[go: up one dir, main page]
More Web Proxy on the site http://driver.im/

CN117951588B - Modeling method, device, terminal and medium for OIP sleeve health state evaluation model - Google Patents

Modeling method, device, terminal and medium for OIP sleeve health state evaluation model Download PDF

Info

Publication number
CN117951588B
CN117951588B CN202410355139.1A CN202410355139A CN117951588B CN 117951588 B CN117951588 B CN 117951588B CN 202410355139 A CN202410355139 A CN 202410355139A CN 117951588 B CN117951588 B CN 117951588B
Authority
CN
China
Prior art keywords
sample data
fault
oip
data
sleeve
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
Application number
CN202410355139.1A
Other languages
Chinese (zh)
Other versions
CN117951588A (en
Inventor
郑哲然
袁耀
喇元
王增超
黎文浩
黄克捷
杨家辉
林显军
高锡明
张曦
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
China South Power Grid International Co ltd
China Southern Power Grid Co Ltd
Original Assignee
China South Power Grid International Co ltd
China Southern Power Grid Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by China South Power Grid International Co ltd, China Southern Power Grid Co Ltd filed Critical China South Power Grid International Co ltd
Priority to CN202410355139.1A priority Critical patent/CN117951588B/en
Publication of CN117951588A publication Critical patent/CN117951588A/en
Application granted granted Critical
Publication of CN117951588B publication Critical patent/CN117951588B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • 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
    • G06N3/0475—Generative 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
    • G06N3/094—Adversarial learning

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Health & Medical Sciences (AREA)
  • Complex Calculations (AREA)

Abstract

The application discloses a modeling method, a modeling device, a modeling terminal and a modeling medium for an OIP sleeve health state assessment model.

Description

Modeling method, device, terminal and medium for OIP sleeve health state evaluation model
Technical Field
The application relates to the technical field of transformers, in particular to a modeling method, a modeling device, a modeling terminal and a modeling medium for an OIP sleeve health state evaluation model.
Background
Transformers play a vital role in the power system. Bushings are a key component of power transformers and are also one of the main contributors to transformer failure. According to CIGRE's data, transformer bushing faults account for about 28% of all transformer faults. The prior researches show that the deterioration of the insulation state inside the bushing is an important cause for the bushing fault, the Oil paper (Oil IMPREGNATED PAPER, OIP) bushing is one of the bushing types which are most widely applied as a transformer, and potential problems can be found in time and targeted maintenance measures can be taken by monitoring key parameters of the bushing on line and evaluating the insulation state inside the bushing, so that the reliability and the safety of the bushing and the transformer are improved.
At present, the method for evaluating the health state of the transformer bushing is a neural network analysis method, the bushing fault cause can be analyzed by utilizing the learning process of a plurality of characteristic quantities of the transformer bushing in the BP neural network, and the transformer bushing fault identification based on multi-source parameters is realized, but the method has the technical problem of unstable evaluation and judgment accuracy in practical application.
Disclosure of Invention
The application provides an OIP sleeve health state evaluation method, an OIP sleeve health state evaluation device, an OIP sleeve health state evaluation terminal and an OIP sleeve health state evaluation medium, which are used for solving the technical problem that the evaluation and judgment accuracy is unstable in the existing transformer sleeve health state evaluation mode.
In order to solve the technical problem, a first aspect of the present application provides a modeling method for an OIP casing health state evaluation model, including:
acquiring sample data of the OIP sleeve, and classifying the sample data according to the state of the OIP sleeve corresponding to each sample data;
Inputting fault sample data as a real data set into the improved GAN network model through a preset improved GAN network model, learning a distribution rule of the fault sample data through a generator in the improved GAN network model, and combining noise vectors to obtain generated sample data, wherein when the generated sample data passes through the judgment of a discriminator, the generated sample data is set as fault expansion sample data so as to be integrated into the fault sample data;
Outputting a sleeve sample data total set of the OIP sleeve when the number relation of the fault sample data and the non-fault sample data corresponding to each fault type meets the preset number relation condition;
inputting the sleeve sample data total set into an SVM classifier model, training the SVM classifier model, and obtaining an OIP sleeve health state assessment model when the training progress of the SVM classifier model meets the preset training termination condition.
Preferably, when the generated sample data passes the determination of the arbiter, setting the generated sample data as the fault extended sample data specifically includes:
The discriminator calculates the distribution similarity between the generated sample data and the fault sample data through a distribution similarity formula based on WGAN networks, and when the distribution similarity meets a preset similarity condition, the generated sample data is set as fault expansion sample data.
Preferably, the distribution similarity formula is specifically:
；
In the method, in the process of the invention, For the distribution similarity between the fault sample data and the generated sample data, P r is the edge distribution data of the fault sample data, P g is the edge distribution data of the generated sample data, K represents the number of faulty sample data x, when the function f (x) satisfies the K-lipschitz continuous condition, the upper bound in the absolute value of its derivative, the ||f L || represents the absolute value of the inverse of f (x).
Preferably, the learning, by the generator in the modified GAN network model, the distribution rule of the fault sample data and the generating sample data in combination with the noise vector, and when the generating sample data passes through the determination of the arbiter, setting the generating sample data as fault extended sample data specifically includes:
Carrying out distribution rule feature extraction on the fault sample data through a generator in the improved GAN network model to obtain first distribution rule data;
carrying out distribution rule feature extraction on the input noise signals through a generator in the improved GAN network model to obtain second distribution rule data;
Establishing a mapping relation from the second distribution rule data to the first distribution rule data through a deep neural network to obtain generated sample data based on the mapping relation;
and when the generated sample data passes the judgment of the discriminator, setting the generated sample data as fault expansion sample data.
Preferably, the process for establishing the SVM classifier model specifically includes:
initializing a plurality of groups of firework particle populations, wherein each group of firework particle populations corresponds to a group of SVM parameters;
According to the fitness of each firework particle population, calculating the explosion range, the explosion spark quantity and the inertia mass of each firework particle population, and applying attractive force to the firework particle population corresponding to the maximum inertia mass as attractive force particles, so that the rest firework particle populations move under the attractive force action of the attractive force particles;
According to SVM parameters corresponding to each group of firework particle population, combining an initial SVM classifier model, respectively calculating classification accuracy of the SVM classifier model in different SVM parameters, optimizing each group of firework particle population according to a firework algorithm based on the classification accuracy, and outputting the current optimal firework particle population when a preset optimizing termination condition is met so as to update parameters of the SVM classifier model based on the SVM parameters of the optimal firework particle population.
Preferably, the classification objective function of the SVM classifier model is specifically:
；
Wherein x i is the i element of the cannula sample data set, y i is the classification identifier corresponding to the i element of the cannula sample data set, x j is the j element of the cannula sample data set, y j is the classification identifier corresponding to the j element of the cannula sample data set, a i is the Lagrange multiplier of the i element of the cannula sample data set, a j is the Lagrange multiplier of the j element of the cannula sample data set, and m is the image number of the cannula sample data set.
Meanwhile, a second aspect of the present application provides an OIP casing health state evaluation model modeling apparatus, including:
The image acquisition and classification unit is used for acquiring sample data of the OIP sleeve and classifying the sample data according to the state of the OIP sleeve corresponding to each sample data;
an expansion sample generation unit, configured to input fault sample data as a real data set to a modified GAN network model through the preset modified GAN network model, learn a distribution rule of the fault sample data through a generator in the modified GAN network model, and combine a noise vector to obtain generated sample data, and when the generated sample data passes through the determination of a discriminator, set the generated sample data as fault expansion sample data so as to be integrated into the fault sample data;
The sleeve total sample output unit is used for outputting a sleeve sample data total set of the OIP sleeve when the number relation of the fault sample data and the non-fault sample data corresponding to each fault type meets the preset number relation condition;
The classification model training unit is used for inputting the sleeve sample data total set into an SVM classifier model, training the SVM classifier model, and obtaining an OIP sleeve health state evaluation model when the training progress of the SVM classifier model meets the preset training termination condition.
Preferably, the method further comprises: an SVM classifier model building unit for:
initializing a plurality of groups of firework particle populations, wherein each group of firework particle populations corresponds to a group of SVM parameters;
According to the fitness of each firework particle population, calculating the explosion range, the explosion spark quantity and the inertia mass of each firework particle population, and applying attractive force to the firework particle population corresponding to the maximum inertia mass as attractive force particles, so that the rest firework particle populations move under the attractive force action of the attractive force particles;
According to SVM parameters corresponding to each group of firework particle population, combining an initial SVM classifier model, respectively calculating classification accuracy of the SVM classifier model in different SVM parameters, optimizing each group of firework particle population according to a firework algorithm based on the classification accuracy, and outputting the current optimal firework particle population when a preset optimizing termination condition is met so as to update parameters of the SVM classifier model based on the SVM parameters of the optimal firework particle population.
A third aspect of the present application provides an OIP casing health status assessment model modeling terminal, including: a memory and a processor;
the memory is used for storing program codes corresponding to the modeling method of the OIP sleeve health state assessment model provided by the first aspect of the application;
the processor is configured to execute the program code to implement an OIP casing health assessment model modeling method as provided in the first aspect of the present application.
A fourth aspect of the present application provides a computer readable storage medium having stored therein program code corresponding to an OIP casing health assessment model modeling method as provided in the first aspect of the present application.
From the above technical solutions, the embodiment of the present application has the following advantages:
According to the technical scheme provided by the application, firstly, sample data obtained from an OIP sleeve of a real transformer are based on the sample data and are classified according to the OIP sleeve states corresponding to the sample data, non-fault sample data and sample data with different fault types can be obtained, then the fault sample data is input into a preset GAN network model, a GAN algorithm is adopted to expand abnormal samples in an OIP sleeve information base so that the number of the abnormal samples is matched with the data quantity in a normal state, the sample quantity balance is ensured to ensure the stability of the learning effect of a constructed OIP sleeve health state evaluation model, the output sleeve sample data total set is input into an SVM classifier model, training is carried out on the SVM classifier model, and the OIP sleeve health state evaluation model can be obtained after training is completed, so that the technical problem that the evaluation and judgment accuracy of the existing transformer sleeve health state evaluation mode is unstable is solved.
Drawings
In order to more clearly illustrate the embodiments of the application or the technical solutions of the prior art, the drawings which are used in the description of the embodiments or the prior art will be briefly described, it being obvious that the drawings in the description below are only some embodiments of the application, and that other drawings can be obtained from these drawings without inventive faculty for a person skilled in the art.
Fig. 1 is a schematic flow chart of an embodiment of a modeling method for an OIP casing health state evaluation model provided by the application.
Fig. 2 is a schematic flow chart of a process for establishing an SVM classification model in the modeling method of the OIP sleeve health state assessment model provided by the application.
Fig. 3 is a schematic overall logic flow diagram of an embodiment of a modeling method for an OIP casing health state evaluation model provided by the present application.
Fig. 4 is a schematic structural diagram of an embodiment of an OIP casing health state evaluation model modeling apparatus provided by the present application.
Fig. 5 is a schematic structural diagram of an embodiment of a modeling terminal of an OIP casing health state evaluation model provided by the application.
Detailed Description
In view of the technical problem that the existing transformer bushing fault recognition mode based on the multi-source parameters and the BP neural network is unstable in evaluation judgment accuracy in practical application, the applicant finds that although the existing evaluation mode adopts an extended sample type improvement direction, the evaluation accuracy is improved to a certain extent by constructing a neural network model and fusing the multi-source parameter evaluation mode, the balance problem of different sample parameters is not considered, the number of normal bushings in practical engineering is far more than that of the fault bushings, so that the data of the fault bushings are relatively sparse, when the number of samples of the fault and non-fault bushings of a training model is unbalanced, the phenomenon that the evaluation judgment accuracy is reduced due to overfitting occurs when machine learning is classified is easy, and the technical problem that the evaluation judgment accuracy is unstable in practical application is caused.
The embodiment of the application provides an OIP sleeve health state evaluation method, an OIP sleeve health state evaluation device, an OIP sleeve health state evaluation terminal and an OIP sleeve health state evaluation medium, which are used for solving the technical problem that the evaluation and judgment accuracy is unstable in the existing transformer sleeve health state evaluation mode.
In order to make the objects, features and advantages of the present application more comprehensible, the technical solutions in the embodiments of the present application are described in detail below with reference to the accompanying drawings, and it is apparent that the embodiments described below are only some embodiments of the present application, but not all embodiments of the present application. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.
Firstly, the detailed description of the embodiment of the OIP sleeve health state assessment method provided by the application is as follows:
referring to fig. 1, the embodiment provides a modeling method for an OIP casing health state evaluation model, which includes:
and 101, acquiring sample data of the OIP sleeve, and classifying the sample data according to the state of the OIP sleeve corresponding to each sample data.
It should be noted that, first, sample data of these OIP bushings are obtained from real transformer OIP bushings, and the sample data are classified according to OIP bushing states corresponding to the respective sample data, for example DGA and FDS analyses are performed on bushings shipped during 1990-2017 of the power grid 254, DGA measurement gas includes H 2, C2H2, CH4, and FDS analyses are performed on measurements of bushing dielectric loss tan δ at frequencies of 0.001Hz, 0.01Hz, 0.1Hz, 1Hz, 10Hz, 50Hz, etc., to primarily determine that these bushings belong to abnormal bushings or normal bushings. In addition, in this embodiment, the casing pipe of which 13 abnormal casings and 3 normal retired casings are extracted from 254 casings is disassembled, and the casings are roughly classified into the following four types according to the internal characteristics of the disassembled casing pipe: x wax closely related to the occurrence of partial discharge is distributed in the first sleeve, and corresponding black discharge marks are also found; the second type of sleeve, namely 3 normally retired sleeves, is not abnormal in the sleeve; the third type of sleeve has serious phenomena of damp and rust, and meanwhile, breakdown discharge marks which hurt a capacitor core are found; the welding position of the fourth type sleeve head connecting sheet is provided with a discharge blackening trace, but the capacitor core is not damaged. According to the above four kinds of phenomena, the OIP bushing insulation state can be correspondingly classified into the following four kinds:
① Partial discharge faults exist in the sleeve;
② No discharge fault exists in the sleeve;
③ Serious discharge faults which damage the capacitor core exist in the sleeve;
④ The inside of the sleeve has discharge faults which do not damage the capacitor core.
Step 102, through a preset improved GAN network model, fault sample data is input into the improved GAN network model as a real data set, so that a generator in the improved GAN network model learns the distribution rule of the fault sample data, and combines with a noise vector to obtain generated sample data, and when the generated sample data passes through the judgment of a discriminator, the generated sample data is set as fault expansion sample data so as to be integrated into the fault sample data.
It should be noted that GAN is a deep learning framework composed of a generator and a arbiter, which compete with each other and learn. Wherein the generator is used for learning potential distribution rules of the real sample data and generating new data based on the input noise vector; the discriminator is used for judging whether the input data is real data or generated data. When the OIP sleeve anomaly data is input as a real dataset, the eigenvectors of each sample are DGA and FDS data for a single sleeve. The real abnormal data of the OIP sleeve is recorded as X r={x1,x2, …,xm, m represents the number of samples of the real abnormal data of the OIP sleeve, the distribution relation existing in X r is recorded as P r, and the distribution satisfied by the noise vector z input to the generator is recorded as P z. The noise z is input into a generator, a mapping relation between P z and P r is established through a deep neural network, the generated fault data of the sleeve pipe which approximately meets the distribution of P r is output, and when the generated sample data can be judged by a judging device, the generated sample data is set as fault expansion sample data so as to be integrated into the fault sample data.
The objective of the generator G is to generate data that approximates the real sample to the greatest extent, so that the arbiter cannot determine whether the input is the real sample or the generated sample, and the objective function V G is defined as:
（1）
the smaller the value of V G is, the closer the distribution of the generated sample data is to the distribution P r of the real data.
The objective of the discriminator is to distinguish the difference between the sample data generated by the generator and the real sample data, the input of the discriminator comprises the real sample data and the generated sample data, the output is the probability that the input data belongs to the real sample data, and the objective function V D is defined as:
（2）
The larger the value of V D, the more discriminatory the actual sample data and the generated sample data the discriminator has.
The training process of the GAN is a process that a generator and a discriminator play games with each other and resist each other, the generated sample data of the generator and the real sample data are more and more similar, and the capability of the discriminator for distinguishing the real sample data from the generated sample data is more and more powerful. The objective function of this process is defined as V G, D:
（3）
In some embodiments, the loss function of a conventional GAN is calculated based on Jensen-Shannon distance, which is prone to generator gradient instability and even gradient extinction. In order to solve the problem, in this embodiment, to further improve the effect of generating fault data expansion of the countermeasure network model, WGAN is preferably adopted as a more stable and robust data generation model, and meanwhile, a waserstein distance formula is introduced as a distribution similarity formula, and the distribution similarity calculated based on the formula is used as a measurement basis to solve the problems of instability of GAN training, gradient disappearance and the like. The expression of Wasserstein distance is:
（4）
In the formula (4) For distribution similarity between the failure sample data and the generated sample data, P r is edge distribution data of the failure sample data, P g is edge distribution data of the generated sample data,Representing a set of all combined distributions of the edge distributions P r and P g, from which (x, x ') - δ can be sampled for each distribution δ to obtain a real casing fault data x and a generated casing fault data x'; the I x-x' I represents the distance of the real casing fault data x and the generated casing fault data y; e (x,y)~δ [ ||x-x' | ] represents the expected value of sample versus distance for the joint distribution delta,Representing the lower bound taken by the expected value in all possible joint distributions. The smaller the value of W (P r,Pg), the higher the distribution similarity between the actual casing failure data and the generated casing failure data. In general, (4) is difficult to solve, and a Kantorovich-Rubinstein dual form is generally adopted:
（5）
In the formula (5), K represents the number of fault sample data x, K represents an upward-bound value in the absolute value of the derivative of the function f (x) when the function f (x) satisfies the K-lipschitz continuous condition, i f L i represents the absolute value of the inverse of f (x), i f L i.ltoreq.k represents the function f (x) satisfies the K-lipschitz continuous condition, the absolute value of the derivative thereof has an upward-bound, it is to be noted that f (x) has no explicit function expression, i f L i i.ltoreq.k is a limitation on her, and f (x) is satisfied. WGAN approximates the Wasserstein distance by a parameter range limited arbiter and minimizes the Wasserstein distance by continuously optimizing the generator, thereby effectively improving the distribution similarity of the generated casing fault data and the real casing fault data.
And 103, outputting a sleeve sample data total set of the OIP sleeve when the number relation of the fault sample data and the non-fault sample data corresponding to each fault type meets the preset number relation condition.
It should be noted that, when the number relationship between the fault sample data and the non-fault sample data corresponding to each fault type satisfies a preset number relationship condition, for example, when the number of each type of the 1 st, 3 rd and 4 th type of sleeve data in the abnormal state is expanded to be consistent with the number of the type 2 normal state, the sleeve sample data total set T for performing subsequent classification training can be output.
Classification between OIP sleeve samples is achieved by mapping the data within the sample set from the original space into a higher-dimensional feature space and constructing a classification hyperplane H therein based on the WGAN obtained OIP sleeve sample set t= { x i,yi |i=1, …, m }. The hyperplane can be expressed as:
（6）
In the formula (6), w is a normal vector for determining the hyperplane direction. Let two of the hyperplanes H 1、H2 be denoted as:
（7）
The distance of H 1、H2 is referred to as the classification interval d h, with d h =2 i/w, the term "w" represents the norm of the normal vector w. When the classification interval is the largest, the classification hyperplane at the moment is called as the optimal classification hyperplane, and the sample set T is provided with:
（8）
The relaxation variable xi i is introduced to measure the difference value between the SVM output value and the class identifier of the OIP sleeve sample set, and the process of solving the optimal classification hyperplane is to solve the minimum value of the following formula:
（9）
C in equation (9) is a penalty factor that controls the degree of sample classification error penalty. Lagrange multiplier a i was introduced and formula (9) was converted to according to Karush-Kuhn-Tucker conditions:
（10）
In order for SVM to better address the problem of classification of non-linearities, a kernel function is typically introduced instead of the inner product operation in equation (10), where the Radial Basis Function (RBF) is most widely used because of its wide applicability and the need for excessive scaling, which can be expressed as:
（11）
then equation (10) translates into:
（12）
The optimal classification function can be obtained by solving equation (12).
Further, as shown in fig. 2 and 3, the process of building the SVM classifier model according to the present embodiment specifically includes:
Step 1001, initializing a plurality of firework particle populations, wherein each firework particle population corresponds to a set of SVM parameters.
Step 1002, according to the fitness of each firework particle population, calculating the explosion range, the explosion spark quantity and the inertial mass of each firework particle population, and applying attractive force to the firework particle population corresponding to the maximum inertial mass as attractive force particles, so that the rest firework particle populations move under the attractive force action of the attractive force particles.
Step 1003, respectively calculating classification accuracy of the SVM classifier model in different SVM parameters according to SVM parameters corresponding to each group of firework particle populations and combining with an initial SVM classifier model, optimizing each group of firework particle populations according to a firework algorithm based on the classification accuracy, and outputting the current optimal firework particle populations when preset optimizing termination conditions are met so as to update parameters of the SVM classifier model based on the SVM parameters of the optimal firework particle populations.
It should be noted that the performance of the SVM has a great relationship with parameters of the kernel function (e.g., σ in RBF) and the penalty coefficient C, and the optimal parameters required for classifying different data models using the SVM are not the same. In order to enable the SVM model to have better casing state data classification capability, an intelligent optimizing algorithm is introduced to conduct parameter optimization. When the traditional intelligent algorithm (such as a particle swarm algorithm) is optimized, the problems that a local optimal solution is easy to fall into, the internal parameters are extremely sensitive and the like exist due to the limitations of the algorithm, and therefore, the embodiment adopts a firework algorithm (fireworks algorithm with gravitational search operator, FAGSO) with a guiding operator to optimize the parameters of the SVM. Compared with the traditional optimizing algorithm, FAGSO has higher searching precision and convergence speed, the FAGSO algorithm is adopted to improve the SVM algorithm, the classification error of the SVM is set as the fitness function of FAGSO, and the error is gradually reduced in iteration, so that a FAGSO-SVM classification model with higher classification accuracy is obtained, and the state evaluation accuracy of the classification model on the expanded OIP sleeve sample information is improved.
FAGSO the optimization mainly comprises the following processes:
① N initial firework particles are randomly generated in the solution space and serve as starting points of algorithm iteration.
② For each firework particle w i, according to the adaptability f (w i), the explosion range A i of the spark generated by explosion and the spark quantity g i are calculated according to the following calculation formula:
（13）
（14）
A. m is a constant that is the magnitude of the explosion and the number of sparks generated; epsilon is a very small constant to prevent the denominator of formulas (13), (14) from becoming zero. The firework with poor adaptability has a larger explosion range and generates fewer explosion sparks, so that the algorithm has certain global optimizing capability; the firework with better adaptability has smaller explosion range and more generated explosion sparks, so as to ensure that the algorithm has certain local convergence capacity near the solution.
③ And introducing a Gaussian variation process to the firework and spark population to further improve the global optimizing capability of the algorithm.
④ For spark particles that may be out of the solution space after explosion, the mapping rules of modulo arithmetic are used to pull the spark particles out of the solution space back into the feasible region.
⑤ For each particle, calculating its inertial mass according to its fitness:
（15）
Particles with poor fitness have smaller inertial mass and particles with better fitness have larger inertial mass. 2 XN particles with the largest mass are selected from the particle population to be used as excellent particles for exerting attractive force, and other particles move towards the excellent particles under the action of the attractive force, so that the particles with better adaptability have the opportunity to perform position information interaction with the particles and update the positions of the particles, and the particles are prevented from being trapped into local optimum.
⑥ 1 Particle with optimal fitness and N-1 particles randomly selected through roulette are selected from a particle set to serve as initial firework particles of the next iteration, the probability of selecting the particle w i is based on the sum of the distances between the particle and other particles, and the probability of selecting the more sparse particle is higher, so that the diversity of the population is increased. The probability P that the particle w i is selected (w i) is expressed as:
（16）
Based on the global optimizing capability of FAGSO, the SVM can obtain the optimal classifying capability under the corresponding data sample by setting the classifying error of the SVM as the fitness function of FAGSO and searching the minimum value of the fitness and the corresponding optimal parameter in the training iteration process. The objective function of FAGSO-SVM may be set as:
（17）
In the formula (17), f i represents the classification identifier of the SVM, and y i represents the actual classification identifier.
And 104, inputting the sleeve sample data total set into an SVM classifier model, training the SVM classifier model, and obtaining an OIP sleeve health state evaluation model when the training progress of the SVM classifier model meets the preset training termination condition.
Finally, as shown in fig. 3, the sleeve sample data aggregate is input into the SVM classifier model, the SVM classifier model is trained, and when the training progress of the SVM classifier model meets the preset training termination condition, if the training iteration number reaches the maximum value or the convergence degree of the training result reaches the preset convergence threshold, the training is completed, and the OIP sleeve health state assessment model is obtained and can be used for assessing the OIP sleeve health state.
The above is a detailed description of an embodiment of a modeling method for an OIP casing health state evaluation model provided by the present application, and the following is a detailed description of an embodiment of an apparatus for modeling an OIP casing health state evaluation model provided by the present application.
Referring to fig. 4, the present embodiment provides an OIP casing health status evaluation model modeling apparatus, including:
An image obtaining and classifying unit 201, configured to obtain sample data of the OIP sleeves, and classify the sample data according to the OIP sleeve states corresponding to each sample data;
an extended sample generating unit 202, configured to input the fault sample data as a real data set to the modified GAN network model through a preset modified GAN network model, learn a distribution rule of the fault sample data through a generator in the modified GAN network model, and combine the noise vector to obtain generated sample data, and when the generated sample data passes through the determination of the arbiter, set the generated sample data as fault extended sample data so as to be integrated into the fault sample data;
a casing total sample output unit 203, configured to output a casing sample data total set of the OIP casing when the number relation between the fault sample data and the non-fault sample data corresponding to each fault type meets a preset number relation condition;
The classification model training unit 204 is configured to input the sleeve sample data aggregate to the SVM classifier model, train the SVM classifier model, and obtain the OIP sleeve health state assessment model when the training progress of the SVM classifier model meets a preset training termination condition.
Further, the method further comprises the following steps: an SVM classifier model building unit 200 for:
initializing a plurality of groups of firework particle populations, wherein each group of firework particle populations corresponds to a group of SVM parameters;
According to the fitness of each firework particle population, calculating the explosion range, the explosion spark quantity and the inertial mass of the firework particle population, and applying attractive force to the firework particle population corresponding to the maximum inertial mass as attractive force particles, so that the rest firework particle populations move under the attractive force action of the attractive force particles;
According to SVM parameters corresponding to each group of firework particle population, combining an initial SVM classifier model, respectively calculating classification accuracy of the SVM classifier model in different SVM parameters, optimizing each group of firework particle population according to a firework algorithm based on the classification accuracy, and outputting the current optimal firework particle population when a preset optimizing termination condition is met so as to update parameters of the SVM classifier model based on the SVM parameters of the optimal firework particle population.
In addition to the above-described embodiment of the modeling apparatus for the OIP casing health status assessment model, the present application further provides a detailed description of an embodiment of an OIP casing health status assessment model modeling terminal and an embodiment of a computer-readable storage medium related to the above-described method for modeling an OIP casing health status assessment model.
Referring to fig. 5, the embodiment provides an OIP sleeve health state evaluation model modeling terminal, wherein the types of the terminal include, but are not limited to, a personal computer, an industrial computer, a server and an embedded intelligent terminal, and the main components of the terminal include: a memory 33 and a processor 31, wherein the memory 33 and the processor 31 are connectable through a communication bus 34;
the memory 33 is configured to store program codes corresponding to an OIP casing health assessment model modeling method as provided in the above embodiment;
The processor 31 is configured to execute program code to implement an OIP casing health assessment model modeling method as provided in the above embodiments.
Meanwhile, the application also provides a computer readable storage medium, and the computer readable storage medium stores program codes corresponding to the modeling method of the OIP sleeve health state assessment model provided by the first aspect of the application.
It will be clearly understood by those skilled in the art that, for convenience and brevity of description, specific working procedures of the terminal, apparatus and unit described above may refer to corresponding procedures in the foregoing method embodiments, which are not repeated herein.
In the several embodiments provided in the present application, it should be understood that the disclosed terminal, apparatus and method may be implemented in other manners. For example, the apparatus embodiments described above are merely illustrative, e.g., the division of the units is merely a logical function division, and there may be additional divisions when actually implemented, e.g., multiple units or components may be combined or integrated into another system, or some features may be omitted or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices or units, which may be in electrical, mechanical or other form.
The terms "first," "second," "third," "fourth," and the like in the description of the application and in the above figures, if any, are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments of the application described herein may be implemented, for example, in sequences other than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
It should be understood that in the present application, "at least one (item)" means one or more, and "a plurality" means two or more. "and/or" for describing the association relationship of the association object, the representation may have three relationships, for example, "a and/or B" may represent: only a, only B and both a and B are present, wherein a, B may be singular or plural. The character "/" generally indicates that the context-dependent object is an "or" relationship. "at least one of" or the like means any combination of these items, including any combination of single item(s) or plural items(s). For example, at least one (one) of a, b or c may represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", wherein a, b, c may be single or plural.
The units described as separate units may or may not be physically separate, and units shown as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
In addition, each functional unit in the embodiments of the present invention may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit. The integrated units may be implemented in hardware or in software functional units.
The integrated units, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied essentially or in part or all of the technical solution or in part in the form of a software product stored in a storage medium, including instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a usb disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
The above embodiments are only for illustrating the technical solution of the present application, and not for limiting the same; although the application has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims (7)

1. An OIP casing health state assessment model modeling method, which is characterized by comprising the following steps:
acquiring sample data of the OIP sleeve, and classifying the sample data according to the state of the OIP sleeve corresponding to each sample data;
Inputting fault sample data as a real data set to the improved GAN network model through a preset improved GAN network model, learning a distribution rule of the fault sample data through a generator in the improved GAN network model, and combining noise vectors to obtain generated sample data, wherein when the generated sample data passes through judgment of a discriminator, the generated sample data is set as fault expansion sample data so as to integrate the fault expansion sample data into the fault sample data;
Outputting a sleeve sample data total set of the OIP sleeve when the number relation of the fault sample data and the non-fault sample data corresponding to each fault type meets the preset number relation condition;
Inputting the sleeve sample data total set into an SVM classifier model, training the SVM classifier model, and obtaining an OIP sleeve health state assessment model when the training progress of the SVM classifier model meets the preset training termination condition;
Wherein when the generated sample data passes the judgment of the discriminator, setting the generated sample data as the fault expansion sample data specifically comprises:
the discriminator calculates the distribution similarity between the generated sample data and the fault sample data through a distribution similarity formula based on WGAN networks, and when the distribution similarity meets a preset similarity condition, the generated sample data is set as fault expansion sample data;
The distribution similarity formula specifically comprises the following steps:
；
In the method, in the process of the invention, For the distribution similarity between the fault sample data and the generated sample data, P r is the edge distribution data of the fault sample data, P g is the edge distribution data of the generated sample data, K represents the number of faulty sample data x, when the function f (x) satisfies the K-lipschitz continuous condition, the absolute value of the inverse f (x) is represented by the absolute value of the derivative, i f L i, of the upper bound value;
The generating means for learning the distribution rule of the fault sample data by the generating means in the modified GAN network model, and obtaining the generated sample data by combining a noise vector, and when the generated sample data passes the determination of the determining means, setting the generated sample data as fault extended sample data specifically includes:
Carrying out distribution rule feature extraction on the fault sample data through a generator in the improved GAN network model to obtain first distribution rule data;
carrying out distribution rule feature extraction on the input noise signals through a generator in the improved GAN network model to obtain second distribution rule data;
Establishing a mapping relation from the second distribution rule data to the first distribution rule data through a deep neural network to obtain generated sample data based on the mapping relation;
and when the generated sample data passes the judgment of the discriminator, setting the generated sample data as fault expansion sample data.
2. The modeling method of an OIP casing health state evaluation model according to claim 1, wherein the process of establishing the SVM classifier model specifically comprises:
initializing a plurality of groups of firework particle populations, wherein each group of firework particle populations corresponds to a group of SVM parameters;
According to the fitness of each firework particle population, calculating the explosion range, the explosion spark quantity and the inertia mass of each firework particle population, and applying attractive force to the firework particle population corresponding to the maximum inertia mass as attractive force particles, so that the rest firework particle populations move under the attractive force action of the attractive force particles;
According to SVM parameters corresponding to each group of firework particle population, combining an initial SVM classifier model, respectively calculating classification accuracy of the SVM classifier model in different SVM parameters, optimizing each group of firework particle population according to a firework algorithm based on the classification accuracy, and outputting the current optimal firework particle population when a preset optimizing termination condition is met so as to update parameters of the SVM classifier model based on the SVM parameters of the optimal firework particle population.
3. The modeling method of an OIP casing health state assessment model according to claim 2, wherein the classification objective function of the SVM classifier model is specifically:
；
Wherein x i is the i element of the cannula sample data set, y i is the classification identifier corresponding to the i element of the cannula sample data set, x j is the j element of the cannula sample data set, y j is the classification identifier corresponding to the j element of the cannula sample data set, a i is the Lagrange multiplier of the i element of the cannula sample data set, a j is the Lagrange multiplier of the j element of the cannula sample data set, and m is the image number of the cannula sample data set.
4. An OIP casing health state assessment model modeling apparatus, comprising:
The image acquisition and classification unit is used for acquiring sample data of the OIP sleeve and classifying the sample data according to the state of the OIP sleeve corresponding to each sample data;
an expansion sample generation unit, configured to input fault sample data as a real data set to a modified GAN network model through the preset modified GAN network model, learn a distribution rule of the fault sample data through a generator in the modified GAN network model, and combine a noise vector to obtain generated sample data, and when the generated sample data passes through the determination of a discriminator, set the generated sample data as fault expansion sample data so as to be integrated into the fault sample data;
The sleeve total sample output unit is used for outputting a sleeve sample data total set of the OIP sleeve when the number relation of the fault sample data and the non-fault sample data corresponding to each fault type meets the preset number relation condition;
The classification model training unit is used for inputting the sleeve sample data total set into an SVM classifier model, training the SVM classifier model, and obtaining an OIP sleeve health state evaluation model when the training progress of the SVM classifier model meets the preset training termination condition;
Wherein when the generated sample data passes the judgment of the discriminator, setting the generated sample data as the fault expansion sample data specifically comprises:
the discriminator calculates the distribution similarity between the generated sample data and the fault sample data through a distribution similarity formula based on WGAN networks, and when the distribution similarity meets a preset similarity condition, the generated sample data is set as fault expansion sample data;
The distribution similarity formula specifically comprises the following steps:
；
In the method, in the process of the invention, For the distribution similarity between the fault sample data and the generated sample data, P r is the edge distribution data of the fault sample data, P g is the edge distribution data of the generated sample data, K represents the number of faulty sample data x, when the function f (x) satisfies the K-lipschitz continuous condition, the absolute value of the inverse f (x) is represented by the absolute value of the derivative, i f L i, of the upper bound value;
The generating means for learning the distribution rule of the fault sample data by the generating means in the modified GAN network model, and obtaining the generated sample data by combining a noise vector, and when the generated sample data passes the determination of the determining means, setting the generated sample data as fault extended sample data specifically includes:
Carrying out distribution rule feature extraction on the fault sample data through a generator in the improved GAN network model to obtain first distribution rule data;
carrying out distribution rule feature extraction on the input noise signals through a generator in the improved GAN network model to obtain second distribution rule data;
Establishing a mapping relation from the second distribution rule data to the first distribution rule data through a deep neural network to obtain generated sample data based on the mapping relation;
and when the generated sample data passes the judgment of the discriminator, setting the generated sample data as fault expansion sample data.
5. The OIP casing health assessment model modeling apparatus of claim 4, further comprising: an SVM classifier model building unit for:
initializing a plurality of groups of firework particle populations, wherein each group of firework particle populations corresponds to a group of SVM parameters;
According to the fitness of each firework particle population, calculating the explosion range, the explosion spark quantity and the inertia mass of each firework particle population, and applying attractive force to the firework particle population corresponding to the maximum inertia mass as attractive force particles, so that the rest firework particle populations move under the attractive force action of the attractive force particles;
According to SVM parameters corresponding to each group of firework particle population, combining an initial SVM classifier model, respectively calculating classification accuracy of the SVM classifier model in different SVM parameters, optimizing each group of firework particle population according to a firework algorithm based on the classification accuracy, and outputting the current optimal firework particle population when a preset optimizing termination condition is met so as to update parameters of the SVM classifier model based on the SVM parameters of the optimal firework particle population.
6. An OIP casing health assessment model modeling terminal, comprising: a memory and a processor;
The memory is used for storing program codes corresponding to an OIP casing health state assessment model modeling method according to any one of claims 1 to 3;
The processor is configured to execute the program code to implement an OIP casing health assessment model modeling method according to any one of claims 1 to 3.
7. A computer readable storage medium, wherein the computer readable storage medium stores program code corresponding to an OIP casing health assessment model modeling method according to any one of claims 1 to 3.
CN202410355139.1A 2024-03-27 2024-03-27 Modeling method, device, terminal and medium for OIP sleeve health state evaluation model Active CN117951588B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202410355139.1A CN117951588B (en) 2024-03-27 2024-03-27 Modeling method, device, terminal and medium for OIP sleeve health state evaluation model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202410355139.1A CN117951588B (en) 2024-03-27 2024-03-27 Modeling method, device, terminal and medium for OIP sleeve health state evaluation model

Publications (2)

Publication Number Publication Date
CN117951588A CN117951588A (en) 2024-04-30
CN117951588B true CN117951588B (en) 2024-07-30

Family

ID=90805170

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202410355139.1A Active CN117951588B (en) 2024-03-27 2024-03-27 Modeling method, device, terminal and medium for OIP sleeve health state evaluation model

Country Status (1)

Country Link
CN (1) CN117951588B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118171180B (en) * 2024-05-15 2024-07-23 山东理工职业学院 Equipment state prediction method and device based on artificial intelligence

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115796843A (en) * 2022-12-12 2023-03-14 南方电网科学研究院有限责任公司 Operation and maintenance strategy generation method of power transformer and related device

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
ES2660013T3 (en) * 2010-08-13 2018-03-20 Respiratory Motion, Inc. Devices and methods for monitoring respiratory variation by measuring respiratory volumes, movement and variability
CN115222111A (en) * 2022-06-29 2022-10-21 南方电网科学研究院有限责任公司 Transformer health state evaluation method, device, equipment and readable storage medium

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115796843A (en) * 2022-12-12 2023-03-14 南方电网科学研究院有限责任公司 Operation and maintenance strategy generation method of power transformer and related device

Also Published As

Publication number Publication date
CN117951588A (en) 2024-04-30

Similar Documents

Publication Publication Date Title
CN106503689A (en) Neutral net local discharge signal mode identification method based on particle cluster algorithm
CN117951588B (en) Modeling method, device, terminal and medium for OIP sleeve health state evaluation model
CN111914253B (en) Method, system, equipment and readable storage medium for intrusion detection
CN106060008B (en) A kind of network intrusions method for detecting abnormality
CN108875783A (en) A kind of extreme learning machine Diagnosis Method of Transformer Faults towards unbalanced dataset
CN113792754B (en) Converter transformer DGA online monitoring data processing method for firstly removing abnormal state and then repairing
CN111144548A (en) Method and device for identifying working condition of pumping well
CN109918720A (en) Transformer fault diagnosis method based on krill swarm optimization support vector machine
CN112362292B (en) A method for anomaly detection in wind tunnel test data
CN105137354B (en) One kind is based on neutral net electrical fault detection method
CN113162037B (en) Power system transient voltage stability self-adaptive evaluation method and system
CN108090615A (en) Low-limit frequency Forecasting Methodology after electric power system fault based on cross entropy integrated study
CN115729761A (en) Hard disk fault prediction method, system, device and medium
CN117235565A (en) A method and device for constructing a transformer fault diagnosis model
CN113949549A (en) Real-time traffic anomaly detection method for intrusion and attack defense
CN111725802B (en) Transient stability judgment method of AC-DC hybrid power grid based on deep neural network
Shi et al. Dissolved gas analysis for power transformer fault diagnosis combining domain knowledge and capsule network
CN110097120A (en) Network flow data classification method, equipment and computer storage medium
CN112595918A (en) Low-voltage meter reading fault detection method and device
CN109918905B (en) Behavior inference model generation device and behavior inference model generation method
CN118230198A (en) Power transmission and transformation equipment inspection system and method based on visible light image recognition technology
CN117350170A (en) Nonlinear oscillation analysis method based on KOOPMAN deep neural network
CN116522121A (en) An Online Transformer Fault Diagnosis Method Under Unbalanced Small Sample Conditions
CN116204771A (en) Power system transient stability key feature selection method, device and product
Fu et al. Fault Diagnosis of an Excitation System Using a Fuzzy Neural Network Optimized by a Novel Adaptive Grey Wolf Optimizer.

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
GR01 Patent grant
GR01 Patent grant