CN115081811A - Relay protection system risk assessment method and system based on semi-supervised MD algorithm - Google Patents

Relay protection system risk assessment method and system based on semi-supervised MD algorithm Download PDF

Info

Publication number
CN115081811A
CN115081811A CN202210550103.XA CN202210550103A CN115081811A CN 115081811 A CN115081811 A CN 115081811A CN 202210550103 A CN202210550103 A CN 202210550103A CN 115081811 A CN115081811 A CN 115081811A
Authority
CN
China
Prior art keywords
algorithm
evaluation
relay protection
matrix
protection system
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
CN202210550103.XA
Other languages
Chinese (zh)
Inventor
叶远波
章昊
李端超
王同文
汪胜和
汪伟
程晓平
王栋
王薇
项忠华
陈晓东
刘宏君
赵子根
丛雷
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.)
State Grid Anhui Electric Power Co Ltd
CYG Sunri Co Ltd
Original Assignee
State Grid Anhui Electric Power Co Ltd
CYG Sunri 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 State Grid Anhui Electric Power Co Ltd, CYG Sunri Co Ltd filed Critical State Grid Anhui Electric Power Co Ltd
Priority to CN202210550103.XA priority Critical patent/CN115081811A/en
Publication of CN115081811A publication Critical patent/CN115081811A/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
    • G06Q10/0635—Risk analysis of enterprise or organisation activities
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • 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
    • G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
    • 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
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Landscapes

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

Abstract

A relay protection system risk assessment method and system based on a semi-supervised MD algorithm belong to the technical field of relay protection of power systems and solve the problems of poor adaptability and self-correction capability and strong subjectivity of the existing assessment method; the method comprises the steps of establishing an operation state evaluation model of a relay protection system based on an analytic hierarchy process as a training set of a semi-supervised MD algorithm, carrying out weight calculation on the operation state evaluation model, carrying out ridge function fuzzy processing on the operation state of a relay protection device, carrying out machine algorithm learning on the established training set to eliminate subjectivity defects after fuzzy processing, and comparing with other machine learning algorithms.

Description

Relay protection system risk assessment method and system based on semi-supervised MD algorithm
Technical Field
The invention belongs to the technical field of relay protection of power systems, and relates to a relay protection system risk assessment method and system based on a semi-supervised (MD) algorithm.
Background
In recent years, more and more intelligent power electronic devices and new energy devices are connected to a power system, and the improvement of the maintenance efficiency of the power system equipment has important significance for the stable operation of the power system. The relay protection device is correct, rapid and reliable in action, so that the fault range is effectively prevented from being enlarged after the power system is in fault, the stability of the power system is effectively improved, and the economic loss is reduced.
With the development of big data technology, machine learning is widely applied to power systems, for example, an improved classification method for genetic algorithm optimization support vector machine parameters is proposed in the document "evaluation method for relay protection device state based on GA optimization SVM parameters and cloud model" (chen haitao et al, intelligent power 2020,48(07):88-92+117.), and uncertainty mapping between health degree and comment domain is realized by using randomness and stable tendency of the cloud model, so that an evaluation result is more in line with actual conditions. However, the existing methods have not negligible limitations: 1) advanced algorithms are required to improve the accuracy of pattern recognition; the adaptability and self-correction capability of the evaluation method are poor; 2) the scheme has strong subjectivity and is not supported by too much data; 3) the simulation results of some models may not be as helpful to actual operation and maintenance.
In order to solve the defects, the invention researches a relay protection system risk assessment method based on a semi-supervised machine learning (MD) (Mahalanobis distance) algorithm.
Disclosure of Invention
The invention aims to design a relay protection system risk assessment method and system based on a semi-supervised (MD) algorithm, so as to solve the problems of poor adaptability and self-correction capability and strong subjectivity of the conventional assessment method.
The invention solves the technical problems through the following technical scheme:
the relay protection system risk assessment method based on the semi-supervised MD algorithm comprises the following steps:
s1, selecting static and real-time parameters of the M-C device, the merging unit, the protection device and the intelligent terminal as evaluation indexes, carrying out weight calculation on the evaluation indexes, and establishing a state evaluation model based on an AHP fuzzy comprehensive algorithm; the method for calculating the weight of the evaluation index specifically comprises the following steps: comparing every two of the nine-level judgment tables to obtain an importance comparison matrix, and processing the importance comparison matrix to obtain a judgment matrix; carrying out consistency check on the judgment matrix by using the safety random index, obtaining the judgment matrixes of the M-C device, the merging unit, the protection device and the intelligent terminal by transforming the matrixes, and calculating the evaluation index weights of the M-C device, the merging unit, the protection device and the intelligent terminal;
s2, establishing a relay protection system risk assessment model based on a semi-supervised MD algorithm to overcome the subjectivity defect of a fuzzy comprehensive algorithm.
The technical scheme of the invention is that an operation state evaluation model of a relay protection system based on an analytic hierarchy process is established to be used as a training set of a semi-supervised MD algorithm, weight calculation is carried out on the training set, ridge function fuzzy processing is carried out on the operation state of a relay protection device, machine algorithm learning is carried out on the established training set to eliminate subjectivity defects after fuzzy processing, compared with other machine learning algorithms, the semi-supervised MD algorithm based on the analytic hierarchy process is provided to describe the relevance between variables without being influenced by dimensionality, great advantages are provided in the aspect of distance measurement learning processing among multivariable high-dimensional samples, the requirement on the scale of the training set is lowest, the prediction result is good, and the result is accurate.
Further, the formula of the judgment matrix is as follows:
Figure BDA0003654622030000021
the maximum eigenvalue of the judgment matrix is as follows:
Figure BDA0003654622030000022
the characteristic vector of the judgment matrix is as follows:
W＝[ω’ 1 ,ω’ 2 ,...ω’ n ]
wherein the comparison coefficient is defined as a ij ＝x i /x j ，i＝1,2,3,...,m；j＝1,2,3,...,m。
Further, the method for performing consistency check on the judgment matrix by using the security random index comprises the following steps: consistency is defined as: the method comprises the following steps that CR is CI/RI, RI is a random index, CI is a consistency check standard, CI is (lambda-N)/(N-1), lambda is the maximum characteristic value of a judgment matrix, and N is a nonzero characteristic root of an N-order consistency matrix;
further, the method for establishing the state evaluation model based on the AHP fuzzy synthesis algorithm specifically comprises the following steps:
s11, considering fuzzy relations among all evaluation index variables, when the running state of the relay protection system is evaluated, the fuzzy relation of a continuous interval [0,1] is adopted to represent a fuzzy characteristic domain of the relay protection system, and the running state of the relay protection system comprises the following steps: normal, warning, error and hidden danger; the fuzzy characteristic domain of the relay protection system comprises: index set, weight set and evaluation set;
s12, carrying out standardization processing on the data of the fuzzy characteristic domain of the relay protection system, converting the actual value into an interval [0,1], and dividing the state evaluation index into: benefit type, cost type, interval type;
s13, calculating membership degree matrixes of each state evaluation index by using ridge distribution;
and S14, setting an index set, a weight set and an evaluation set according to the statistical data, taking the index set, the weight set and the evaluation set as a training set, and calculating the evaluation result of the protection device of the training set.
Further, the method for establishing the relay protection system risk assessment model based on the semi-supervised MD algorithm to overcome the subjectivity defect of the fuzzy comprehensive algorithm in step S2 is as follows:
the cost function of the semi-supervised MD algorithm is as follows:
Figure BDA0003654622030000031
calculated to satisfy the following formula
Figure BDA0003654622030000034
The value of (c):
Figure BDA0003654622030000032
the G index of the overall sample is m-24, and the mean vector of the high-dimensional data sample is mu (mu) 1 ,μ 2 ,…,μ 24 ) ', global sample G and individual sample X ═ X (X) 1 ,x 2 ,…,x 24 ) The MD between' is:
d 2 (X,G)＝(X-μ)’∑(X-μ) -1
the covariance between the evaluation indices is:
Figure BDA0003654622030000033
the solution is formulated as:
Figure BDA0003654622030000041
in the formula, N 0 Is the number of marked samples;
Figure BDA0003654622030000042
is the number of implicit layer nodes;
Figure BDA0003654622030000043
is a sample vector of markers; n is the dimension of the sample vector;
Figure BDA0003654622030000044
is that
Figure BDA0003654622030000045
Class label sample vector of (1); c is the number of output nodes of the network; w is a i ＝[w i1 ,…,w in ] T Is an input weight vector between the input node and the implicit layer node; b i Is the offset of node i; g is an activation function of an implicit layer node; beta is a i ＝[β i1 ,β i2 ,…,β iC ] T An output weight vector between the hidden layer node and the output node; n is a radical of e Is the number of extended samples;
Figure BDA0003654622030000046
is an extended sample vector;
Figure BDA0003654622030000047
is that
Figure BDA0003654622030000048
Class label sample vector of (1); h 0 Outputting a matrix for the hidden layer of the marked sample set; β is the output matrix of weights; t is 0 Labeling a matrix for the category of the labeled sample; h e Outputting a matrix for a hidden layer of the extended sample; t is e Is a class label matrix of extended samples.
A relay protection system risk assessment system based on a semi-supervised MD algorithm comprises: an operation state evaluation module and a risk evaluation module; the running state evaluation module is used for selecting static and real-time parameters of the M-C device, the merging unit, the protection device and the intelligent terminal as evaluation indexes, carrying out weight calculation on the evaluation indexes and establishing a state evaluation model based on an AHP fuzzy comprehensive algorithm; the method for calculating the weight of the evaluation index specifically comprises the following steps: comparing every two of the nine-level judgment tables to obtain an importance comparison matrix, and processing the importance comparison matrix to obtain a judgment matrix; carrying out consistency check on the judgment matrix by using the safety random index, obtaining the judgment matrixes of the M-C device, the merging unit, the protection device and the intelligent terminal by transforming the matrixes, and calculating the evaluation index weights of the M-C device, the merging unit, the protection device and the intelligent terminal; the risk evaluation module is used for establishing a relay protection system risk evaluation model based on a semi-supervised MD algorithm so as to overcome the subjectivity defect of a fuzzy comprehensive algorithm.
Further, the formula of the judgment matrix is as follows:
Figure BDA0003654622030000049
the maximum eigenvalue of the judgment matrix is as follows:
Figure BDA0003654622030000051
the characteristic vector of the judgment matrix is as follows:
W＝[ω’ 1 ,ω’ 2 ,…ω’ n ]
wherein the comparison coefficient is defined as a ij ＝x i /x j ，i＝1,2,3,...,m；j＝1,2,3,...,m；
Further, the method for performing consistency check on the judgment matrix by using the security random index comprises the following steps: consistency is defined as: and CR is CI/RI, RI is a random index, CI is a consistency check standard, CI is (lambda-N)/(N-1), lambda is the maximum characteristic value of the judgment matrix, and N is a nonzero characteristic root of the N-order consistency matrix.
Further, the method for establishing the state evaluation model based on the AHP fuzzy synthesis algorithm specifically comprises the following steps:
(1) considering fuzzy relations existing among all evaluation index variables, when the running state of the relay protection system is evaluated, the fuzzy relation of a continuous interval [0,1] is adopted to represent a fuzzy characteristic domain of the relay protection system, and the running state of the relay protection system comprises the following steps: normal, warning, error and hidden danger; the fuzzy characteristic domain of the relay protection system comprises: index set, weight set and evaluation set;
(2) carrying out standardization processing on data of a fuzzy characteristic domain of the relay protection system, converting an actual value into an interval [0,1], and dividing state evaluation indexes into the following parts according to an exchange mode: benefit type, cost type, interval type;
(3) calculating a membership matrix of each state evaluation index by using ridge distribution;
(4) and setting an index set, a weight set and an evaluation set according to the statistical data, taking the index set, the weight set and the evaluation set as a training set, and calculating the evaluation result of the protection device of the training set.
Further, the method for establishing the relay protection system risk assessment model based on the semi-supervised MD algorithm to overcome the subjectivity defect of the fuzzy comprehensive algorithm comprises the following steps:
the cost function of the semi-supervised MD algorithm is as follows:
Figure BDA0003654622030000052
calculated to satisfy the following formula
Figure BDA0003654622030000053
The value of (c):
Figure BDA0003654622030000061
the G index of the overall sample is m-24, and the mean vector of the high-dimensional data sample is mu (mu) 1 ,μ 2 ,…,μ 24 ) ', global sample G and individual sample X ═ X (X) 1 ,x 2 ,…,x 24 ) The MD between' is:
d 2 (X,G)＝(X-μ)’∑(X-μ) -1
the covariance between the evaluation indices is:
Figure BDA0003654622030000062
the solution is formulated as:
Figure BDA0003654622030000063
in the formula, N 0 Is the number of marked samples;
Figure BDA0003654622030000064
is the number of implicit layer nodes;
Figure BDA0003654622030000065
is a sample vector of markers; n is the dimension of the sample vector;
Figure BDA0003654622030000066
is that
Figure BDA0003654622030000067
Class label sample vector of (1); c is the number of output nodes of the network; w is a i ＝[w i1 ,…,w in ] T Is an input weight vector between the input node and the implicit layer node; b i Is the offset of node i; g is an activation function of an implicit layer node; beta is a i ＝[β i1 ,β i2 ,…,β iC ] T An output weight vector between the hidden layer node and the output node; n is a radical of e Is the number of extended samples;
Figure BDA0003654622030000068
is an extended sample vector;
Figure BDA0003654622030000069
is that
Figure BDA00036546220300000610
Class label sample vector of (1); h 0 Outputting a matrix for the hidden layer of the marked sample set; β is the output matrix of weights; t is 0 Labeling a matrix for the category of the labeled sample; h e Outputting a matrix for a hidden layer of the extended sample; t is e Is a class label matrix of extended samples.
The invention has the advantages that:
the technical scheme of the invention is that an operation state evaluation model of a relay protection system based on an analytic hierarchy process is established to be used as a training set of a semi-supervised MD algorithm, weight calculation is carried out on the training set, ridge function fuzzy processing is carried out on the operation state of a relay protection device, machine algorithm learning is carried out on the established training set to eliminate subjectivity defects after fuzzy processing, compared with other machine learning algorithms, the semi-supervised MD algorithm based on the analytic hierarchy process is provided to describe the relevance between variables without being influenced by dimensionality, great advantages are provided in the aspect of distance measurement learning processing among multivariable high-dimensional samples, the requirement on the scale of the training set is lowest, the prediction result is good, and the result is accurate.
Drawings
Fig. 1 is a network topology diagram of a relay protection system risk assessment method based on a semi-supervised MD algorithm in an embodiment of the present invention;
FIG. 2 is a triangular distribution membership function diagram of the relay protection system risk assessment method based on the semi-supervised MD algorithm in the embodiment of the present invention;
FIG. 3 is a diagram of a ridge distribution membership function of a semi-supervised MD algorithm-based risk assessment method for a relay protection system in the embodiment of the present invention;
fig. 4 is an evaluation diagram of the protection operation state of a local 110kV loop I transformer of the risk evaluation method of the relay protection system based on the semi-supervised MD algorithm of the embodiment of the present invention, wherein (a) is an index diagram without considering the weight, and (b) is an index diagram after considering the weight;
FIG. 5 is a residual error diagram of the relay protection system risk assessment method based on the semi-supervised MD algorithm in the embodiment of the present invention;
fig. 6 is a corrected residual error diagram of the relay protection system risk assessment method based on the semi-supervised MD algorithm in the embodiment of the present invention;
FIG. 7 is a ROC evaluation curve diagram of the relay protection system risk evaluation method based on the semi-supervised MD algorithm in the embodiment of the present invention;
fig. 8 is a machine learning algorithm comparative analysis diagram of the relay protection system risk assessment method based on the semi-supervised MD algorithm in the embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the embodiments of the present invention, and it is obvious that the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The technical scheme of the invention is further described by combining the drawings and specific embodiments in the specification:
example one
1. Establishing an Analytic Hierarchy Process (AHP) -based relay protection system operation state evaluation model
1.1 evaluation index selection
According to the DL/T860 engineering implementation standard, the network topology of the intelligent substation of the relay protection system is divided into three layers, namely a station control layer, a bay layer and a process layer, as shown in FIG. 1.
A station control layer: the station control layer comprises a background monitor, a remote control machine, a clock system, a communication power supply system and the like, and mainly realizes the monitoring function of the running state, the alarm information, the CPU, the memory utilization rate, the hard disk utilization rate, the network interface and the like.
Spacing layer: the bay level equipment comprises a protection device, a measurement and control (M-C) device, a fault recorder, a network analyzer and the like. Its main functions are monitoring a/D, Generic Object Oriented Substation Events (GOOSE), Sampling Values (SV) and I/O.
A process layer: the intelligent power amplifier consists of a merging unit, an intelligent terminal, a mutual inductor and the like, and is mainly used for collecting information such as clock matching, chip state, optical power and the like.
After extensive research, referring to 'secondary equipment state evaluation and risk evaluation guide rules' and 'intelligent substation relay protection information standard' of southern power grid company, static and real-time parameters of an M-C device, a merging unit, a protection device and an intelligent terminal which can reflect the operation state of a relay protection system most are selected as modeling objects.
The M-C device has important significance for a dispatcher to master and control the transformer substation information, and the evaluation indexes of the M-C device comprise 28 indexes: ROM fault, RAM fault, output fault, input fault, parameter setting fault, protection pressing plate fault, EEPROM fault, serial port communication interruption, system configuration error, CPU communication interruption, configuration table error, logic table error, inconsistent indication, indicator lamp fault, system operation fault, circuit breaker no-fault tripping, communication abnormality, optical power abnormality, crystal oscillator frequency deviation, GOOSE alarm information (abnormal), GOOSE alarm information (serious), GOOSE alarm information (wrong), SV alarm information (abnormal), SV alarm information (serious), SV alarm information (wrong), multimedia message alarm information (abnormal), multimedia message alarm information (serious), multimedia message alarm information (wrong).
The merging unit can realize the acquisition and transmission of voltage and current data according to time sequence. Contains 18 evaluation indexes: ROM fault, RAM fault, output fault, input fault, parameter setting fault, protection pressing plate fault, EEPROM fault, serial port communication interruption, system configuration error, CPU communication interruption, configuration table error, logic table error, indication inconsistency, indicator lamp fault, system operation fault, circuit breaker no-fault tripping, communication abnormity and optical power abnormity.
The comprehensive protection device for the high, medium and low voltage lines has the functions of removing fault equipment and preventing expanded faults, and comprises 24 evaluation indexes: ROM fault, RAM fault, output fault, input fault, parameter setting fault, protection pressing plate fault, EEPROM fault, serial port communication interruption, system configuration error, CPU communication interruption, configuration table error, logic table error, indication inconsistency, indicator lamp fault, system operation fault, circuit breaker no-fault tripping, communication abnormity, optical power abnormity, crystal oscillator frequency deviation, GOOSE alarm information (abnormity), GOOSE alarm information (serious), GOOSE alarm information (error), SV alarm information (abnormity), SV alarm information (serious), and SV alarm information (error).
The intelligent terminal is responsible for circuit breaker instruction and equipment location, contains 23 evaluation indexes: ROM fault, RAM fault, output fault, input fault, parameter setting fault, protection pressing plate fault, EEPROM fault, serial port communication interruption, system configuration error, CPU communication interruption, configuration table error, logic table error, indication inconsistency, indicator lamp fault, system operation fault, circuit breaker no-fault tripping, communication abnormity, optical power abnormity, crystal oscillator frequency deviation, GOOSE alarm information (abnormity), GOOSE alarm information (serious), GOOSE alarm information (error), SV alarm information (abnormity), SV alarm information (serious).
1.2, calculating index weight value
1.2.1 evaluation index of relay protection system
The operation assessment indexes of the relay protection system consist of dynamic assessment indexes and static assessment indexes. The dynamic assessment indexes of the intelligent relay protection device are derived from self-checking information and mainly comprise a memory, a setting value, signal input and output, a relay protection board, a communication interface, a CPU and the like. The static assessment indexes are derived from historical operating records and mainly comprise alarm information records of a GOOSE web, an SV web and an MMS web. The alarm information is divided into normal, alarm, fault and critical information.
1.2.2, decision matrix
The relevant importance of each index is calculated by using a nine-level judgment table as shown in Table 1, wherein "1" represents x i And x j Equally important; "3" means x i Is slightly more important than x j (ii) a "5" means x i Ratio x j Importance; "7" means x i Ratio x j More important; "9" means x i Far exceeds x j 。
TABLE 1 nine-grade judging table
Figure BDA0003654622030000091
In order to obtain the influence degree of the index X containing m indexes on the index Y, an importance comparison matrix is established through pairwise comparison.
Defining the comparison coefficient as a ij ＝x i /x j Through which is passed
Figure BDA0003654622030000101
And performing secondary comparison processing to obtain a judgment matrix A:
Figure BDA0003654622030000102
solving the maximum eigenvalue lambda of the judgment matrix A and the corresponding eigenvector W according to the formulas (2) to (6):
Figure BDA0003654622030000103
M i the square root of degree n is:
Figure BDA0003654622030000104
will be provided with
Figure BDA0003654622030000105
Normalizing to obtain:
Figure BDA0003654622030000106
the maximum eigenvalue may be expressed as:
Figure BDA0003654622030000107
the feature vector is represented as:
W＝[ω’ 1 ,ω’ 2 ,...ω’ n ] (6)
1.2.3 weight calculation and consistency check
The weight calculation result is considered correct only if the decision matrix meets the consistency check criterion. The consistency check criteria were customized as:
CI＝(λ-N)/(N-1) (7)
wherein, λ is the maximum eigenvalue of the n-order reciprocal matrix A; n is the non-zero characteristic root of the uniform matrix of order N.
Consistency (CR) is defined as:
CR＝CI/RI (8)
wherein, RI is a random index, and the random index of security is shown in Table 2.
Table 2 random index table
N RI N RI N RI N RI
1 0 8 1.41 15 1.59 22 1.64
2 0 9 1.46 16 1.60 23 1.65
3 0.58 10 1.49 17 1.61 24 1.65
4 0.90 11 1.52 18 1.61 25 1.66
5 1.12 12 1.54 19 1.62 26 1.66
6 1.24 13 1.56 20 1.63 27 1.66
7 1.32 14 1.58 21 1.64 28 1.67
If CR < 0.10, then the consistency check of the decision matrix is deemed acceptable. The CR calculation results for the devices in the station are as follows: CR m-c ＝0.0069、CR m ＝0.0123、CR p ＝0.0118、CR i And (4) all meet the consistency check standard of 0.0122.
The decision matrices for different devices can be derived by transforming the matrices:
Figure BDA0003654622030000111
Figure BDA0003654622030000112
Figure BDA0003654622030000113
Figure BDA0003654622030000121
the calculation of index weights for different devices using equations (2) - (6) is shown in table 3.
TABLE 3 index weight calculation for different devices
Figure BDA0003654622030000122
Figure BDA0003654622030000131
1.3 fuzzy comprehensive evaluation
Considering the fuzzy relation existing among all index variables, when the relay protection running state is evaluated, the actual running cannot be simply described by true and false binary logic of 0-1, but the actual running can be represented by the fuzzy relation of a continuous interval [0,1 ]. The relay protection fuzzy characteristic domain comprises an index set, an index weight set and an evaluation set. For data normalization, the actual value is converted into an interval [0,1], and the state evaluation index is divided into: benefit type, cost type, and interval type. These three types correspond to three evaluation methods, respectively: 1. the higher the index value, the better the benefit. 2. Cost type: the smaller the value, the better. 3. The interval type: the values should be within the ideal interval.
In a relay protection system, the operation state is generally divided into four levels, namely journey, warning, error and critical state. Comparing the trigonometric function distribution with the ridge function distribution, we find that the ridge distribution has higher membership information when the index is at a high membership position. At the same time, at lower membership, the ridge distribution eliminates less membership information. Triangularly distributed fuzzy membership tends to be a conservative result, so membership matrices for each state are computed for the gate-select ridge distribution.
Finally, 80 devices in the statistical data of the southern power grid company in 2017 and 2020 are taken as examples for analysis, and an evaluation index set with the index number of 24 and the evaluation grade type of 4 is set; the method comprises the following steps of (1) classifying into four grades of evaluation sets of normal, alarm, fault and hidden danger; and a weight set of 80 devices. And taking the result as a training set, and calculating the evaluation result of the protection device of the training set.
Considering the fuzzy relationship between variables, the actual operation cannot be described simply by 0-1 logic when evaluating the operation state of the relay protection system. Fuzzy relations should be represented by a continuous interval [0,1] rather than by binary logic of 0 (false) and 1 (true).
1.3.1 fuzzy comprehensive evaluation characteristic domain
The fuzzy characteristic domain of the relay protection system comprises an index set, a weight set and an evaluation set, wherein I is ═ I 1 ,i 2 ,…,i n Indicating an index set of state evaluation of the relay protection system; calculating according to analytic hierarchy process to obtain index weight value represented as W ═ W 1 ,w 2 ,…,w n }. Hypothesis evaluation set m The results of the evaluation are combined, and may then be expressed as E ═ E 1 ,e 2 ,…,e n }。
1.3.2 data normalization
For data standardization, the actual value is converted into an interval [0,1], and the state evaluation indexes are classified into benefit type, cost type and interval type according to the exchange mode.
Benefit type: benefit type means the greater the value the better. Its transfer function g b (x) Can be expressed as
Figure BDA0003654622030000141
Wherein x is the actual running value, [ x ] min ,x max ]Are the corresponding lower and upper limits.
Cost type: the cost type is that a smaller value is better. Cost type g c (x) Has a conversion function of
Figure BDA0003654622030000142
Wherein x is the actual running value, [ x ] min ,x max ]Are the corresponding lower and upper limits.
The interval type: the interval type indicates that the value should be within the ideal interval. Interval type g i (x) Has a conversion function of
Figure BDA0003654622030000143
Where x is the actual running value, [ x ] i_min ,x i_max ]Is the ideal running interval, [ x ] min ,x max ]Is the desired operating interval.
1.3.3 fuzzy comprehensive membership function
Trigonometric distribution and ridge distribution membership functions have found wide application in engineering practice. The running state of the relay protection system is divided into four grades of normal, warning, error and hidden danger. The trigonometric distribution membership functions for the different states are:
Figure BDA0003654622030000144
Figure BDA0003654622030000145
Figure BDA0003654622030000151
Figure BDA0003654622030000152
combining equations (16) - (19), the calculation model of the trigonometric distribution membership function is shown in fig. 2.
The ridge distribution membership functions for the different states are:
Figure BDA0003654622030000153
Figure BDA0003654622030000154
Figure BDA0003654622030000155
Figure BDA0003654622030000156
in conjunction with equations (20) - (23), the model for the computation of the ridge distribution membership function is shown in FIG. 3.
As can be seen from fig. 2 and 3, the information of the degree of membership of the ridge distribution is higher when the index is at a high degree of membership position, as compared to the degree of membership function of the triangular distribution. Meanwhile, at lower membership positions, the ridge distribution eliminates less membership information. Triangularly distributed fuzzy membership tends to yield conservative results. In view of the above analysis, the present invention employs a ridge distribution to compute a membership matrix for each state.
1.3.4 obtaining training set
The data information used by the invention is taken from the statistics data of the southern power grid company 2017 plus 2020, and comprises the following steps: mobile communication equipment, merger group, protection device and intelligent terminal. Taking 80 devices such as 110kV Qingshan transformer substation and 110kV purple light transformer substation in Guizhou province as an example for analysis.
The number of indices for evaluating the protection device is n p 24; evaluation class type m p 4; index setting I p Device failure, ROM failure, parameter setting failure, … }; evaluation set is E p 1, getting { normal, alarm, fault, hidden danger }; set of weights W p ＝{0.1049,0.1049,0.1049,0.0057,…,0.1049}。
The data standardization processing standard of the protection device is shown in table 4. The index database shown in table 4 is from the station level, bay level and process level: remote control personnel sense running states and warnings, and data interaction between station levels is achieved through DL/860; the M-C equipment and the protection equipment collect data from the MMS network, and the switcher acquires the data through DL/T860; and the self-monitoring information of the intelligent terminal and the merging unit acquired by the equipment chip is sent to the M-C equipment through the GOOSE. And the data acquired by the MC equipment is uploaded to the station-level online state monitoring module through multimedia messages. The network analyzer acquires and analyzes GOOSE and SV information at the process level.
Table 4 protective device data standardization standard
Figure BDA0003654622030000161
Figure BDA0003654622030000171
Taking a certain device in a station of 6 months and 1 day in 2020 as an example, the evaluation calculation of the protection operation state of the 110kV loop I of the Qingshan transformer is shown in FIG. 4. Fig. 4a shows the correspondence relationship of each index of the protection device, and fig. 4b shows the relationship in consideration of the weight. The broken line indicates the upper limit of the ideal upper limit of the evaluation index, and the solid line indicates the actual evaluation result.
Taking a certain device at 6.1.2020 as an example, the evaluation of the 110kV I loop protection operating state of the qingshan substation is obtained by calculation and is shown in fig. 4 (the dotted line represents the upper limit of the ideal upper limit of the evaluation index, and the solid line represents the actual evaluation result).
The training set provided for the machine learning algorithm consisted of 80 protector evaluations, as shown in table 5.
And S is the scale of the machine learning training set and represents the proportion of the whole sample. The training set contains samples of the sequence [1,8000s ]. Table 5 shows the test set consisting of samples with the sequence [8000s +1,8000 ].
TABLE 5 evaluation results of protection devices
Evaluating time Index 1 … Index 24 Score of
2020.06.01 0 … 4 94.99
2020.06.13 0 … 10 90.93
2020.07.02 0 … 5 88.36
… … … … …
2020.07.06 0 … 20 94.29
2. Machine learning algorithm for running state of relay protection system
Adopting an AHP fuzzy comprehensive evaluation method to evaluate four devices, wherein the number d of each device i 80,10,24, 33. Evaluation period of c i ，p i Represents the proportion of the total sample. d i c i p i Representing the number of training sets of an individual device, d i c i (1-p i ) Refers to the number of one test set per device. In order to overcome the subjectivity defect of the fuzzy comprehensive algorithm, the invention adopts a new machine learning method, so that the calculation result is more objective and real.
2.1 supervised multiple regression analysis Algorithm
Aiming at different types of equipment, the multiple regression equations of the protection device, the M-C device, the merging device and the intelligent terminal are as follows:
Y i ＝f(X i ,B i ,ε i )(i＝1,2,3,4) (24)
wherein Y is i Is the result of the evaluation of the different devices; x i Is an evaluation index; b i Is a regression coefficient; epsilon i Is a random error.
In order to minimize the sum of squares in the historical evaluation period, an optimization objective function is proposed as follows:
Figure BDA0003654622030000181
wherein, Y 1j Is the observed protector value;
Figure BDA0003654622030000182
is an estimate of the protection device.
Considering the extreme principle, the constraint function is:
Figure BDA0003654622030000183
where j is the number of the training set.
Taking the statistical data of 100 evaluation periods of 80 protection devices as a training set, wherein the training set is in proportion p 1 A residual level map can be obtained using a supervised multivariate regression analysis algorithm on the training set, 80%, as shown in fig. 5.
When the significance level α is 0.05, there are 308 abnormal sample points in the residual map, and as shown in fig. 5, the yield is 95.19%.
After removing the red outliers in the graph, the corrected residual graph after data clean-up is shown in fig. 6. The number of abnormal value samples in the regression model after correction is 274, and the yield is improved to 95.50%.
2.2 ED-based unsupervised k-means algorithm
Taking the protection device as an example, an ED (empirical distance) -based unsupervised k-means algorithm is adopted to process unmarked numbersAccording to the evaluation index of X 1 ＝{x 1 ,x 2 ,…,x 24 }。
The ED between samples a and b in the training set is:
Figure BDA0003654622030000184
where n is 24, the objective function is:
Figure BDA0003654622030000185
wherein k is the number of clusters; n is k Is the sample number; m is k Is the average of the classification samples.
The flow of the ED-based unsupervised k-means algorithm can be described as:
(1) the number of initial center clusters is k-4. To select the initial center point of the training set of protection devices, the two farthest distant samples x are selected from the training set i1 And x i2 As the first and second initial clustering points satisfying the objective function:
Figure BDA0003654622030000191
wherein x ij And x il Are any two samples in the training set T. Then, a third initial clustering point satisfying the following conditions is selected:
Figure BDA0003654622030000192
also, a fourth initial clustering point x satisfying the following condition is selected i4 ：
Figure BDA0003654622030000193
Thus, the initial values of the training set are expressed as:
Figure BDA0003654622030000194
(2) initial classification by ED calculation
Figure BDA0003654622030000195
(3) Recalculating new cluster centers based on the initial classification results obtained in step (2)
Figure BDA0003654622030000196
(4) Obtaining a new clustering result which meets the condition function:
Figure BDA0003654622030000197
(5) repeating the steps t times until two continuous calculations are the same, and meeting the following conditions: if it is not
Figure BDA0003654622030000198
Then t ═ t +1 and return to step (3); if it is not
Figure BDA0003654622030000199
The iterative process is ended and the final grouping result is recorded.
2.3 MD Algorithm based on semi-supervision
Compared with ED, available covariance σ based on MD algorithm ij ＝cov(x i ,x j ) To describe the correlation between variables without being affected by the dimension. It has advantages in processing distance metric learning between multivariate high-dimensional samples.
In order to fully utilize the useful information contained in the unlabeled data, the present invention proposes an online sequential ELM-MD (OSELM-MD). And applying the correct marked sample to the expansion of the new semi-marked sample by using a semi-supervised learning algorithm with reversible reasoning capability, judging that the wrong training sample is marked as an invalid sample, and then correcting the parameters.
The training set T contains a label set T 1 And unlabeled setT 2 The cost function of the semi-supervised OSELM-MD is expressed as:
Figure BDA0003654622030000201
in the formula, N 0 Is the number of marked samples;
Figure BDA0003654622030000202
is the number of implicit layer nodes;
Figure BDA0003654622030000203
is a sample vector of markers; n is the dimension of the sample vector;
Figure BDA0003654622030000204
is that
Figure BDA0003654622030000205
Class label sample vector of (1); c is the number of output nodes of the network; w is a i ＝[w i1 ,…,w in ] T Is an input weight vector between the input node and the implicit layer node; b i Is the offset of node i; g is an activation function of an implicit layer node; beta is a i ＝[β i1 ,β i2 ,…,β iC ] T An output weight vector between the hidden layer node and the output node; n is a radical of e Is the number of extended samples;
Figure BDA0003654622030000206
is an extended sample vector;
Figure BDA0003654622030000207
is that
Figure BDA0003654622030000208
The class of (2) marks the sample vector.
By calculation, the compound satisfying the formula (33) is obtained
Figure BDA0003654622030000209
Figure BDA00036546220300002010
In the formula, H 0 Outputting a matrix for the hidden layer of the marked sample set; β is the output matrix of weights; t is 0 Labeling a matrix for the category of the labeled sample; h e Outputting a matrix for a hidden layer of the extended sample; t is e Is a class label matrix of extended samples.
The G index of the overall sample is m-24, and the mean vector of the high-dimensional data sample is mu (mu) 1 ,μ 2 ,…,μ 24 ) ', global sample G equals (X) with single sample X 1 ,x 2 ,…,x 24 ) The MD between' is:
d 2 (X,G)＝(X-μ)’∑(X-μ) -1 (34)
the covariance between the evaluation indices is:
Figure BDA00036546220300002011
the solution is formulated as:
Figure BDA0003654622030000211
2.4 comparative analysis
Compared with ED, available covariance σ based on MD algorithm ij ＝cov(x i ,x j ) To describe the correlation between variables without being affected by the dimension. It has advantages in processing distance metric learning between multivariate high-dimensional samples.
In order to verify the validity and accuracy of these algorithms, the above supervised, unsupervised and semi-supervised machine learning methods were compared and analyzed. In order to ensure the fairness of the comparative analysis, the simulation test is completed on the same computer and the same sample.
Taking the protection device as an example, when the training set is s-80%, the results of roc (receiver operating characteristic) characteristic evaluation curves of the three algorithms are shown in fig. 7 (in the figure, the black dotted line is a random reference curve, which is a reference curve for 50% probability classification). The closer the ROC curve is to the upper left corner, the closer the value of the Area covered under the curve (AUC) is to 1, which means that the corresponding machine learning method has higher prediction performance. Calculation and analysis show that the AUC value of the supervised multiple regression analysis algorithm (blue curve) is 0.6084 e (0.5,1), so that the prediction result is superior to random prediction, and the method has certain reference significance. The AUC value of an ED-based unsupervised k-means algorithm (yellow curve) is 0.4952-0.5, namely the prediction result is similar to random prediction and has no reference meaning. The AUC value of the semi-supervised MD learning algorithm is 0.9464-1, namely the prediction result of the algorithm is very good and the result is relatively accurate. Table 6 shows the results of the comparative analysis of the three algorithms in terms of accuracy, processing time and reliability.
TABLE 6 evaluation results of protection devices
Figure BDA0003654622030000212
Figure BDA0003654622030000221
As can be seen from Table 6, the supervised multiple regression analysis algorithm takes 1.8864-2.0159 s, the ED-based unsupervised k-means algorithm takes 2.0135-2.9846 s, and the semi-supervised-based MD learning algorithm takes 460.1249-501.3124 seconds to obtain the result. Compared with the MD learning algorithm based on semi-supervision, the simulation calculation time of the supervised multiple regression analysis algorithm and the unsupervised K-means algorithm based on ED is shorter, but the accuracy rate of the algorithms is far lower than that of the MD learning algorithm based on semi-supervision (the AUC value is about 0.95, and is obviously superior to other two methods).
In addition, considering that the size of the training set is one of the main factors influencing the accuracy of the machine learning algorithm, 9 training subsets with the size of 10% to 90% are selected from the same sample set, and the accuracy analysis result is shown in a graphShown in fig. 8. The change trend of the regression analysis algorithm and the semi-supervised-based MD learning algorithm is observed, so that the scale of the training set influences the calculation precision, and when the samples are insufficient, the precision is sensitive to the scale of the training set. Δ r ═ 1-r i /r max ( i 10,20, …, 90%) represents the sensitivity threshold of the algorithm. When the training set size exceeds 70%, the accuracy of the evaluation of algorithm 1 (multiple regression analysis algorithm) will no longer be sensitive to training set size (Δ r) 1 < 5%). When the training set size exceeds 60%, the accuracy of the evaluation of algorithm 3 (semi-supervised based MD learning algorithm) is no longer sensitive to the training size (Δ r) 3 < 5%). The accuracy of the observation algorithm 2 (the non-supervision k-means algorithm based on ED) is not limited by the training scale. As can be seen from the figure, algorithm 3 has the lowest requirement for training set size. When the scale reaches 60%, the requirement of obtaining excellent evaluation precision is met.
3. Conclusion
In summary, the invention firstly provides a training set for the machine learning method, and establishes a state evaluation model based on an AHP fuzzy comprehensive algorithm for the M-C device, the merging unit, the protection device and the intelligent terminal; secondly, machine learning algorithms such as supervised multivariate regression analysis, unsupervised k-means and semi-supervised MD algorithms are applied to state evaluation of the relay protection system for more objective and accurate evaluation results, and comparison is carried out. Simulation results show that the machine learning algorithm provided by the invention has higher precision and lower data scale requirements.
The above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention 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 solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.

Claims (10)

1. The relay protection system risk assessment method based on the semi-supervised MD algorithm is characterized by comprising the following steps of:
s1, selecting static and real-time parameters of the M-C device, the merging unit, the protection device and the intelligent terminal as evaluation indexes, carrying out weight calculation on the evaluation indexes, and establishing a state evaluation model based on an AHP fuzzy comprehensive algorithm; the method for calculating the weight of the evaluation index specifically comprises the following steps: comparing every two of the nine-level judgment tables to obtain an importance comparison matrix, and processing the importance comparison matrix to obtain a judgment matrix; carrying out consistency check on the judgment matrix by using the safety random index, obtaining the judgment matrixes of the M-C device, the merging unit, the protection device and the intelligent terminal by transforming the matrixes, and calculating the evaluation index weights of the M-C device, the merging unit, the protection device and the intelligent terminal;
s2, establishing a relay protection system risk assessment model based on a semi-supervised MD algorithm to overcome the subjectivity defect of a fuzzy comprehensive algorithm.
2. The semi-supervised MD algorithm based risk assessment method for the relay protection system according to claim 1, wherein the formula of the judgment matrix is as follows:
Figure FDA0003654622020000011
the maximum eigenvalue of the judgment matrix is as follows:
Figure FDA0003654622020000012
the characteristic vector of the judgment matrix is as follows:
W＝[ω′ 1 ,ω' 2 ,...ω' n ]
wherein the comparison coefficient is defined as a ij ＝x i /x j ，i＝1,2,3,...,m；j＝1,2,3,...,m。
3. The semi-supervised MD algorithm based risk assessment method for the relay protection system according to claim 2, wherein the method for performing consistency check on the judgment matrix by using the safety random index comprises the following steps: consistency is defined as: and CR is CI/RI, RI is a random index, CI is a consistency check standard, CI is (lambda-N)/(N-1), lambda is the maximum characteristic value of the judgment matrix, and N is a nonzero characteristic root of the N-order consistency matrix.
4. The semi-supervised MD algorithm-based risk assessment method for the relay protection system according to claim 3, wherein the method for establishing the AHP fuzzy comprehensive algorithm-based state assessment model specifically comprises the following steps:
s11, considering fuzzy relations among all evaluation index variables, when the running state of the relay protection system is evaluated, the fuzzy relation of a continuous interval [0,1] is adopted to represent a fuzzy characteristic domain of the relay protection system, and the running state of the relay protection system comprises the following steps: normal, warning, error and hidden danger; the fuzzy characteristic domain of the relay protection system comprises: index set, weight set and evaluation set;
s12, carrying out standardization processing on the data of the fuzzy characteristic domain of the relay protection system, converting the actual value into an interval [0,1], and dividing the state evaluation index into: benefit type, cost type, interval type;
s13, calculating membership degree matrixes of each state evaluation index by using ridge distribution;
and S14, setting an index set, a weight set and an evaluation set according to the statistical data, taking the index set, the weight set and the evaluation set as a training set, and calculating the evaluation result of the protection device of the training set.
5. The semi-supervised MD algorithm based risk assessment method for the relay protection system according to claim 4, wherein the step S2 is implemented by establishing a semi-supervised MD algorithm based risk assessment model for the relay protection system to overcome the subjectivity defect of the fuzzy comprehensive algorithm as follows:
the cost function of the semi-supervised MD algorithm is as follows:
Figure FDA0003654622020000021
calculated to satisfy the following formula
Figure FDA0003654622020000022
The value of (c):
Figure FDA0003654622020000023
the G index of the overall sample is m-24, and the mean vector of the high-dimensional data sample is mu (mu) 1 ,μ 2 ,…,μ 24 ) ', global sample G and individual sample X ═ X (X) 1 ,x 2 ,…,x 24 ) The MD between' is:
d 2 (X,G)＝(X-μ)'∑(X-μ) -1
the covariance between the evaluation indices is:
Figure FDA0003654622020000024
the solution is formulated as:
Figure FDA0003654622020000031
in the formula, N 0 Is the number of marked samples;
Figure FDA0003654622020000032
is the number of implicit layer nodes;
Figure FDA0003654622020000033
is a sample vector of markers; n is the dimension of the sample vector;
Figure FDA0003654622020000034
is that
Figure FDA0003654622020000035
Class label sample vector of (1); c is the number of output nodes of the network; w is a i ＝[w i1 ,…,w in ] T Is an input weight vector between the input node and the implicit layer node; b i Is the offset of node i; g is an activation function of an implicit layer node; beta is a i ＝[β i1 ,β i2 ,…,β iC ] T An output weight vector between the hidden layer node and the output node; n is a radical of e Is the number of extended samples;
Figure FDA0003654622020000036
is an extended sample vector;
Figure FDA0003654622020000037
is that
Figure FDA0003654622020000038
Class label sample vector of (1); h 0 Outputting a matrix for the hidden layer of the marked sample set; β is the output matrix of weights; t is 0 Labeling a matrix for the category of the labeled sample; h e Outputting a matrix for a hidden layer of the extended sample; t is a unit of e Is a class label matrix of extended samples.
6. Relay protection system risk assessment system based on semi-supervised MD algorithm, its characterized in that includes: an operation state evaluation module and a risk evaluation module; the running state evaluation module is used for selecting static and real-time parameters of the M-C device, the merging unit, the protection device and the intelligent terminal as evaluation indexes, carrying out weight calculation on the evaluation indexes and establishing a state evaluation model based on an AHP fuzzy comprehensive algorithm; the method for calculating the weight of the evaluation index specifically comprises the following steps: comparing every two of the nine-level judgment tables to obtain an importance comparison matrix, and processing the importance comparison matrix to obtain a judgment matrix; carrying out consistency check on the judgment matrix by using the safety random index, obtaining the judgment matrixes of the M-C device, the merging unit, the protection device and the intelligent terminal by transforming the matrixes, and calculating the evaluation index weights of the M-C device, the merging unit, the protection device and the intelligent terminal; the risk evaluation module is used for establishing a relay protection system risk evaluation model based on a semi-supervised MD algorithm so as to overcome the subjectivity defect of a fuzzy comprehensive algorithm.
7. The semi-supervised MD algorithm based relay protection system risk assessment system as claimed in claim 6, wherein the formula of the judgment matrix is as follows:
Figure FDA0003654622020000039
the maximum eigenvalue of the judgment matrix is as follows:
Figure FDA0003654622020000041
the characteristic vector of the judgment matrix is as follows:
W＝[ω′ 1 ,ω' 2 ,…ω' n ]
wherein the comparison coefficient is defined as a ij ＝x i /x j ，i＝1,2,3,...,m；j＝1,2,3,...,m。
8. The semi-supervised MD algorithm based risk assessment system for the relay protection system as claimed in claim 7, wherein the method for performing consistency check on the judgment matrix by using the safety random index comprises the following steps: consistency is defined as: and CR is CI/RI, RI is a random index, CI is a consistency check standard, CI is (lambda-N)/(N-1), lambda is the maximum eigenvalue of the judgment matrix, and N is a nonzero characteristic root of the N-order consistency matrix.
9. The relay protection system risk assessment system based on semi-supervised MD algorithm as recited in claim 8, wherein the method for establishing the state assessment model based on the AHP fuzzy comprehensive algorithm specifically comprises the following steps:
(1) considering fuzzy relations existing among all evaluation index variables, when the running state of the relay protection system is evaluated, the fuzzy relation of a continuous interval [0,1] is adopted to represent a fuzzy characteristic domain of the relay protection system, and the running state of the relay protection system comprises the following steps: normal, warning, error and hidden danger; the fuzzy characteristic domain of the relay protection system comprises: index set, weight set and evaluation set;
(2) carrying out standardization processing on data of a fuzzy characteristic domain of the relay protection system, converting an actual value into an interval [0,1], and dividing state evaluation indexes into the following parts according to an exchange mode: benefit type, cost type, interval type;
(3) calculating a membership matrix of each state evaluation index by using ridge distribution;
(4) and setting an index set, a weight set and an evaluation set according to the statistical data, taking the index set, the weight set and the evaluation set as a training set, and calculating the evaluation result of the protection device of the training set.
10. The relay protection system risk assessment system based on semi-supervised MD algorithm as claimed in claim 9, wherein the method for establishing the relay protection system risk assessment model based on semi-supervised MD algorithm to overcome the subjectivity defect of the fuzzy comprehensive algorithm is as follows:
the cost function of the semi-supervised MD algorithm is as follows:
Figure FDA0003654622020000042
calculated to satisfy the following formula
Figure FDA0003654622020000043
The value of (c):
Figure FDA0003654622020000051
the G index of the overall sample is m-24, and the mean vector of the high-dimensional data sample is mu (mu) 1 ,μ 2 ,…,μ 24 ) ', global sample G equals (X) with single sample X 1 ,x 2 ,…,x 24 ) The MD between' is:
d 2 (X,G)＝(X-μ)'∑(X-μ) -1
the covariance between the evaluation indices is:
Figure FDA0003654622020000052
the solution is formulated as:
Figure FDA0003654622020000053
in the formula, N 0 Is the number of marked samples;
Figure FDA0003654622020000054
is the number of implicit layer nodes;
Figure FDA0003654622020000055
is a sample vector of markers; n is the dimension of the sample vector;
Figure FDA0003654622020000056
is that
Figure FDA0003654622020000057
Class label sample vector of (1); c is the number of output nodes of the network; w is a i ＝[w i1 ,…,w in ] T Is an input weight vector between the input node and the implicit layer node; b i Is the offset of node i; g is an activation function of an implicit layer node; beta is a beta i ＝[β i1 ,β i2 ,…,β iC ] T As hidden layer nodesAn output weight vector with an output node; n is a radical of e Is the number of extended samples;
Figure FDA0003654622020000058
is an extended sample vector;
Figure FDA0003654622020000059
is that
Figure FDA00036546220200000510
Class label sample vector of (a); h 0 Outputting a matrix for the hidden layer of the marked sample set; β is the output matrix of weights; t is 0 Labeling a matrix for the category of the labeled sample; h e Outputting a matrix for a hidden layer of an extended sample; t is e Is a class label matrix of extended samples.
CN202210550103.XA 2022-05-20 2022-05-20 Relay protection system risk assessment method and system based on semi-supervised MD algorithm Pending CN115081811A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210550103.XA CN115081811A (en) 2022-05-20 2022-05-20 Relay protection system risk assessment method and system based on semi-supervised MD algorithm

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210550103.XA CN115081811A (en) 2022-05-20 2022-05-20 Relay protection system risk assessment method and system based on semi-supervised MD algorithm

Publications (1)

Publication Number Publication Date
CN115081811A true CN115081811A (en) 2022-09-20

Family

ID=83249042

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210550103.XA Pending CN115081811A (en) 2022-05-20 2022-05-20 Relay protection system risk assessment method and system based on semi-supervised MD algorithm

Country Status (1)

Country Link
CN (1) CN115081811A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115795276A (en) * 2022-11-22 2023-03-14 南京电力设计研究院有限公司 Secondary circuit pressing plate state evaluation method based on wavelet analysis and machine learning
CN116976682A (en) * 2023-09-22 2023-10-31 安徽融兆智能有限公司 Fuzzy algorithm-based operation state evaluation method for electricity consumption information acquisition system

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115795276A (en) * 2022-11-22 2023-03-14 南京电力设计研究院有限公司 Secondary circuit pressing plate state evaluation method based on wavelet analysis and machine learning
CN115795276B (en) * 2022-11-22 2024-04-26 南京电力设计研究院有限公司 Secondary circuit pressing plate state evaluation method based on wavelet analysis and machine learning
CN116976682A (en) * 2023-09-22 2023-10-31 安徽融兆智能有限公司 Fuzzy algorithm-based operation state evaluation method for electricity consumption information acquisition system
CN116976682B (en) * 2023-09-22 2023-12-26 安徽融兆智能有限公司 Fuzzy algorithm-based operation state evaluation method for electricity consumption information acquisition system

Similar Documents

Publication Publication Date Title
CN115081811A (en) Relay protection system risk assessment method and system based on semi-supervised MD algorithm
CN106371427B (en) Industrial process Fault Classification based on analytic hierarchy process (AHP) and fuzzy Fusion
CN105117602B (en) A kind of metering device running status method for early warning
CN109933905B (en) Mechanical equipment health state assessment method based on multi-dimensional early warning analysis
CN106651169A (en) Fuzzy comprehensive evaluation-based distribution automation terminal state evaluation method and system
CN104299115B (en) Secondary system of intelligent substation state analysis method based on Fuzzy C-Means Cluster Algorithm
CN110837866A (en) XGboost-based electric power secondary equipment defect degree evaluation method
CN103793854A (en) Multiple combination optimization overhead transmission line operation risk informatization assessment method
CN109491339B (en) Big data-based substation equipment running state early warning system
CN109711664B (en) Power transmission and transformation equipment health assessment system based on big data
CN113569462A (en) Distribution network fault level prediction method and system considering weather factors
CN112417627A (en) Power distribution network operation reliability analysis method based on four-dimensional index system
CN110705859A (en) PCA-self-organizing neural network-based method for evaluating running state of medium and low voltage distribution network
CN116148753A (en) Intelligent electric energy meter operation error monitoring system
Ying et al. Research on state evaluation and risk assessment for relay protection system based on machine learning algorithm
CN110428191B (en) Method for identifying fragile nodes of power distribution network
CN114386753A (en) Equipment risk comprehensive analysis early warning method based on main transformer load condition
Dong Combining unsupervised and supervised learning for asset class failure prediction in power systems
CN114553671A (en) Diagnosis method for power communication network fault alarm
CN112434887B (en) Water supply network risk prediction method combining network kernel density estimation and SVM
CN117154716B (en) Planning method and system for accessing distributed power supply into power distribution network
CN108446202A (en) A kind of judgment method of the safe condition of calculator room equipment
CN114167837B (en) Intelligent fault diagnosis method and system for railway signal system
CN116151799A (en) BP neural network-based distribution line multi-working-condition fault rate rapid assessment method
CN110580578A (en) Intelligent substation secondary system operation quality multi-layer evaluation method

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