt+ωFj, wherein M i and F j are the positions of the moths and the flames respectively, D i is the distance between the updated moths and the flames, p is a spiral shape constant, ω is a dynamic self-adaptive weight factor, the dynamic self-adaptive weight factor is multiplied by j flames, and the nonlinear self-adaption reduction from 1 to 0 along with the increase of iteration times ensures that the moths advance towards the correct searching direction, and the accuracy of an algorithm is effectively improved;
step 4-8: continuously performing iterative updating until a stopping condition is met, outputting the optimal position of flame and a corresponding fitness value in the whole iterative process, stopping iterative searching, namely finishing the algorithm, and outputting an optimal parameter value;
step 4-9: and constructing an SVM model according to the optimal parameters.
Specifically, the prediction process of life prediction of the SVM model includes the following steps:
Step 4-10-1: normalizing the original performance parameter data to establish a training sample set;
Step 4-10-2: selecting a kernel function and parameters: in order to improve the calculation accuracy of the SVM prediction model, proper model parameters including insensitive parameters, penalty factors, kernel functions and the like are required to be selected. The penalty factor is a parameter that maintains the training sample error and model complexity balance. The common kernel functions include radial basis kernel functions and polynomial kernel functions, different SVM models can appear in different kernel function selections, proper regular parameters can be selected to ensure that the SVM generalization capability is best, and the calculation formula of the parameters is provided In the middle ofAnd σ y are mean and variance, respectively;
Step 4-10-3: sample training is carried out according to the established SVM model, and then the SVM model is predicted, so that a related prediction result is obtained, an input sample is substituted into the model for analysis, and a prediction value is output;
Step 4-10-4: and (4) checking the predicted value and the actual value, and returning to the step (4-10-3) if the predicted value is not satisfied, otherwise, performing error analysis.
And 4, determining optimal parameters by adopting a self-adaptive moth flame optimization algorithm, and constructing an SVM model according to the optimal parameters. The moths are actually search individuals moving within the search space, each of which surrounds a flame, and once a better solution is found, are updated to the location of the flames in the next generation.
The MFO algorithm is a triplet that approximates the global best in the optimization problem: mfo= (I, P, T), I is a function that generates a random population of moths and corresponding fitness values. The system model is as follows: P is the primary function that moves the moths in the search space P accepts the matrix M and returns updated M, P: M.fwdarw.M. If the termination criterion is met, the T function returns true; if not, the T function returns false, T: m→ { true, false }.
After initializing the function I, the position of the moth in the search space is changed by utilizing the function P, the updated moth matrix is output, and the individual position of the updated moth is judged. AMFO introduces a dynamic self-adaptive step factor alpha and a dynamic self-adaptive weight factor omega, expressed in the form of
Where l is the current iteration number and T is the maximum iteration number.
Step 5: the failure threshold value, namely the service life end point is set, test samples in different intervals are selected according to experimental waveforms, and in the embodiment, data in different time periods can be selected for testing, for example, a data time point with steep rising contact resistance value is taken as an initial prediction point, then a data time point with gentle rising is taken as the initial prediction point, so that the initial prediction time points are different, and a degradation curve reaching the failure threshold value is predicted. The predicted trends of four different sets of temperature stresses were compared.
Step 6: the final predicted lifetime is derived by calculating the probability density function. When the predicted degradation curve reaches the set failure threshold, in order to obtain the final life more accurately, the probability density at the point of arrival is calculated, and finally, the error analysis is performed in comparison with the actual relay storage life.
The MSE, the MAE and the MRE are adopted as evaluation indexes to compare and verify the experimental prediction results, wherein the average relative Error (MEAN RELATIVE Error, MRE) is used for measuring the ratio of the average absolute Error value to the target average value of all test samples, and the evaluation indexes can well solve the problem of magnitude difference among the samples. The correlation formula is as follows:
Wherein y is the true residual life value of all samples in the test data; y' is the predicted lifetime value of all samples in the test data; y i is the true remaining life value of the ith sample in the test data; y i' is the predicted lifetime value of the ith sample in the test data; n is the total number of samples. The smaller the value of the evaluation index MAE, MSE, MRE, the higher the prediction accuracy of the representative model, and the better the fitting effect of the model.
The foregoing description is only illustrative of the invention and is not to be construed as limiting the invention. Various modifications and variations of the present invention will be apparent to those skilled in the art. Any modification, equivalent replacement, improvement, or the like, which is within the spirit and principles of the present invention, should be included in the scope of the claims of the present invention.
Claims (4)
1. The relay storage life prediction method based on AMFO algorithm and SVM algorithm is characterized in that: the relay storage life prediction method comprises the following steps:
step 1: carrying out an accelerated degradation storage test on the electromagnetic relay to obtain material performance parameters required by building training samples;
Step 2: performing principal component analysis on the original performance parameter data in the step 1, performing dimension reduction processing, removing redundant data, and reserving factors with larger overall contribution to the sample to obtain principal component variables;
Step 3: respectively taking the main component variables obtained in the step 2 as training set data and test set data, and inputting the training set data into an SVM model for training and learning;
Step 4: optimizing parameters in a kernel function of the SVM model by adopting a self-adaptive moth flame optimization algorithm, determining optimal parameters C and sigma, constructing the SVM model by utilizing the optimal parameters, improving the accuracy of prediction data of the SVM model, and constructing an optimized model for life prediction;
Step 5: setting an invalidation threshold, selecting test samples in different intervals to enable initial prediction time points to be different, and predicting a degradation curve reaching the invalidation threshold;
Step 6: the final predicted life is obtained by calculating the probability density function and compared with the actual relay storage life, an error analysis is performed, wherein,
In the step 4, an adaptive moth flame optimization algorithm is adopted to determine optimal parameters C and sigma, and an SVM model is constructed by utilizing the optimal parameters, and the method comprises the following steps:
Step 4-1: initializing parameters, setting the population size N of the moths, the searched space dimension d, the maximum iteration number T and the flame number N, and initializing the positions of the moths in the space;
step 4-2: the matrix OM stores fitness values of the moths, and the matrix M represents positions of the moths:
mij＝[bub(i)-blb(i)]rand()+blb(i)
Wherein b ub (i) and b lb (i) are the upper and lower limits, respectively, of the ith moth position;
step 4-3: the matrix OF stores the fitness value OF the flame, and the matrix F represents the position OF the flame:
step 4-4: calculating the fitness value of the individuals, sequencing all the individuals according to the sequence from small to large, finding out the optimal position of the moths and assigning the optimal position to the flame;
step 4-5: using the formula Updating the dynamic adaptive inertial weights and flame position using the formulaReducing the number of flames, wherein ω is a dynamic adaptive weight factor, wherein N is the maximum of the number of flames, l is the current iteration number, T is the maximum iteration number, and the iteration number l=l+1;
Step 4-6: will adapt the step formula Applied to the distance formula of the moth M i and the flame F j Updating the distance D i between the moth and the flame, wherein l is the current iteration number, T is the maximum iteration number and θ is a parameter;
Step 4-7: updating the positions of the moths by using a logarithmic spiral function formula S (M i,Fj)＝Di*ept*cos2πt+ωFj, wherein M i and F j are the positions of the moths and the flames respectively, D i is the distance between the updated moths and the flames, p is a spiral shape constant, ω is a dynamic self-adaptive weight factor, the dynamic self-adaptive weight factor is multiplied by j flames, and the nonlinear self-adaption reduction from 1 to 0 along with the increase of iteration times ensures that the moths advance towards the correct searching direction, and the accuracy of an algorithm is effectively improved;
step 4-8: continuously performing iterative updating until a stopping condition is met, outputting the optimal position of flame and a corresponding fitness value in the whole iterative process, stopping iterative searching, namely finishing the algorithm, and outputting an optimal parameter value;
step 4-9: and constructing an SVM model according to the optimal parameters.
2. The relay shelf life prediction method based on AMFO algorithm and SVM algorithm according to claim 1, wherein: the prediction process of the life prediction of the SVM model in the step 4 comprises the following steps:
Step 4-10-1: normalizing the original performance parameter data to establish a training sample set;
step 4-10-2: selecting a kernel function and parameters: different kernel function selections can generate different SVM models and calculation formulas of parameters In the middle ofAnd σ y are mean and variance, respectively;
Step 4-10-3: sample training is carried out according to the established SVM model, and then the SVM model is predicted, so that a related prediction result is obtained, an input sample is substituted into the model for analysis, and a prediction value is output;
Step 4-10-4: and (4) checking the predicted value and the actual value, and returning to the step (4-10-3) if the predicted value is not satisfied, otherwise, performing error analysis.
3. The relay shelf life prediction method based on AMFO algorithm and SVM algorithm according to claim 1, wherein: the dimension reduction processing in the step2 specifically comprises the following steps:
Step 2-1: vector normalization is performed on the original performance parameter data to form a P random vector x= (X 1,x2,...,xp)T, total n sample sizes, X i＝(xi1,xi2,...,xip)T), and then standard transformation is performed: the final construction matrix Z implements matrix normalization, wherein
Step 2-2: calculating the normalized matrix to obtain a correlation coefficient matrix
Step 2-3: solving a characteristic equation |R-gamma p |=0 of R to obtain P characteristic roots, determining a principal component information value to ensure that the accumulated contribution rate of the principal component can exceed 85%, and solving an equation set Rb=gamma p b for each gamma p to obtain a unit characteristic vector b;
Step 2-4: the normalized index variable z ij is converted into a principal component: Wherein b j is the corresponding j-th column single feature vector;
Step 2-5: and (3) carrying out weighted summation on m principal components in the step (2-4) through scoring, obtaining a final comprehensive evaluation value, wherein the weighted ratio is defined as the variance contribution rate of each principal component, and the principal component is analyzed before:
Wherein the method comprises the steps of
After principal component analysis:
where (a '1,a′2,a′3,...a′p)' is the unitized feature vector of the feature root, F i is the ith principal component of x and F i is a set of linearly independent vectors.
4. The relay shelf life prediction method based on AMFO algorithm and SVM algorithm according to claim 1, wherein: the training set data input SVM model training learning step in the step 3 comprises the following steps:
Step 3-1: for a given sample set (x i,yi),i＝1,2,...,l,x∈Rn, y e { ±1}, the hyperplane is denoted as (ω x) +b=0, which is required to satisfy the following constraint in order for the classification to be correctly classified for all samples and to have a classification interval: y i [ (ω x) +b ]. Gtoreq.1, i=1, 2,.. the classification interval is calculated to be 2/||ω|| 2, the problem of constructing an optimal hyperplane translates into solving under constraints:
wherein ω is the normal vector of the hyperplane, b is the hyperplane deviation, ω' is the transposed vector of ω;
Step 3-2: solving the constraint optimization problem, and introducing a Lagrangian function: a i >0 is a lagrangian multiplier, the solution of the constraint optimization problem is determined by the saddle point of the lagrangian function, and the solution of the optimization problem satisfies the bias of 0 to the hyperplane normal vector w and the hyperplane bias b at the saddle point, and the problem is converted into corresponding dual problem, namely:
and alpha i in the formula is the Lagrangian multiplier corresponding to each data sample, and the factors have the following relationship
Obtaining the optimal solutionWherein α * is a support vector, and the calculated optimal weight vector w * and optimal bias b * are respectively:
An optimal classification hyperplane (ω **x)+b* =0, and the optimal classification function is:
Wherein: w * is the optimal weight vector, b * is the optimal bias, a * is the support vector, Is the j-th value in the support vector.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111353432.7A CN114021470B (en) | 2021-11-16 | Relay storage life prediction method based on AMFO algorithm and SVM algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111353432.7A CN114021470B (en) | 2021-11-16 | Relay storage life prediction method based on AMFO algorithm and SVM algorithm |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114021470A CN114021470A (en) | 2022-02-08 |
| CN114021470B true CN114021470B (en) | 2024-07-02 |
Family
ID=
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110991494A (en) * | 2019-11-13 | 2020-04-10 | 温州大学 | Method for constructing prediction model based on improved moth optimization algorithm |
| CN111898507A (en) * | 2020-07-22 | 2020-11-06 | 武汉大学 | Deep learning method for predicting earth surface coverage category of label-free remote sensing image |
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110991494A (en) * | 2019-11-13 | 2020-04-10 | 温州大学 | Method for constructing prediction model based on improved moth optimization algorithm |
| CN111898507A (en) * | 2020-07-22 | 2020-11-06 | 武汉大学 | Deep learning method for predicting earth surface coverage category of label-free remote sensing image |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108734355B (en) | Short-term power load parallel prediction method and system applied to power quality comprehensive management scene | |
| CN104881706B (en) | A kind of power-system short-term load forecasting method based on big data technology | |
| CN112699913A (en) | Transformer area household variable relation abnormity diagnosis method and device | |
| Zhao et al. | A novel hybrid method of parameters tuning in support vector regression for reliability prediction: particle swarm optimization combined with analytical selection | |
| KR102215690B1 (en) | Method and apparatus for time series data monitoring | |
| CN109740859A (en) | Transformer condition evaluation and system based on Principal Component Analysis and support vector machines | |
| CN113987033B (en) | Main transformer online monitoring data group deviation identification and calibration method | |
| CN114019370B (en) | Motor fault detection method based on gray level image and lightweight CNN-SVM model | |
| CN114881101B (en) | Bionic search-based power system typical scene association feature selection method | |
| CN113205233A (en) | Lithium battery life prediction method based on wolf algorithm and multi-core support vector regression | |
| CN117214637B (en) | Charging gun withstand voltage test method, device, equipment and storage medium | |
| CN116169670A (en) | Short-term non-resident load prediction method and system based on improved neural network | |
| CN116737510A (en) | Data analysis-based intelligent keyboard monitoring method and system | |
| CN115688581A (en) | Oil gas gathering and transportation station equipment parameter early warning method, system, electronic equipment and medium | |
| Zhang et al. | Fault diagnosis of oil-immersed power transformer based on difference-mutation brain storm optimized catboost model | |
| CN114936514A (en) | Lithium ion battery life prediction method and device based on improved grey model | |
| CN114021470B (en) | Relay storage life prediction method based on AMFO algorithm and SVM algorithm | |
| CN110334508A (en) | A kind of host sequence intrusion detection method | |
| CN113033898A (en) | Electrical load prediction method and system based on K-means clustering and BI-LSTM neural network | |
| CN117113852A (en) | GIS life prediction method based on electric, chemical, structural and vibration parameters | |
| CN115618743B (en) | State evaluation method and state evaluation system of sighting telescope system | |
| CN115759395A (en) | Training of photovoltaic detection model, detection method of photovoltaic power generation and related device | |
| CN116128148A (en) | Photovoltaic power station output prediction network training method, prediction method, device and equipment | |
| CN110852370A (en) | Clustering algorithm-based large-industry user segmentation method | |
| CN115936773A (en) | Internet financial black product identification method and system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant |