CN113379251A - IFOA-SVM-based high-voltage switch cabinet state evaluation method - Google Patents

IFOA-SVM-based high-voltage switch cabinet state evaluation method Download PDF

Info

Publication number
CN113379251A
CN113379251A CN202110654396.1A CN202110654396A CN113379251A CN 113379251 A CN113379251 A CN 113379251A CN 202110654396 A CN202110654396 A CN 202110654396A CN 113379251 A CN113379251 A CN 113379251A
Authority
CN
China
Prior art keywords
data
voltage switchgear
fruit fly
svm
kernel function
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202110654396.1A
Other languages
Chinese (zh)
Inventor
金学成
周敦有
陈博
王鑫
吴麒
丁年礼
张文安
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.)
Zhejiang University of Technology ZJUT
Original Assignee
Zhejiang University of Technology ZJUT
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 Zhejiang University of Technology ZJUT filed Critical Zhejiang University of Technology ZJUT
Priority to CN202110654396.1A priority Critical patent/CN113379251A/en
Publication of CN113379251A publication Critical patent/CN113379251A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—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
    • 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
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Human Resources & Organizations (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Economics (AREA)
  • Strategic Management (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Health & Medical Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Development Economics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Educational Administration (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Marketing (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Primary Health Care (AREA)
  • Biomedical Technology (AREA)
  • Computational Linguistics (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Biophysics (AREA)
  • Water Supply & Treatment (AREA)
  • Public Health (AREA)
  • Game Theory and Decision Science (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

一种基于IFOA‑SVM的高压开关柜状态评估方法，首先，采集高压开关柜运行数据，通过删除极端异常值和补全缺失值的方法对数据进行预处理；以现场调研结果和专家经验为依据，选取评价高压开关柜的特征量为局放值、超声波值、红外诊断、电缆接头温度和母排温度；接着，结合多项式函数和高斯径向基函数构造混合核函数，将原始输入空间映射到高维特征空间；然后，利用改进果蝇优化算法寻优核函数比例系数、核函数宽度参数和惩罚因子；最后，通过训练OVO‑SVM高压开关柜状态估计分类器，实现高压开关柜状态的非线性分类，即高压开关柜状态评估。本发明利用多源信息对配电高压开关柜进行状态综合评价，理论框架简单，易于实现且泛化能力强。

Figure 202110654396

A high-voltage switchgear state assessment method based on IFOA‑SVM. First, collect high-voltage switchgear operation data, and preprocess the data by deleting extreme outliers and filling missing values; based on on-site investigation results and expert experience , select the feature quantities to evaluate the high-voltage switchgear as partial discharge value, ultrasonic value, infrared diagnosis, cable joint temperature and busbar temperature; then, combine the polynomial function and Gaussian radial basis function to construct a mixed kernel function, and map the original input space to High-dimensional feature space; then, the improved fruit fly optimization algorithm is used to optimize the kernel function scale coefficient, kernel function width parameter and penalty factor; finally, by training the OVO‑SVM high-voltage switchgear state estimation classifier, the non-discrimination of the state of the high-voltage switchgear is realized. Linear classification, i.e. high-voltage switchgear condition assessment. The invention uses multi-source information to comprehensively evaluate the state of the distribution high-voltage switch cabinet, has a simple theoretical framework, is easy to implement and has strong generalization ability.

Figure 202110654396

Description

IFOA-SVM-based high-voltage switch cabinet state evaluation method
Technical Field
The invention is applied to the field of high-voltage switch cabinet state evaluation, and relates to a high-voltage switch cabinet state evaluation method based on an improved fruit fly optimization algorithm and an improved support vector machine (IFOA-SVM).
Background
The high-voltage switch cabinet is one of main equipment of a power distribution system and plays a role in controlling and protecting other power equipment in an intelligent power grid. At present, the number of high-voltage switch cabinets installed is huge, the maintenance task is heavy, the fault of the high-voltage switch cabinet becomes a main factor influencing the safe and stable economical operation of the power distribution network, and the health condition assessment of the high-voltage switch cabinet is just an urgent research focus.
Because the internal structure of the high-voltage switch cabinet is very complicated, and the mutual influence among all the components is large, the current state and the future state of the equipment cannot be accurately evaluated through the traditional single analysis on all characteristic quantities influencing the state of the high-voltage switch cabinet. The method has the advantages that the multi-dimensional characteristic quantity of the high-voltage switch cabinet is selected by effectively integrating multiple information sources such as online monitoring, live detection and power failure experiments, and the key for improving the state evaluation precision of the high-voltage switch cabinet is achieved.
The Support Vector Machine (SVM) is a novel machine learning method based on a statistical learning theory, and has a plurality of specific advantages for solving the problems of non-linearity and high-dimensional pattern recognition, so that the method for estimating the state of the high-voltage switch cabinet by using an SVM classification model is an effective method. The traditional method for evaluating the state based on the SVM classification model mainly improves the classification recognition precision and efficiency by searching a better single kernel function or optimizing related parameters of the SVM through a modern optimization algorithm, cannot give consideration to the improvement of the SVM kernel function structure and the parameter optimization algorithm at the same time, and has certain limitation. In order to improve the classification effect of the SVM, the method provided by the invention optimizes the relevant parameters of the SVM by constructing a mixed kernel function and using an improved drosophila optimization algorithm so as to improve the accuracy of state estimation of the high-voltage switch cabinet.
Disclosure of Invention
In order to overcome the defects of the existing high-voltage switch cabinet state evaluation technology, the invention provides a high-voltage switch cabinet state evaluation method based on an improved fruit fly optimization algorithm and an improved support vector machine (IFOA-SVM).
The technical scheme adopted by the invention for solving the technical problems is as follows:
a high-voltage switch cabinet state evaluation method based on an improved fruit fly optimization algorithm and a support vector machine (IFOA-SVM), comprising the following steps:
1) the method comprises the following steps of (1) preprocessing the state evaluation data acquisition of the high-voltage switch cabinet, wherein the process comprises the following steps:
data acquisition is carried out through an online monitoring device and various sensors which are equipped with a high-voltage switch cabinet, firstly, the mean value of the same type of data and the mean value of measured values before and after the missing position data are used for completing the missing value, and secondly, the extreme abnormal value is defined through the comparison of the data of the abnormal condition and the normal condition, so that the extreme abnormality is eliminated;
2) the method comprises the following steps of (1) normalizing the state evaluation data of the high-voltage switch cabinet, wherein the process comprises the following steps:
carrying out normalization processing on the data of the high-voltage switch cabinet processed by the abnormal values and the missing values by adopting a formula (1);
Figure BDA0003112028330000021
in the formula, xiThe data of the high-voltage switch cabinet are not normalized; x is the number ofmaxThe maximum value is the maximum value in the data of the same group of the high-voltage switch cabinets which are not normalized; x is the number ofminThe minimum value in the high-voltage switch cabinet data which are not in the same group and are not normalized;
Figure BDA0003112028330000022
the data are the high-voltage switch cabinet data after normalization;
3) the inputs and outputs of the model are determined as follows:
selecting high-voltage switch cabinet characteristic quantities from multiple information sources for on-line monitoring, live detection and power failure experiments, wherein the selected characteristic quantities are an partial discharge value, an ultrasonic value, infrared diagnosis, cable joint temperature and busbar temperature, the sample data characteristics of the preprocessed high-voltage switch cabinet are used as the input of a model, and the state of the high-voltage switch cabinet is used as the output;
4) determining the SVM multi-classification strategy, wherein the process is as follows:
the OVO-SVM high-voltage switch cabinet state estimation classifier is formed by combining a one-to-one strategy and an SVM (support vector machine), when the K-type high-voltage switch cabinet state multi-classification problem is solved, a two-classification SVM sub-classifier is constructed between any two classification samples, K (K-1)/2 sub-classifiers are required to be constructed in total, and only two types of data in training samples are required to be used for generating each sub-classifier, so that the two types of data in the K-type multi-classification problem are distinguished; after constructing all sub-classifiers, combining the sub-classifiers to make decisions on the classes of new samples, and judging the classes by adopting a 'voting method', wherein each sub-classifier performs 'voting' on samples to be classified, only 1 vote can be cast for a certain sample, after traversing all the sub-classifiers, the samples are divided into the classes with the highest 'number of votes', a sub-classifier is constructed between any two classes of states based on a 'one-to-one' classification strategy state estimation classifier, and different sub-classifiers are combined together to form a final classifier;
5) dividing the state evaluation sample data of the high-voltage switch cabinet, wherein the process is as follows:
immediately dividing sample data of the high-voltage switch cabinet into a training data set and a testing data set, wherein the training data set accounts for 60% -80% and is used for supervised training learning of the model, and the rest data is used as the testing data set and is used for testing the trained classification model;
6) constructing a mixed kernel function by the following process:
for multi-dimensional features selected by the high-voltage switch cabinet, introducing a kernel function into an SVM classification model, and mapping original data features of the nonlinear high-voltage switch cabinet into a high-dimensional linear separable feature space so as to enable the high-dimensional linear separable feature space to be linearly separable; according to the characteristics of each kernel function, a mixed kernel function is constructed to replace a single kernel function, so that the state estimation accuracy is improved, namely
K1(x,z)＝((xTz)+1)
Figure BDA0003112028330000041
0≤λ≤1
In the formula, K1(x, z) is a polynomial kernel function; k2(x, z) is a radial basis kernel function; lambda is a kernel function proportionality coefficient, sigma is a radial basis kernel function width parameter, and x and z respectively correspond to original data characteristics and characteristics in a mapping space; k (x, z) is a final mixed kernel function expression;
7) substituting the new mixed kernel function obtained in the step 6) into the SVM classification model to obtain a state estimation model of the high-voltage switch cabinet, wherein the state estimation model comprises the following steps:
Figure BDA0003112028330000042
in the formula, yiIs the output of the ith example, αiIs Lagrange multiplier, b is bias, m is characteristic quantity, f (x) represents the final classification result of the classifier;
8) the parameter optimization is carried out on the SVM based on the improved drosophila optimization algorithm IFOA algorithm, and the process is as follows:
8.1) initializing IFOA parameters: defining a fruit fly population in a plane coordinate system, initializing population scale s and maximum iteration times maxgenDetermining the position coordinates of the individual drosophila; the kernel function proportionality coefficient lambda of the parameter to be optimized, the radial basis kernel function width parameter sigma and a penalty factor C, and the random initialization population position is (lambda)0，σ0，C0) And endowing the fruit fly with random movement direction and distance, thereby generating seedsThe individual positions of the fruit flies in the population are defined as l, the coordinates of the jth fruit fly are,
Figure BDA0003112028330000051
in the formula, R is a random search distance;
8.2) calculating the fitness value of the individual fruit fly optimization: substituting the position coordinates of the fruit flies into a fitness function F, evaluating the taste concentration of the individual positions of the fruit flies according to fitness values F (lambda, sigma, C), wherein the fitness values are determined according to the accuracy A of an SVM model, solving the fruit Fly closest to the food position in a population, namely solving the maximum value max F of the fitness corresponding to the individual BI of the fruit flies,
Figure BDA0003112028330000052
in the formula, DTFor all samples with correct prediction, D is a total sample, BS is an optimal concentration value, and BI is a drosophila individual;
8.3) preservation of optimal taste concentration values: storing the best fitness value BS executed for the first time in Sb, and simultaneously storing the coordinates of lambda, sigma and C to obtain
Figure BDA0003112028330000053
8.4) determining the average position coordinates of the induced learning fruit flies: sorting the fruit fly taste fitness values searched by the generation, taking two fruit fly individuals subBI1 and subBI2 next to the optimal fruit fly as induced learning individuals, respectively calculating the positions of coordinates of the induced learning individuals, and calculating an arithmetic mean value to obtain lambdae、σe、CeI.e. by
Figure BDA0003112028330000061
8.5) calculating the target position of the convergence direction of the fruit fly colony: updating the position of the fruit fly population by using the formula (8);
Figure BDA0003112028330000062
in the formula, eta is a learning factor, lambdaj′，σj′，Cj' is a new drosophila location coordinate;
in the fruit fly optimizing process, a learning factor is introduced to promote a fruit fly population to learn to the elaiopsis eligua individuals, so that the algorithm is ensured to be finally converged at the global optimal position, in the fruit fly population iterative optimizing process, the direction and the distance of population movement are guided by the elaiopsis eligua individuals, so that the fruit fly population is continuously close to the global optimal position, the influence of the elaiopsis eligua individuals of the previous iteration on the population is gradually reduced along with the increase of the iteration times T, and a fractional attenuation mechanism is adopted to set a learning factor eta and a learning factor attenuation amplitude control parameter cf；
η＝η0/(1+cf(T-1)) (9)
In the formula eta0As an initial learning factor, cfAttenuating the amplitude control parameter for the learning factor;
8.6) carrying out iterative optimization: the BS of the present generation is given to the preBS and then enters an iterative loop, the steps are repeated, the Sb value of each generation is stored, and when the algorithm reaches maxgenThe loop is ended when the set value is reached, and the position coordinates (lambda, sigma, C) of the best fruit fly individual, namely the optimizing result, are obtained
Sb＝max{BS,preBS} (10)
9) Obtaining optimal parameter combinations according to IFOA algorithm
Figure BDA0003112028330000063
Therefore, an SVM optimal model is determined, test set data are used for verifying the optimal model, and new state data of the high-voltage switch cabinet are input into the SVM classification model after data preprocessing, so that the SVM optimal model can be obtained
Compared with the prior art, the invention has the advantages that: 1) the method is different from the traditional state single information evaluation, and multi-state comprehensive evaluation is carried out on the high-voltage switch cabinet by utilizing multi-source information. 2) The method is different from the traditional SVM which only uses a single kernel function, and the accuracy of state estimation is further improved by constructing a mixed kernel function to replace the single kernel function and applying the mixed kernel function to a classifier; 3) the improved fruit fly optimization algorithm IFOA is provided, the defect that the algorithm is easy to fall into local optimum when the initial value selection of the fruit fly optimization algorithm is inappropriate is overcome, and the algorithm is guaranteed to be finally converged to the global optimum position. 4) The detection method has a simple theoretical framework, is easy to realize and has good generalization capability.
Drawings
FIG. 1 is a detailed flow chart of a method for estimating the state of a high-voltage switch cabinet based on an IFOA-SVM.
FIG. 2 is a diagram of OVO-SVM state estimation multi-classifier.
Detailed Description
In order to make the technical scheme and the design idea of the present invention clearer, the following detailed description is made with reference to the accompanying drawings.
Referring to fig. 1 and 2, a high-voltage switchgear state evaluation method based on an improved drosophila optimization algorithm and a support vector machine (IFOA-SVM), the method comprising the steps of:
1) the method comprises the following steps of (1) preprocessing the state evaluation data acquisition of the high-voltage switch cabinet, wherein the process comprises the following steps:
data acquisition is carried out through an online monitoring device and various sensors which are equipped in a high-voltage switch cabinet, and the possibility that the obtained data has a missing value and an abnormal value is considered, firstly, the missing value is completed through the mean value of the same type of data and the mean value of the measured values before and after the missing position data, and secondly, the extreme abnormal value is defined through the comparison of the data of the abnormal condition and the normal condition, so that the extreme abnormality is deleted;
2) the method comprises the following steps of (1) normalizing the state evaluation data of the high-voltage switch cabinet, wherein the process comprises the following steps:
the data normalization processing can reduce the influence of dimensions on the algorithm precision, simultaneously reduce the algorithm calculation complexity, accelerate the calculation speed and improve the high-voltage switch cabinet evaluation performance, and the high-voltage switch cabinet data processed by abnormal values and missing values is normalized by adopting the formula (1);
Figure BDA0003112028330000081
in the formula, xiThe data are the high-voltage switch cabinet data which are not normalized; x is the number ofmaxThe maximum value is the maximum value in the data of the same group of high-voltage switch cabinets which are not normalized; x is the number ofminA minimum value in the high voltage switchgear data that is not in the same group without normalization;
Figure BDA0003112028330000082
the data are the high-voltage switch cabinet data after normalization;
3) the inputs and outputs of the model are determined as follows:
in consideration of the practical feasibility of characteristic quantity acquisition, selecting and evaluating the characteristic quantity of the high-voltage switch cabinet according to on-site investigation results and expert experiences, selecting the characteristic quantity of the high-voltage switch cabinet from multiple information sources such as online monitoring, live detection, power failure experiments and the like, wherein the selected characteristic quantity is a partial discharge value, an ultrasonic wave value, infrared diagnosis, cable joint temperature and busbar temperature, the sample data characteristics of the preprocessed high-voltage switch cabinet are used as the input of a model, and the state of the high-voltage switch cabinet is used as the output;
4) determining the SVM multi-classification strategy, wherein the process is as follows:
referring to fig. 2, a OVO-SVM high-voltage switch cabinet state estimation classifier is formed by combining a one-to-one (OVO) strategy and SVM, and when the problem of multi-classification of K-class high-voltage switch cabinet states is solved, a two-classification SVM sub-classifier is constructed between any two classes of samples, and K (K-1)/2 sub-classifiers are required to be constructed. Generating each sub-classifier only needs two types of data in the training sample, so that two types of differentiation in the K-type multi-classification problem is realized; after all the sub-classifiers are constructed, the classification of a new sample is determined by combining the sub-classifiers, the classification is judged by adopting a 'voting method', each sub-classifier carries out 'voting' on the sample to be classified, only 1 ticket can be cast for a certain sample, after all the sub-classifiers are traversed, the sample is divided into the class with the highest 'number of votes', a sub-classifier is constructed between any two classes of states based on the state estimation classifier of the 'one-to-one' classification strategy, and different sub-classifiers are combined to form a final classifier;
5) dividing the state evaluation sample data of the high-voltage switch cabinet, wherein the process is as follows:
in order to give consideration to the training accuracy and the generalization capability of the verification model, sample data of the high-voltage switch cabinet is immediately divided into a training data set and a test data set, wherein the training data set accounts for 60% -80% and is used for supervised training and learning of the model, and the rest data is used as the test data set for testing the trained classification model;
6) constructing a mixed kernel function:
for multi-dimensional features selected by a high-voltage switch cabinet, a kernel function is introduced into an SVM classification model, original data features of the nonlinear high-voltage switch cabinet are mapped into a high-dimensional linear separable feature space, so that the high-dimensional linear separable feature space is linearly separable, common kernel functions comprise polynomial kernel functions and Gaussian radial basis kernel functions
K1(x,z)＝((xTz)+1)
Figure BDA0003112028330000091
0≤λ≤1
In the formula, K1(x, z) is a polynomial kernel function; k2(x, z) is a radial basis kernel function; lambda is a kernel function proportionality coefficient, sigma is a radial basis kernel function width parameter, and x and z respectively correspond to original data characteristics and characteristics in a mapping space; k (x, z) is a final mixed kernel function expression;
7) substituting the new mixed kernel function obtained in the step 6) into the SVM classification model to obtain a state estimation model of the high-voltage switch cabinet, wherein the state estimation model comprises the following steps:
Figure BDA0003112028330000101
in the formula, yiIs the output of the ith example, αiIs lagrange multiplier, b is bias, m is feature quantity, f (x) represents the result of the final classification of the classifier.
8) The parameter optimization is carried out on the SVM based on an improved fruit fly optimization algorithm (IFOA) algorithm, and the process is as follows:
8.1) initializing IFOA parameters: defining a fruit fly population in a plane coordinate system, initializing the population size s as 100 and maximum iteration number maxgenDetermining the position coordinates of individual drosophila 200, a kernel function proportionality coefficient lambda of a parameter to be optimized, a radial basis kernel function width parameter sigma and a penalty factor C, and randomly initializing the population position to be (lambda)0，σ0，C0) Giving the fruit fly individual random movement direction and distance to generate each fruit fly individual position in the population, defining the number of the fruit flies in the population as l, then the coordinates of the jth fruit fly as,
Figure BDA0003112028330000102
in the formula, R is a random search distance;
8.2) calculating the fitness value of the individual fruit fly optimization: and substituting the position coordinates of the fruit flies into a fitness function F, and evaluating the taste concentration of the individual positions of the fruit flies according to fitness values F (lambda, sigma, C), wherein the fitness values are determined according to the accuracy A of the SVM model. Obtaining the fruit Fly nearest to the food position in the population, namely obtaining the maximum value max F of the fitness corresponding to the fruit Fly individual BI,
Figure BDA0003112028330000111
in the formula, DTFor all samples with correct prediction, D is a total sample, BS is an optimal concentration value, and BI is a drosophila individual;
8.3) preservation of optimal taste concentration values: storing the best fitness value BS executed for the first time in Sb, and simultaneously storing the coordinates of lambda, sigma and C to obtain
Figure BDA0003112028330000112
8.4) determining the average position coordinates of the induced learning fruit flies: sorting the fruit fly taste fitness values searched by the generation, taking two fruit fly individuals subBI1 and subBI2 next to the optimal fruit fly as induced learning individuals, respectively calculating the positions of coordinates of the induced learning individuals, and calculating an arithmetic mean value to obtain lambdae、σe、CeI.e. by
Figure BDA0003112028330000113
8.5) calculating the target position of the convergence direction of the fruit fly colony: updating the position of the fruit fly population by using the formula (8);
Figure BDA0003112028330000121
in the formula, eta is a learning factor, lambdaj′，σj′，Cj' is a new drosophila location coordinate;
in the fruit fly optimizing process, a learning factor is introduced to promote a fruit fly population to learn to the elaiopsis eligua individuals, so that the algorithm is ensured to be finally converged at the global optimal position, in the fruit fly population iterative optimizing process, the direction and the distance of population movement are guided by the elaiopsis eligua individuals, so that the fruit fly population is continuously close to the global optimal position, the influence of the elaiopsis eligua individuals of the previous iteration on the population is gradually reduced along with the increase of the iteration times T, and a fractional attenuation mechanism is adopted to set a learning factor eta and a learning factor attenuation amplitude control parameter cf；
η＝η0/(1+cf(T-1)) (9)
In the formula eta0In order to be the initial learning factor,cfattenuating the amplitude control parameter for the learning factor; setting an initial learning factor η00.2, learning factor decay control parameter cf＝0.4；
8.6) carrying out iterative optimization: the BS of the present generation is given to the preBS and then enters an iterative loop, the steps are repeated, the Sb value of each generation is stored, and when the algorithm reaches maxgenThe loop is ended when the set value is reached, and the position coordinates (lambda, sigma, C) of the best fruit fly individual, namely the optimizing result, are obtained
Sb＝max{BS,preBS} (10)
9) Obtaining optimal parameter combinations according to IFOA algorithm
Figure BDA0003112028330000122
Therefore, an SVM optimal model is determined, test set data are used for verifying the optimal model, and the new state data of the high-voltage switch cabinet are input into the SVM classification model after data preprocessing, so that accurate state evaluation can be performed on the high-voltage switch cabinet.

Claims (1)

1.一种基于IFOA-SVM的高压高压开关柜状态评估方法，其特征在于，所述方法包括以下步骤：1. a high-voltage high-voltage switchgear state assessment method based on IFOA-SVM, is characterized in that, described method comprises the following steps: 1)高压开关柜状态评估数据采集预处理，过程如下：1) Preprocessing of high-voltage switchgear status evaluation data collection, the process is as follows: 通过高压开关柜所配备的在线监测装置、各类传感器进行数据采集，首先，通过使用同类数据的均值和缺失位置数据前后测量值均值对缺失值进行补全，其次，通过对异常情况和正常情况的数据的对比，实现对极端异常值进行界定，从而删除极端异常；Data collection is carried out through the online monitoring devices and various sensors provided in the high-voltage switchgear. First, the missing values are completed by using the mean value of similar data and the mean value of the missing position data before and after the measurement value. The comparison of the data, realize the definition of extreme outliers, so as to delete extreme anomalies; 2)高压开关柜状态评估数据归一化处理，过程如下：2) Normalization processing of high-voltage switchgear status evaluation data, the process is as follows: 采用式(1)对经过异常值和缺失值处理后的高压开关柜数据进行归一化处理；Use formula (1) to normalize the high-voltage switchgear data after processing outliers and missing values;
Figure FDA0003112028320000011
Figure FDA0003112028320000011
 式中，xi为未归一化的高压高压开关柜数据；xmax为同组未归一化的高压开关柜数据中的最大值；xmin未同组未归一化的高压开关柜数据中的最小值；
Figure FDA0003112028320000012
为归一化之后的高压开关柜数据；In the formula, x i is the unnormalized high-voltage switchgear data; x max is the maximum value in the same group of unnormalized high-voltage switchgear data; x min is not the same group of unnormalized high-voltage switchgear data the minimum value of ;
Figure FDA0003112028320000012
is the data of the high-voltage switchgear after normalization; 3)确定模型的输入和输出，过程如下：3) Determine the input and output of the model, the process is as follows: 从在线监测、带电检测和停电实验多信息源中选取高压高压开关柜特征量，选取的特征量为局放值、超声波值、红外诊断、电缆接头温度和母排温度，采用预处理过的高压开关柜样本数据特征作为模型的输入，并将高压开关柜的状态作为输出；The characteristic quantities of high-voltage and high-voltage switchgear are selected from multiple information sources of online monitoring, live detection and power failure experiments. The selected characteristic quantities are partial discharge value, ultrasonic value, infrared diagnosis, cable joint temperature and busbar temperature. The switchgear sample data features are used as the input of the model, and the state of the high-voltage switchgear is used as the output; 4)确定SVM多分类策略，过程如下：4) Determine the SVM multi-classification strategy, the process is as follows: 采用“一对一”策略与SVM结合拓展构成OVO-SVM高压开关柜状态估计分类器，在解决K类高压开关柜状态多分类问题时，在任意两个类别样本间构造一个二分类的SVM子分类器，一共需要构造K(K-1)/2个子分类器，每个子分类器的生成只需要用到训练样本中的两类数据，从而实现K类多分类问题中的两类区分；构造好所有的子分类器后，联合它们对新样本的类别进行决策，并采用“投票法”进行类别的判断，每个子分类器对待分类样本进行“投票”，对于某一个样本只能投1票，当遍历完所有的子分类器后，将样本划分到“得票数”最高的类别中，基于“一对一”分类策略的状态估计分类器在任意两类状态之间构建一个子分类器，将不同的子分类器组合在一起构成最终的分类器；The OVO-SVM high-voltage switchgear state estimation classifier is formed by combining the "one-to-one" strategy with SVM. When solving the multi-classification problem of K-type high-voltage switchgear states, a binary SVM subclass is constructed between any two categories of samples. Classifier, a total of K(K-1)/2 sub-classifiers need to be constructed. The generation of each sub-classifier only needs to use two types of data in the training samples, so as to realize the two types of distinction in K-type multi-classification problems; After all the sub-classifiers are finished, they are combined to make decisions on the category of the new sample, and the "voting method" is used to judge the category. , after traversing all the sub-classifiers, the samples are divided into the categories with the highest "votes", and the state estimation classifier based on the "one-to-one" classification strategy constructs a sub-classifier between any two types of states, Combine different sub-classifiers together to form the final classifier; 5)高压开关柜状态评估样本数据划分，过程如下：5) The sample data division of high-voltage switchgear status evaluation, the process is as follows: 把高压开关柜样本数据随即划分为训练数据集和测试数据集，其中，训练数据集所占比例为60％～80％，用于对模型的有监督训练学习，剩余的数据作为测试数据集，用于对训练后分类模型的测试；The high-voltage switchgear sample data is then divided into training data sets and test data sets. The training data set accounts for 60% to 80%, which is used for supervised training and learning of the model, and the remaining data is used as the test data set. For testing the post-training classification model; 6)构建混合核函数，过程如下：6) Build a mixed kernel function, the process is as follows: 对于高压开关柜所选取的多维特征，在SVM分类模型中引入核函数，将非线性高压开关柜原始数据特征映射到高维的线性可分特征空间中，从而使之线性可分；根据各个核函数的特点，构造混合核函数代替单一核函数，提高状态估计准确率，即For the multi-dimensional features selected by the high-voltage switchgear, a kernel function is introduced into the SVM classification model to map the original data features of the nonlinear high-voltage switchgear into a high-dimensional linearly separable feature space, so as to make it linearly separable; The characteristics of the function, construct a mixed kernel function instead of a single kernel function, and improve the accuracy of state estimation, that is
Figure FDA0003112028320000021
Figure FDA0003112028320000021
 式中，K1(x,z)为多项式核函数；K2(x,z)为径向基核函数；λ为核函数比例系数，σ为径向基核函数宽度参数，x,z分别对应原始数据特征和在映射空间中的特征；K(x,z)为最终的混合核函数表达式；In the formula, K 1 (x, z) is the polynomial kernel function; K 2 (x, z) is the radial basis kernel function; λ is the kernel function scale coefficient, σ is the radial basis kernel function width parameter, x and z are respectively Corresponding to the original data features and the features in the mapping space; K(x,z) is the final mixed kernel function expression; 7)根据步骤6)所得到的新的混合核函数，代入SVM分类模型之中，得到高压开关柜的状态估计模型为：7) According to the new hybrid kernel function obtained in step 6), substitute it into the SVM classification model, and obtain the state estimation model of the high-voltage switchgear as:
Figure FDA0003112028320000031
Figure FDA0003112028320000031
 式中，yi为第i个实例的输出，αi为拉格朗日乘子，b为偏置，m为特征数量，f(x)表示分类器最终分类的结果；In the formula, y i is the output of the ith instance, α i is the Lagrange multiplier, b is the bias, m is the number of features, and f(x) represents the final classification result of the classifier; 8)基于改进果蝇优化算法IFOA算法对SVM进行参数优化，过程如下：8) Based on the improved fruit fly optimization algorithm IFOA algorithm, the parameters of SVM are optimized, and the process is as follows: 8.1)初始化IFOA参数：定义一个果蝇种群位于平面坐标系中，初始化种群规模s，最大迭代次数maxgen,确定果蝇个体位置坐标；待优化参数核函数比例系数λ、径向基核函数宽度参数σ和惩罚因子C，随机初始化种群位置为(λ0，σ0，C0)，并赋予果蝇个体随机的运动方向和距离，从而生成种群中各个果蝇个体位置，定义种群中果蝇的只数为l，则第j只果蝇的坐标为，8.1) Initialize IFOA parameters: define a fruit fly population in a plane coordinate system, initialize the population size s, the maximum number of iterations max gen , and determine the individual position coordinates of the fruit fly; the parameters to be optimized are kernel function proportional coefficient λ, radial basis kernel function width Parameter σ and penalty factor C, randomly initialize the population position as (λ 0 , σ 0 , C 0 ), and assign random movement directions and distances to the individual fruit flies, thereby generating the positions of individual fruit flies in the population, defining the fruit flies in the population The number of is l, then the coordinate of the jth fruit fly is,
Figure FDA0003112028320000032
Figure FDA0003112028320000032
 式中，R为随机搜索距离；where R is the random search distance; 8.2)计算果蝇个体寻优的适应度值：将果蝇位置坐标代入适应度函数F，根据适应度值F(λ,σ,C)评价果蝇个体位置的味道浓度，其中，适应度值根据SVM模型准确率A确定，求得种群中离食物位置最近的果蝇Fly，即求适应度的极大值maxF对应果蝇个体BI,8.2) Calculate the fitness value of the fruit fly individual optimization: Substitute the position coordinates of the fruit fly into the fitness function F, and evaluate the taste concentration of the individual position of the fruit fly according to the fitness value F (λ, σ, C), where the fitness value Determined according to the accuracy rate A of the SVM model, the fruit fly Fly closest to the food position in the population is obtained, that is, the maximum fitness value maxF corresponds to the individual fruit fly BI,
Figure FDA0003112028320000041
Figure FDA0003112028320000041
 式中，DT为所有预测正确的样本，D为总样本，BS为最佳浓度值，BI为果蝇个体；In the formula, D T is all the correctly predicted samples, D is the total sample, BS is the optimal concentration value, and BI is the Drosophila individual; 8.3)保存最佳味道浓度值：将第一次执行的最佳适应度值BS保存于Sb，同时保存λ，σ，C坐标，得8.3) Save the best taste concentration value: save the best fitness value BS of the first execution in Sb, and save the λ, σ, C coordinates at the same time, get
Figure FDA0003112028320000042
Figure FDA0003112028320000042
 8.4)确定诱导学习果蝇的平均位置坐标：对本代搜寻完成的果蝇味道适应度值进行排序，以仅次于最优果蝇的两个果蝇个体subBI1和subBI2作为诱导学习个体，分别计算其坐标的位置并求算术平均值得到λe、σe、Ce，即8.4) Determine the average position coordinates of the induced learning fruit flies: Rank the taste fitness values of the fruit flies that have completed the search in this generation, and use the two fruit flies subBI1 and subBI2, which are second only to the best fruit fly, as the induced learning individuals, and calculate respectively. The position of its coordinates and the arithmetic mean are obtained to obtain λ e , σ e , C e , namely
Figure FDA0003112028320000043
Figure FDA0003112028320000043
 8.5)计算果蝇群体收敛方向的目标位置：利用式(8)更新果蝇种群位置；8.5) Calculate the target position of the convergence direction of the fruit fly population: use the formula (8) to update the position of the fruit fly population;
Figure FDA0003112028320000051
Figure FDA0003112028320000051
 式中，η为学习因子，λj′，σj′，Cj′为新的果蝇位置坐标；In the formula, η is the learning factor, λ j ', σ j ', C j ' are the new fruit fly position coordinates; 在果蝇寻优过程中，引入一种学习因子促使果蝇种群向精英果蝇个体学习，从而保证算法最终收敛于全局最优位置，在果蝇种群迭代寻优的过程中，利用精英个体对种群移动的方向和距离进行引导，使得果蝇种群不断向全局最优位置靠拢，随着迭代次数T的增加，上一次迭代的精英果蝇个体对种群的影响逐渐减小，采取分数衰减机制设置学习因子η和学习因子衰减幅度控制参数cf；In the fruit fly optimization process, a learning factor is introduced to prompt the fruit fly population to learn from the elite fruit fly individuals, thereby ensuring that the algorithm eventually converges to the global optimal position. In the iterative optimization process of the fruit fly population, the elite individuals are used to The direction and distance of the population movement are guided, so that the fruit fly population continues to move closer to the global optimal position. With the increase of the number of iterations T, the influence of the elite fruit fly individuals of the previous iteration on the population gradually decreases, and the score decay mechanism is used to set learning factor η and learning factor attenuation amplitude control parameter cf ; η＝η0/(1+cf(T-1)) (9)η=η 0 /(1+c f (T-1)) (9) 式中，η0为初始学习因子，cf为学习因子衰减幅度控制参数；In the formula, η 0 is the initial learning factor, cf is the learning factor attenuation amplitude control parameter; 8.6)进行迭代寻优：将本代的BS赋予preBS后进入迭代循环，重复上述步骤，保存每代Sb值，当算法达到maxgen的设定值时结束循环，获得最佳果蝇个体的位置坐标(λ，σ，C)即寻优结果，即8.6) Perform iterative optimization: assign the BS of the current generation to preBS and enter the iterative loop, repeat the above steps, save the Sb value of each generation, and end the loop when the algorithm reaches the set value of max gen to obtain the position of the best fruit fly individual The coordinates (λ, σ, C) are the optimization results, namely Sb＝max{BS,preBS} (10)Sb=max{BS,preBS} (10) 9)根据IFOA算法获得最佳参数组合
Figure FDA0003112028320000052
从而确定SVM最优模型，使用测试集数据对该最优模型进行验证，针对高压开关柜新的状态数据，经过数据预处理后输入到该SVM分类模型中，则能对高压开关柜进行准确的状态评估。9) Obtain the best parameter combination according to the IFOA algorithm
Figure FDA0003112028320000052
Therefore, the optimal model of SVM is determined, and the optimal model is verified by the test set data. For the new state data of the high-voltage switchgear, after data preprocessing, it is input into the SVM classification model, and the high-voltage switchgear can be accurately analyzed. State Assessment.
CN202110654396.1A 2021-06-11 2021-06-11 IFOA-SVM-based high-voltage switch cabinet state evaluation method Pending CN113379251A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110654396.1A CN113379251A (en) 2021-06-11 2021-06-11 IFOA-SVM-based high-voltage switch cabinet state evaluation method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110654396.1A CN113379251A (en) 2021-06-11 2021-06-11 IFOA-SVM-based high-voltage switch cabinet state evaluation method

Publications (1)

Publication Number Publication Date
CN113379251A true CN113379251A (en) 2021-09-10

Family

ID=77574023

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110654396.1A Pending CN113379251A (en) 2021-06-11 2021-06-11 IFOA-SVM-based high-voltage switch cabinet state evaluation method

Country Status (1)

Country Link
CN (1) CN113379251A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114548188A (en) * 2022-04-21 2022-05-27 武汉格蓝若智能技术有限公司 Method for identifying operation mode of capacitive voltage transformer in transformer substation
CN118862991B (en) * 2024-09-27 2025-01-03 浙江伟臻成套柜体有限公司 A lightweight distribution cabinet service life loss prediction optimization method

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107330781A (en) * 2017-06-19 2017-11-07 南京信息工程大学 A kind of individual credit risk appraisal procedure based on IFOA SVM
CN111783866A (en) * 2020-06-28 2020-10-16 中国电子科技集团公司第二十研究所 Production logistics early warning information multi-classification method based on improved FOA-SVM

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107330781A (en) * 2017-06-19 2017-11-07 南京信息工程大学 A kind of individual credit risk appraisal procedure based on IFOA SVM
CN111783866A (en) * 2020-06-28 2020-10-16 中国电子科技集团公司第二十研究所 Production logistics early warning information multi-classification method based on improved FOA-SVM

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
亢超群: ""基于多源信息融合的配电开关柜状态评价方法研究"", 《中国优秀博硕士学位论文全文数据库（硕士） 工程科技II辑》, 15 February 2017 (2017-02-15), pages 8 - 9 *
张维 等: ""基于IFOA-SVM的轴承故障分类识别方法"", 《机械传动》, 28 February 2021 (2021-02-28), pages 3 - 5 *
葛晓霞 等: ""基于改进果蝇算法优化 SVM 的凝汽器真空预测"", 《热能动力工程》, 30 November 2020 (2020-11-30), pages 4 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114548188A (en) * 2022-04-21 2022-05-27 武汉格蓝若智能技术有限公司 Method for identifying operation mode of capacitive voltage transformer in transformer substation
CN118862991B (en) * 2024-09-27 2025-01-03 浙江伟臻成套柜体有限公司 A lightweight distribution cabinet service life loss prediction optimization method

Similar Documents

Publication Publication Date Title
Zhang et al. Load probability density forecasting by transforming and combining quantile forecasts
CN109829587A (en) Zonule grade ultra-short term and method for visualizing based on depth LSTM network
CN112581263A (en) Credit evaluation method for optimizing generalized regression neural network based on wolf algorithm
CN114399032B (en) Method and system for predicting metering error of electric energy meter
CN111931983B (en) Precipitation prediction method and system
CN111310722A (en) Power equipment image fault identification method based on improved neural network
CN113379251A (en) IFOA-SVM-based high-voltage switch cabinet state evaluation method
CN117494573B (en) Wind speed prediction method, system and electronic equipment
CN115454988B (en) Satellite power supply system missing data complement method based on random forest network
CN118336705A (en) Short-term load interval prediction method
CN113033898A (en) Electrical load prediction method and system based on K-means clustering and BI-LSTM neural network
CN109901064B (en) ICA-LVQ-based high-voltage circuit breaker fault diagnosis method
Zhang et al. Ensemble optimization approach based on hybrid mode decomposition and intelligent technology for wind power prediction system
CN113536662B (en) Error state prediction method of electronic transformer based on firefly optimization LightGBM algorithm
CN115660167A (en) Short-term air quality prediction method based on sparrow search algorithm and decomposition error correction
Fan et al. Wind power time series missing data imputation based on generative adversarial network
US20200349495A1 (en) Analytical model training method for customer experience estimation
Tan et al. Time Series Prediction using Backpropagation Network Optimized by Hybrid K-means-Greedy Algorithm.
CN111160419A (en) Electronic transformer data classification prediction method and device based on deep learning
CN118569445B (en) Intelligent dynamic calibration multidimensional air quality monitoring data optimization system
Kashirina et al. Development of information and analytical tools based on adaptive classifier Cascade ARTMAP
Wei et al. TST: An Effective Approach to Extract Trend Feature in Stock Time Series
Wang et al. E-commerce growth prediction model based on grey Markov chain
Zhao et al. Research on the Prediction Application of Multiple Classification Datasets Based on Random Forest Model
CN118817547A (en) A PM2.5 prediction method based on TCN neural network

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
RJ01 Rejection of invention patent application after publication
RJ01 Rejection of invention patent application after publication

Application publication date: 20210910