An intelligent power grid security evaluation method oriented to index system screening optimization

Abstract

The application discloses a kind of index system screening optimization-oriented intelligent power grid security evaluation method, belong to information security field;Specifically:First, for each intelligent power grid system, input the security requirements and authoritative security standards of this kind of system, and formulate each initial security evaluation index system;Then, select N similar intelligent power grid systems as the object to be evaluated, optimize the initial security evaluation index system, and obtain the reduced index set X';By further calculating the weight of each index in the reduced index set X';Finally, using each index in the reduced index set X', combined with their weights, fuzzy comprehensive evaluation of N objects to be evaluated;The application proposes a measurement index system optimization algorithm based on the discrimination degree for the optimization problem of index system, combines the qualitative and quantitative requirements, optimizes the index system structure on the basis of maintaining the original index information amount, and reduces the calculation cost.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of information security, and mainly relates to an intelligent power grid security evaluation method oriented to index system screening and optimization. BACKGROUND

[0002] The intelligent power grid is a modernized power infrastructure ensuring the safety and reliability of the national power transmission and distribution system, and capable of meeting the future growth, energy saving and diversity service demand as well as environmental constraints, and has the characteristics of informatization, automation and interactivity. With the rapid improvement of the comprehensive national strength of China, a cross-regional interconnected national power grid with ultrahigh voltage as the backbone has been formed, which has expanded the coverage of the power grid and significantly improved the power transmission capacity, thus providing strong support for the development of the national economy. The intelligent power grid system involves various complex technologies including power grid basic technology, intelligent power generation, intelligent power transmission, intelligent power distribution and intelligent power utilization technology.

[0003] At present, with the wide application of the intelligent power grid in the national power transmission and distribution system, it faces many security risks, and the security evaluation demand of the intelligent power grid system is increasingly prominent. The security condition evaluation of the intelligent power grid system is the basis for ensuring its safe operation and the premise for effectively preventing accidents and improving the security condition of the system.

[0004] In the current overall architecture of information system security represented by the intelligent power grid, it usually includes information security protection and defense, security risk and threat identification and monitoring, security threat response and management, information system recovery and disaster recovery, and system security quantification and evaluation. The security evaluation of the information system, as a necessary part of the overall architecture of the information system security, can identify and analyze the risks faced by the information system, give the security level and vulnerable items of the information system, and provide security reinforcement strategy suggestions for the managers.

[0005] For the security evaluation and assessment of the information system, it usually includes the analysis of the security demand of the evaluation target, the construction of the security evaluation index system, the construction of the security evaluation algorithm model, the quantitative calculation of the security of the evaluation target, and the analysis and management of the security evaluation results. The selection and construction of the security evaluation index system is one of the important factors determining whether the information system security evaluation measurement framework is reasonable. Based on the reasonable construction strategy of the index system, the index system is optimized according to the index discrimination degree, rough set and information theory; and with the weight algorithm of subjective and objective fusion and the comprehensive evaluation algorithm based on fuzzy theory, a complete information system security evaluation model is constructed, which is of great significance for maintaining and managing the overall security of the information system

[0006] Existing information security standards constrain the evaluation dimensions of information security from different perspectives. For example, traditional information security attributes include confidentiality, integrity, availability, controllability, and non-repudiation. The security function requirements defined in the CC common criteria include security audit, communication, cryptographic support, user data protection, identification and authentication, security management, privacy, trusted security function (TSF) protection, resource usage, trusted evaluation target (TOE) access, and trusted path / channel. Meanwhile, the national standard GB / T 28448-2019 defines ten aspects for security evaluation of Internet of Things information systems, including security physical environment, secure communication network, security regional boundary, secure computing environment, security management center, security management system, security management organization, security management personnel, security construction management, and security operation management, including general requirements and extended requirements.

[0007] Rough set theory is based on the classification mechanism and can be widely applied to solve the classification problem of multi-attribute decision table. In reality, the indicators are not explicit attributes of "yes" or "no", but more likely to have fuzzy membership. This leads to some data sets that cannot be divided into mutually exclusive categories, and the attributes of the data are difficult to distinguish. Rough set is commonly used in machine learning, data mining, data statistical analysis, etc. to remove redundant items in data, optimize data content, and improve algorithm performance.

[0008] In information security evaluation, the determination of indicator weight is an important process. The current mainstream subjective weight determination algorithms include:

[0009] Expert survey method, also known as "Delphi method", analytic hierarchy process (AHP), PageRank function, comprehensive quantitative calculation, and OPIS method.

[0010] The closest prior art is an information security evaluation model based on combined weighting and fuzzy comprehensive evaluation. After analyzing the evaluation target requirements and specifying the security indicator system, the AHP / TOPSIS is used to calculate the subjective weight, the entropy weight method is used to calculate the objective weight, and a certain proportion of the two weights is combined to form the final weight result. Then, the deterministic data is converted into fuzzy set data, and the overall security membership degree and the final security level are calculated based on fuzzy comprehensive evaluation.

[0011] However, the above prior art has the following disadvantages:

[0012] 1. The evaluation indicator system in the traditional information system security evaluation model has coarse granularity, is difficult to quantify, and has low general applicability, which can cause evaluation problems.

[0013] 2). The existing security evaluation process lacks further optimization and evaluation of the index system. The lack of indicators will affect the analysis and evaluation results due to insufficient information, and too many indicators will produce information redundancy and increase the difficulty of analysis and calculation.

[0014] 3). In the traditional information system security evaluation process, the weight of the measurement index system is mostly determined according to the experience of expert evaluation, which is easy to cause deviation. A subjective and objective weight algorithm is needed to consider the correlation between indicators and the amount of information carried by indicators, and to optimize the subjectivity of the weight.

[0015] 4). The index construction and selection are relatively disconnected from the subsequent calculation, which makes the overall coupling of the security evaluation model worse. A certain calculation basis is needed to support the construction of the index system, and to provide guidance for the subsequent weight and comprehensive evaluation calculation. SUMMARY

[0016] The present application aims at the problems of difficulty in constructing the intelligent power grid security evaluation index system, strong subjectivity of the security evaluation algorithm, and poor coupling, and proposes an intelligent power grid security evaluation method for index system screening and optimization. The standardized intelligent power grid system security index system is constructed, the index system is optimized by using rough set and information theory, the system security quantification calculation is completed based on combination weighting and fuzzy comprehensive evaluation, and the scientific intelligent power grid system security level is obtained.

[0017] Step one, for each intelligent power grid system, input the security requirements and authoritative security standards of the system, and develop each initial security evaluation index system;

[0018] The top-level index of the security evaluation index system includes data transmission confidentiality, power grid function availability, power grid risk controllability, system access identifiability, power grid personnel organization, power grid risk management, enterprise grading, and power grid operation continuity.

[0019] Each top-level index includes several lower-level indexes. By analogy, the set of all bottom-level indexes is defined as U.

[0020] Step two, select N intelligent power grid systems of the same type as the evaluation objects, optimize the initial security evaluation index system, and obtain the reduced index set X'.

[0021] The specific process is as follows:

[0022] Step 201, develop the initial score of each bottom-level index of the initial security evaluation index system;

[0023] Step 202, divide the bottom layer indicators according to the number of sub-bottom layer indicators, and perform coarse-grained data mapping to obtain a plurality of indicator subsets and coarse-grained data corresponding to each subset;

[0024] N evaluation objects {M 001 ,M 002 ,...,M 00N} each corresponding to the initial score of all bottom layer indicators, form the numerical data D = {D 001 ,D 002 ,...,D 00N}; the indicator subset U i corresponds to the coarse-grained data The value range of i is the number of sub-bottom layer indicators.

[0025] Step 203, respectively substitute the coarse-grained data corresponding to each indicator subset into the condition attribute set in the rough set reduction algorithm, substitute N evaluation objects {M 001 ,M 002 ,...,M 00N} into the domain, select one of the most important indicators in the indicator subset as the decision attribute. Perform rough set reduction to obtain the results corresponding to each indicator subset after reduction, and combine to obtain the set X' rough .

[0026] The result corresponding to the indicator subset U i is B i ;

[0027] The rough set algorithm is as follows:

[0028] First, input {M 001 ,M 002 ,...,M 00N} as the domain; the indicator subset U i as the condition attribute set c, the value of the domain on the condition attribute set; artificially designate one of the indicators in U i as the decision attribute d. In the initial state, let the set B = c, and start calculation.

[0029] Then, traverse the coarse-grained data a ∈ B corresponding to each indicator in the set B, and calculate the dependency degree δ B (d) and δ B-a (d) of the set B on d before and after removing the element a;

[0030] The dependency degree δ B (d) calculation process is as follows:

[0031] POS B (d) = U X∈U / d BX

[0032]

[0033] wherein, B X is the approximation of X with respect to B, indicating that according to the value of the elements in the set B, the set of elements that must be classified, i.e., the largest definable set contained in X; card(X) represents the cardinality of the set X.

[0034] After the calculation is completed, if δ B (d) = δ B-a (d), the data a and the corresponding indicators are removed from B, and the reduced B is used to iterate again; otherwise, the remaining unvisited elements are iterated, and if all elements in B are visited, there is no element a such that δ B (d) = δ B-a (d), then B at this time is the final output B i of the present reduction algorithm.

[0035] Step 204, the information entropy and weight of each indicator in the bottom layer are calculated using the entropy weight method on the numerical data D, and the discrimination is further calculated, the indicators with discrimination lower than the threshold are deleted, and a new set X' entropy is obtained.

[0036] The specific calculation process is as follows:

[0037] For the indicator u i , the calculation formula of the information entropy is as follows:

[0038]

[0039]

[0040] p ij corresponds to the jth to-be-tested object in the ith indicator; d j is the jth vector value in the row vector of the numerical data D;

[0041] The entropy weight w i of the indicator u i is calculated as follows:

[0042]

[0043] m is the number of indicators in the bottom layer;

[0044] The discrimination ξ i of the indicator u i is calculated as follows:

[0045]

[0046] Step 205, according to the qualitative and quantitative comprehensive decision, the set X' rough and X' entropy Take the union to get the final index set X' after reduction.

[0047] Step three, calculate the weight of each index in the reduced index set X';

[0048] The calculation process is as follows:

[0049] First, for the i-th index in the reduced index set X', repeat the entropy weight method to calculate the information entropy and objective weight w bi .

[0050] Then, use the PageRank algorithm to iteratively calculate the subjective weight w ai of the index.

[0051] Specifically, according to the risk probability influence between different indexes in the reduced index set X', construct the risk probability transition matrix Q n is the number of indexes in the reduced index set X';

[0052] Then, normalize the column vector in Q n×n matrix to get the probability transition matrix S, and get the probability transition matrix G through iteration.

[0053] α is a fixed probability value.

[0054] After the convergence of the probability transition matrix G, the value of each element at the corresponding position is the subjective weight of the corresponding index, and the subjective weight of the i-th index is w ai .

[0055] Finally, combine the weight coefficients γ and β to get the final weight w i of the index. ai = γw bi .

[0056] The weight coefficient is calculated by the correction function shown in the following formula:

[0057]

[0058] Co(γ,β) = |γ-β|, γ<β

[0059] Step four, use the indexes in the reduced index set X' to combine their respective weights for fuzzy comprehensive evaluation.

[0060] First, set the fuzzy evaluation set V = {V1, V2, V3, V4, V5} to represent five levels of very high, high, medium, low, and very low, respectively.

[0061] Then, the simple index set X' reconstitutes the numerical data D', and each numerical data d is converted into a fuzzy evaluation vector using an isosceles triangle membership function:

[0062] v={r v1 (d),r v2 (d),r v3 (d),r v4 (d),r v5 (d)}

[0063] All fuzzy evaluation vectors corresponding to the N objects to be evaluated form a set V

[0064] For the set D V , the n*5 matrix elements are Using the weight W corresponding to the element = {w i}, i∈[1,n], the corresponding V 00I is calculated according to the following formula, wherein each position respectively corresponds to the membership values of high, higher, medium, lower and very low safety levels.

[0065]

[0066] Finally, the comprehensive safety level of the Ith object to be evaluated is calculated using the maximum membership principle.

[0067] Similarly, N V 00I are obtained, which form a set V N×5 , that is, the safety evaluation levels of the N objects to be evaluated are obtained as the results.

[0068] The present application has the following advantages:

[0069] 1) An intelligent power grid safety evaluation method oriented to index system screening and optimization, which refers to domestic and international information security evaluation standards and can be expanded and changed according to the specific target system characteristics of the implementation evaluation work.

[0070] 2) An intelligent power grid safety evaluation method oriented to index system screening and optimization, in the subjective and objective combined index weight algorithm, the application scene needs and restrictions can be flexibly modified to other commonly used subjective weight algorithms, such as AHP method, Delphi method, etc.

[0071] 3) An intelligent power grid safety evaluation method oriented to index system screening and optimization, the fuzzy comprehensive evaluation algorithm can be selected according to the data type of the actual index, which is suitable for the quantitative calculation of qualitative index, and can be replaced by algorithms such as TOPSIS if the data is quantitative data.

[0072] 4), an index system screening optimization-oriented smart grid security evaluation method, according to national standards, CC general guidelines and other benchmarks, a standardized fine-grained security measurement index system construction strategy is proposed to support the security evaluation and measurement of information systems.

[0073] 5), an index system screening optimization-oriented smart grid security evaluation method, for the optimization problem of the index system, a measurement index system optimization algorithm based on discriminability is proposed, combined with qualitative and quantitative requirements, the index system structure is optimized on the basis of maintaining the original index information amount, and the calculation cost is reduced.

[0074] 6), an index system screening optimization-oriented smart grid security evaluation method, a fuzzy comprehensive evaluation algorithm based on subjective and objective combination right establishment is constructed to complete index right establishment and comprehensive evaluation calculation. The weight is determined by fully considering the subjective opinions of users and experts and the characteristics of objective data, and the quantitative calculation of qualitative indexes is completed by applying fuzzy theory.

[0075] 7), an index system screening optimization-oriented smart grid security evaluation method, the encapsulation and integration of modules are completed, and an index system screening optimization-oriented smart grid security evaluation model is proposed to provide method support for smart grid system security evaluation. BRIEF DESCRIPTION OF DRAWINGS

[0076] Figure 1 The index system screening optimization-oriented smart grid security evaluation method of the present application is a flowchart;

[0077] Figure 2 The subject security target formalization mechanism based on mapping of the present application;

[0078] Figure 3 The information system security evaluation and measurement index system supporting standardization of the present application;

[0079] Figure 4 The fuzzy comprehensive evaluation algorithm based on subjective and objective combination right establishment of the present application;

[0080] Figure 5 The index system screening optimization-oriented smart grid security evaluation model of the present application;

[0081] Figure 6 The security evaluation index system of the smart grid system of the present application. DETAILED DESCRIPTION

[0082] The present application will be further described in detail below with reference to the accompanying drawings.

[0083] The application discloses an intelligent power grid security evaluation method based on index system screening and optimization, relates to security quantification and evaluation in an information system security architecture, focuses on problems in index selection, construction, synthesis and calculation in an information security evaluation process, innovatively proposes a construction strategy of a fine-grained security measurement index system based on standardization, and is connected with index right confirmation, quantitative calculation and other evaluation processes to form a whole, and an intelligent power grid security evaluation model based on index system screening and optimization is constructed.

[0084] As shown in the figure, the specific steps are as follows: Figure 1

[0085] Step one, for each intelligent power grid system, an initial security evaluation index system is formulated by summarizing security requirements of the system, in combination with relevant authoritative security standards of the intelligent power grid;

[0086] The uppermost index of the security evaluation index system includes data transmission confidentiality, power grid function availability, power grid risk controllability, system access identifiability, power grid personnel organization, power grid risk management, enterprise grading and power grid operation continuity.

[0087] Each upper index contains a plurality of lower indexes, and a set of the bottom indexes is defined as U.

[0088] Step two, N intelligent power grid systems of the same type are selected as evaluation objects, and the initial security evaluation index system is optimized to obtain a simple index set X'.

[0089] The N intelligent power grid systems of the same type are used as the evaluation objects, and the purpose is to have enough sample systems and corresponding index data to support subsequent algorithms.

[0090] The specific process is as follows:

[0091] Step 201, initial scores of the bottom indexes of the initial security evaluation index system are respectively formulated;

[0092] Step 202, the bottom indexes are divided according to the number of the next bottom indexes, coarse-grained data mapping is performed, a plurality of index subsets and coarse-grained data corresponding to the subsets are obtained, and the coarse-grained data is obtained by using the following formula:

[0093] N evaluation objects {M 001 ,M 002 ,...,M 00N} are scored according to the bottom index set U by using expert evaluation, and corresponding numerical data D={D 001 ,D 002 ,...,D 00N ​} According to the coarse-grained level mapping relationship, all data in D is coarse-grained to obtain coarse-grained data D rough , and the bottom layer index is divided according to each time one layer index to obtain a plurality of index subsets U i and corresponding coarse-grained data The value range of i is the number of the next bottom layer index.

[0094] Step 203, for each index subset set , a rough set reduction algorithm is used for reduction, and the reduced index subset is recombined to obtain set X' rough .

[0095] The rough set reduction algorithm is defined as follows:

[0096] Definition 1: For a given formal information system M=(U,A,V,f);

[0097] U is non-empty, representing the domain, that is, the finite set of evaluation objects; A is a non-empty finite set of all objects in U, A=c∪d and Where c is the condition attribute, and d is the decision attribute. The value domain of attribute A i is V, and f:U×A→V is an information function, for If the decision attribute of the system , the system is called a decision system.

[0098] For system M, B∈A, the binary relation ind(A) is called the indistinguishable relation of M:

[0099]

[0100] U / ind(A) is all equivalence classes of the equivalence relation ind(A).

[0101] Definition 2: In information system M, if , then the positive domain POS P (Q) of Q is defined as:

[0102] POS P (Q)=U X∈U / Q P X

[0103] Where, P X is the lower approximation of X about P, which represents the set of elements that must be classified under the knowledge P, that is, the largest definable set contained in X.

[0104] Definition 3: The dependency degree δ represents the importance of the attribute, and the dependency degree of the sub-attribute B to the attribute set c is defined as follows: ​

[0105]

[0106] where card(X) denotes the cardinality of set X.

[0107] The rough set algorithm is as follows:

[0108] First, input {M 001 ,M 002 ,…,M 00N} as the domain, the index subset U i as the condition attribute set c, is the coarse-grained data corresponding to the domain on the condition attribute set. One of the most important indexes in U i is manually selected as the decision attribute d. Initially, let the set B = c, and start the calculation.

[0109] Subsequently, traverse the coarse-grained data a e B corresponding to each index in B, and calculate the dependency degree δ B (d) and δ B-a (d) of the set B on d before and after removing the element a.

[0110] After the calculation, if δ B (d) = δ B-a (d), remove the data a and the corresponding index from B, and use the reduced B to re-traverse; otherwise, continue to traverse the remaining elements that have not been accessed. If all elements in B have been accessed, and there is no element a such that δ B (d) = δ B-a (d), then the B at this time is the final output B i of the reduction algorithm.

[0111] Step 203, for each index subset, input U i , and manually specify the decision attribute, execute the rough set algorithm to obtain the result B i , and finally merge all B i to obtain the set X' rough .

[0112] Step 204, use the entropy weight method to calculate the information entropy and weight of each index for the entire numerical data D of the bottom layer index U, further calculate the discrimination degree, delete the indexes with a discrimination degree lower than the threshold, and obtain a new set X' entropy .

[0113] The specific calculation process is as follows:

[0114] First, input the data D, and D is expanded into an m x N matrix, where m is the number of bottom layer indexes in the set U. For each index u i∈U, i∈[1,m], corresponding to a vector D of length N. i ={d1,d2..,d j ,…,d N When the value range of each indicator is different, it is necessary to adjust the data for each indicator D. i Normalization is then performed. Assuming normalization for each indicator has already been completed, all D... i The values ​​in the vector are all in the range [0,1].

[0115] Then, based on the definition of information entropy in information theory, for each index u i The formula for calculating its information entropy is as follows:

[0116]

[0117]

[0118] The index u is calculated using the following formula. i entropy weight w i :

[0119]

[0120] The index u is calculated using the following formula. i Discrimination ξ i :

[0121]

[0122] Based on the above calculation results, the indicator system can eliminate indicator items with a discrimination index lower than the threshold (depending on the actual situation), thus obtaining X′. entropy .

[0123] Step 205: Based on the comprehensive decision-making process combining qualitative and quantitative methods, set X' rough and X' entropy Take the union of the sets to obtain the simplified final index set X'.

[0124] Step 3: Calculate the weight of each indicator in the simplified indicator set X';

[0125] The calculation process is as follows:

[0126] First, for the i-th indicator in the simplified indicator set X', the objective weight w is calculated using the entropy weight method. bi ;

[0127] The calculation process is the same as in step two; simply recalculate under the new simplified index set X' system.

[0128] Then, the subjective weight w of this indicator is calculated iteratively using the PageRank algorithm.ai ;

[0129] Specifically, a risk probability transition matrix is constructed according to the influence of risk probability between different indexes in the simple index set X' n is the number of indexes of the current index system;

[0130] Then, Q n×n The column vector in the matrix is normalized to obtain a probability transition matrix S, and the probability transition matrix G is obtained through iteration

[0131] Alpha is the probability of a user randomly arriving at a node, generally taking 0.85.

[0132] The value corresponding to each node position in the converged probability transition matrix G is w ai .

[0133] Finally, the final weight w is obtained by combining the weight coefficients gamma and beta. i = gamma w ai + beta w bi .

[0134] The weight coefficient is calculated by a correction function as shown in the following formula.

[0135]

[0136] Co(gamma, beta) = |gamma - beta|, gamma < beta

[0137] Step four, using each index in the simple index set X' and combining the respective weights, fuzzy comprehensive evaluation is performed;

[0138] First, set the fuzzy evaluation set V = {V1, V2, V3, V4, V5} to represent five levels of very high, high, medium, low, and very low, respectively;

[0139] Then, for the numerical index evaluation data of N objects to be evaluated under the current index system X':

[0140] D' = {D' 001 , D' 002 ,..., D' 00N}

[0141] For each numerical data d in D', use the isosceles triangle membership function to convert it into a fuzzy evaluation vector:

[0142] v = {r v1 (d), r v2 (d), r v3 (d), r v4 (d), r v5 (d)}

[0143]

[0144]

[0145]

[0146]

[0147]

[0148] All fuzzy evaluation vectors corresponding to N to-be-evaluated objects form a set For the set D V The element of the n*5 matrix Using the weight W corresponding to the element = {w i}, i [1, n], the corresponding V is calculated according to the following formula 00I Wherein each position respectively corresponds to the membership value of high, higher, medium, lower and very low security level:

[0149]

[0150] Finally, the set D V After fuzzy vectorization is comprehensively evaluated by a fuzzy comprehensive evaluation method, N V 00I Form a set V N×5 ; The maximum membership degree principle is used to calculate the comprehensive security level of each system = {M 001 , M 002 ,..., M 00N}, that is, the safety evaluation level of N systems is obtained as the result.

[0151] Embodiment:

[0152] At present, the evaluation index strength in the traditional information security evaluation model is rough and difficult to quantify, and the present example proposes a subject security target formalization mechanism based on mapping to guide the construction of an information security quality evaluation multi-layer index tree system, and the specific idea is as shown in Figure 2 .

[0153] The present application is based on the requirement of security attribute, combined with the requirements of CC general criteria and national standards, and adopts UML language to form an information system security evaluation index measurement system containing a 5-layer index system, and the specific is as shown in Figure 3 .

[0154] The index system is respectively from top to bottom security target layer, security attribute layer, security function layer, security control layer and security index layer.

[0155] Security target layer represents the requirement of the whole information system security, and is the specific performance of the information system security evaluation result.

[0156] Security attribute layer comprehensively considers the information security attribute, CC general rule and national standard requirement, and sets 8 requirements after fusion, including confidentiality, identifiability, controllability, availability, personnel management, risk management, enterprise grading and continuity.

[0157] Security function layer is a further decomposition and subdivision of security attribute layer, and security control layer is a further interpretation and refinement of security function layer. Security index layer is a set of clear and quantifiable security indexes, and their service target values represent the corresponding security capability level that the information system should provide.

[0158] In the index screening link, the qualitative and quantitative combination method can more finely distinguish the importance of the index, and guarantee the controllability of the index screening granularity. Therefore, the embodiment proposes a metric index system optimization algorithm based on the discrimination degree, reduces the indexes through the rough set theory, calculates the discrimination degree of the evaluation index through the entropy weight method, and then completes the optimization of the evaluation index, so that the distribution of the selected index is more reasonable. Finally, by comparing the discrimination degree before and after the index reduction, the index system optimization result is obtained.

[0159] In the actual evaluation process, a subjective index with higher or the highest importance is usually selected as the decision attribute D reference to solve the problem that the evaluation result cannot be generated prior to the index screening process. In addition, the index importance is closely related to the actual application system, and the classification ability of the index is not completely equivalent to the index importance in the evaluation. Directly executing the rough set algorithm on the whole index set may lead to the problem of excessive index reduction. In the index decomposition process, it is assumed that the number of index decomposition layers is n, and the rough set will reduce the index subset composed of the bottom layer sub-index of each n-1 layer index, so as to ensure the comprehensiveness of index decomposition and the scientificity of index reduction.

[0160] The index discrimination degree metric calculation process based on the entropy weight method is as follows:

[0161] Suppose that there are n information systems as evaluation objects and the statistical data is relatively comprehensive, and K security attributes are formed. For the first security attribute, m evaluation indexes (security indexes) x1, x2, …, xm are selected. m The statistical data of the n information systems and the corresponding m indexes form an n*m order matrix X n×m :

[0162]

[0163] wherein x ij represents the value of the ith evaluation index of the jth information system (i = 1, 2,..., m; j = 1, 2,..., n).

[0164] Since the units of measurement of each index are not uniform, the index needs to be standardized, i.e. the relative value of the index is processed into an absolute value, so as to solve the problem of too large difference caused by different dimensions of the index. X m×n is standardized to obtain X' m×n , wherein x' ij represents the value of the ith evaluation index of the jth information system after normalization, considering the positive and negative problems of the index, the normalization of the positive index adopts the following formula:

[0165]

[0166] The normalization of the negative index adopts the following formula:

[0167]

[0168] According to the definition of information entropy in information theory, the calculation formula of the information entropy of a group of data is as follows:

[0169]

[0170]

[0171] wherein p ij represents the probability of each possible event, -ln p ij represents the uncertainty function of the amount of information contained in each possible event. If p ij = 0, then define lim pij→0 p ij ln p ij = 0.

[0172] The value range of information entropy E i is [0, 1], in actual cases, E i = 1 indicates that the index does not provide any valuable information for the evaluation object; E i = 0 indicates that only the index can complete the measurement of the evaluation object, which indicates that the remaining indexes do not contain any value, which is not consistent with the actual situation, therefore, E i ≠ 0.

[0173] Further, in the evaluation problem of (m, n), the entropy weight calculation formula of the ith index is:

[0174]

[0175] where 0≤w i ≤1 and

[0176] From the definition of entropy weight, it can be analyzed that the closer the entropy value is to 1, the closer the entropy weight is to 0, the smaller the change of the index represents, the smaller the amount of information carried, and the failure to provide effective information, which can be considered to be eliminated.

[0177] Further, the calculation formula of the discrimination degree of the i-th index is:

[0178]

[0179] According to the above calculation results, the index system can eliminate the index items with a discrimination degree less than the threshold value (determined according to actual conditions), and form a new safety attribute index system

[0180] Suppose that the index set of the original index system is X, and X' is obtained by screening using the rough set reduction algorithm rough , X' is obtained by screening using the entropy weight method combined with the threshold value entropy , and the final new index set X' = X' rough ∪X' entropy to ensure that the screened-out indexes meet the reduction conditions under both qualitative and quantitative requirements. In order to combine rough set and entropy weight method to reduce the index set, unnecessary evaluation indexes are eliminated, and the best evaluation analysis of the evaluated information system can be made with as few evaluation indexes as possible.

[0181] Since the rough set reduction is based on the fact that the discrimination degree of the index after reduction does not change, the indexes that do not contribute to the classification of the target are screened out. Therefore, the overall discrimination degree of the index system is calculated using the discrimination degree weighted by the entropy weight method. The overall discrimination degree calculation formula of the index system is as follows:

[0182]

[0183] where n is the number of bottom-level indexes, w i and ξ i are the corresponding index weights and discrimination degrees calculated by the entropy weight method. Suppose that the overall discrimination degree before index screening is and the overall discrimination degree after screening is If , it is proved that the index optimization processing is effective.

[0184] For the research of information security evaluation model, the measurement algorithm is the key to quantitative evaluation of the evaluation subject. The data of the evaluation index system obtained in the previous step are integrated into the algorithm to obtain the comprehensive evaluation result. The general form of the information security measurement algorithm is:

[0185] m = f t(x1,x2,…,x i ),i≥1

[0186]

[0187] where f t is the index combination function, x i is the index value of the measure, i is the number of indexes, S is the information security quality degree, f g is the global comprehensive measure function, m n is the local measure value or index combination value, w i is the weight corresponding to the index value, and n is the number of factors describing the information system. Generally, for a specific evaluation object, the greater the n value, the more objectively and accurately S can express the security quality of the system, but too large n value can also cause index redundancy and increase the calculation difficulty.

[0188] The principles and calculation processes of the subjective index weight algorithm based on index correlation and PageRank idea, the objective weight algorithm based on entropy weight method, and the fuzzy comprehensive evaluation method are shown in Figure 4 ;

[0189] 1) Comprehensive weight algorithm based on subjective and objective combination

[0190] The weight calculation of the weight algorithm based on index correlation evaluation ranking is based on the following assumptions: if a node receives a greater weight of other chain-in links, the node is more important, and the quality of the chain-in nodes pointing to node A is different, and the higher quality node transmits more weight to other nodes through the link, so the higher quality node points to node A, and node A is more important.

[0191] The formula for calculating the PR value of each node is

[0192]

[0193] wherein, is the set of all nodes that have outlinks to p i node, is the number of outlinks of node p j , N is the total number of nodes, and a is the probability of a user randomly arriving at a node, generally taken as 0.85. According to the PR value calculation formula, the PR value of each node can be calculated, and when the iteration tends to be stable, the final result is set as the subjective weight w ai .

[0194] The objective weight is constructed by using the entropy weight method, and the entropy weight calculation formula according to the index optimization part can be obtained w bi .

[0195] Therefore, the weight setting of the index follows the following formula:

[0196] w i =αw ai +βw bi

[0197] Among them, w i w represents the new weights after linear data fusion. ai To determine the subjective weights obtained using a weighting method based on indicator correlation evaluation and ranking, w bi The objective weights are obtained using the entropy weighting method, where α and β are weight coefficients and α + β = 1.

[0198] To rationally allocate the weights of subjective and objective factors, the concept of a correction function is introduced. This function ensures that the distance between weights aligns with the distance between their coefficients, resulting in a more reasonable weight distribution. The formula for the correction function is as follows:

[0199]

[0200] Co(α,β)=|α-β|,α<β

[0201] 2) Establish the fuzzy comprehensive evaluation factor set U

[0202] The fuzzy comprehensive evaluation factor set U is established as follows:

[0203] (1) Divide * into n subsets according to its specific attributes and content, denoted as U = {U1, U2, ..., U...} n}

[0204] (2) For each subset, if its factor U i If it contains m sub-indicators, it can be further subdivided according to its characteristics, denoted as U. i ={U i1 U i2 ,…,U im}

[0205] 3) Establish evaluation set V and make comprehensive judgments

[0206] The evaluation set refers to the set of possible evaluation results that evaluators may produce for a given indicator, denoted as V = {V1, V2, ..., V...} t (Assume there are t evaluation results in total).

[0207] First, for deterministic evaluation data, a membership function is used to transform the data into a fuzzy evaluation set V. Then, by mapping a specific indicator to the evaluation results in the evaluation set, the membership degree of the evaluated indicator to the evaluation set results is obtained. Assuming the factor set U contains n lower-level indicators, the i-th indicator is evaluated, and the membership degree of the i-th indicator to the j-th element in the evaluation set V is obtained, denoted as r. ij , with r ijThe membership matrix is constituted by the row, that is

[0208]

[0209] Therefore, the fuzzy comprehensive evaluation model is:

[0210]

[0211] Wherein, b i (i = 1, 2,..., t) represents the membership of the system security to the i-th evaluation level in the evaluation set V.

[0212] 4) Establish the fuzzy comprehensive evaluation factor set U

[0213] For the evaluation result B, the maximum membership degree method is selected as the evaluation index v = {v i |max(b i ) → v i}, that is, the evaluation set with the maximum membership degree in the fuzzy evaluation result B is taken as the final evaluation result, that is, the final security level of the evaluation target.

[0214] The fourth technical means is used to completely construct the intelligent power grid security evaluation model for index system screening and optimization, and the model can meet the security evaluation requirements of the existing intelligent power grid system, and has strong adaptability to system confidentiality, robustness and complexity; through the standardization index system construction strategy and optimization mechanism, the index system can dynamically adapt to the specific demand characteristics of different systems in actual work, and the indexes with large information amount and high calculation efficiency are screened; the fuzzy comprehensive evaluation algorithm combining subjective and objective factors is used to solve the problem that the weight is affected by a single subjective or objective factor, and the quantitative calculation weight of the index optimization is used to fuse the objective factors, thereby enhancing the correlation between system links and saving the calculation cost.

[0215] Taking the intelligent power grid system as an example, the security evaluation model of the application is used for security evaluation, and the technical effect of the application is verified. For the evaluation of the intelligent power grid information system, the influence relationship between the indexes involved and the information difference carried by the indexes are considered, the comprehensive weight calculation algorithm based on the combination of subjective and objective factors is used, and then the fuzzy comprehensive method is selected to calculate the overall security evaluation level. The intelligent power grid security evaluation model for index system screening and optimization is as shown in Figure 5 .

[0216] The security evaluation index system of the intelligent power grid system is as shown in Figure 6 . In order to simplify the verification process, the data transmission confidentiality is selected as the security attribute for evaluation, and the specific index of confidentiality is as shown in Table 1:

[0217] Table 1

[0218]

[0219]

[0220]

[0221] The aforementioned evaluation model was used to evaluate four similar evaluation objects, M001, M002, M003, and M004. The data collection results, simulating the real system through expert representative evaluation, are shown in Table 2.

[0222] Table 2

[0223]

[0224]

[0225] Because the coarse set reduction algorithm has strong filtering capabilities, it is necessary to perform coarse-grained level mapping on the index value D to obtain D. rough To ensure the feasibility of the rough set algorithm, the mapping relationship is shown in Table 3.

[0226] Table 3

[0227]

[0228] Binary index

[0229]

[0230]

[0231] For the current indicator data, according to the security control layer u jp The lowest-level indicators are divided into indicator subsets, and for each indicator subset, its coarse-grained mapping data D is applied. rough The rough set reduction algorithm was used for each. Simultaneously, the entropy weight method was used to calculate the corresponding weights and discrimination of the statistical data D of all bottom-level indicators, with a discrimination threshold γ = 0.015. The calculation process and reduction results are shown in Table 4.

[0232] Table 4

[0233]

[0234]

[0235] Based on the comprehensive decision-making process combining qualitative and quantitative methods, the final reduced set of indicators is as follows:

[0236] {u 112 ,u 113 ,u 121 ,u 122 ,u 123u 131 u 141 u 143 u 145 u 151 u 152 u 161 ,

[0237] u 171 u 213 u 222 u 231 u 241 u 251 u 252 u 261 u 262 u 311 u 321 u 331}

[0238] The overall discriminant degree is calculated for the index system and data before and after reduction At this time The effectiveness of index reduction is theoretically proved.

[0239] Then, the subjective and objective combined weighting is carried out using the reduced index system and simulation data. The objective weight w bi The weight result reserved when the index reduction entropy weight method is calculated is used. The subjective weight w ai is calculated using PageRank algorithm. Considering the influence of risk probability between different indexes of the system, the expert evaluation result is obtained and the risk probability transfer matrix Q

[0240]

[0241] The column vector in Q is normalized to obtain the probability transfer matrix S.

[0242]

[0243] Then, the final probability transfer matrix G is calculated:

[0244]

[0245] Where U is an all-1 matrix of the same size as S, and N is the current number of indexes. According to the algorithm iteration, the subjective weight w ai is obtained. From the current subjective and objective weights, the Co(w ai , w bi ) = 0.14, the proportion coefficient α = 0.43, and β = 0.57. The final subjective and objective combined weighting weight calculation result is shown in Table 5.

[0246] Table 5

[0247]

[0248]

[0249] Then, fuzzy comprehensive evaluation is started.

[0250] The process adopts a fuzzy evaluation set V={V1, V2, V3, V4, V5}={very high, higher, medium, lower, very low}, and uses an isosceles triangle membership function r of the following formula v (u ijk ) to convert the index accurate numerical data into a fuzzy evaluation vector.

[0251]

[0252]

[0253]

[0254]

[0255]

[0256]

[0257] After fuzzy comprehensive evaluation method is used to comprehensively evaluate the fuzzy vectorized data, the fuzzy calculation result vector is shown in Table 6:

[0258] Table 6

[0259]

[0260] According to the maximum membership degree principle, the following is obtained:

[0261] System comprehensive safety level={M001, M002, M003, M004}={higher, higher, very poor, very poor}.

[0262] By now, for the smart grid system, a multi-level decomposition feedback comprehensive safety evaluation model based on index correlation analysis is established, an evaluation index system is established, the information security quality evaluation result is a safety level through the ordering-based right assignment algorithm and the fuzzy comprehensive algorithm, and the effectiveness of the information security evaluation is verified.

Claims

1. A method for smart grid security assessment oriented to index system screening optimization, characterized in that, The specific steps are as follows: First, for each smart grid system, input the security requirements and authoritative security standards of each system, and formulate each initial security evaluation index system; Then, select The same intelligent power grid system as the object to be evaluated, the initial security evaluation index system is optimized, and the simple index set ; The specific process is as follows: Step I, the initial security evaluation index system is divided into several subsets, and each subset is corresponding to a coarse-grained data; The rough set reduction algorithm is as follows: the object to be evaluated the initial scores of all the bottommost indicators of each constitute the numerical data ; indicator subset corresponding to the coarse-grained data , the value range is the number of sub-bottom indicators Step III, the coarse-grained data corresponding to each index subset is respectively substituted into the condition attribute set in the rough set reduction algorithm, and one to be evaluated object is taken as the domain, one most important index in the index subset is selected as the decision attribute, the rough set reduction is executed, the results corresponding to each index subset after reduction are obtained, and the set is obtained after merging. ; The specific calculation process is as follows: First, input a subject to be evaluated as the domain; a subset of indicators as the set of condition attributes , the value of the domain on the set of condition attributes; artificially specified one of the indicators in the middle as the decision attribute ; Initially, let the set Start the calculation: Then, the coarse-grained data corresponding to each index in the set is traversed, and the dependence degree and of the set on is calculated. and ; Dependency The calculation process is shown below: ； ； wherein, is with respect to the lower approximation of , denoted by , is the set of elements in that will necessarily be classified, i.e. the largest definable set contained in denotes the cardinality of the set ; After the calculation is completed, if , the data and the corresponding indicators are removed from , and the traversal is re-performed using the reduced ; otherwise, the traversal of the remaining unvisited elements is continued, and if all elements in are visited, there is no element such that , then the at this time is the final output result of the indicator subset of the current reduction algorithm ; By analogy, the results of the reduction of all subsets of indicators are combined into a set ; Step IV, numerical data The information entropy and weight of each index of the bottom layer are calculated using the entropy weight method, and the discrimination is further calculated to delete the indexes with a discrimination lower than a threshold value to obtain a new set ; The topmost index of the security evaluation index system includes: data transmission confidentiality, grid function availability, grid risk controllability, system access identifiable, grid personnel organization, grid risk management, enterprise classification, and grid operation continuity; To index The formula for calculating the information entropy is as follows: ； ； the first indicator corresponds to the first to-be-evaluated object; the first vector value in the numerical data row vector; Indicators Entropy weight of the indicators The calculation formula is: ; is the number of the lowest level indicators; Indicator Discrimination The calculation formula is: ； Step V. Take the union of the sets and to obtain the final reduced set of indicators ; Next, the simple index set The weight of each index is calculated. For the first index, the weight includes objective weight of entropy weight method and subjective weight calculated by iteration using PageRank algorithm. ;​​ Finally, the fuzzy comprehensive evaluation of each object to be evaluated is carried out by using the simple index set and combining the respective weights of each index.

2. The index system-oriented screening optimization intelligent power grid security evaluation method according to claim 1, characterized in that, The process of fuzzy comprehensive evaluation of each object to be evaluated is as follows: Each upper-level indicator contains several lower-level indicators; and so on, the set of all the bottom-level indicators is defined as .

3. The index system-oriented screening optimization intelligent power grid security evaluation method according to claim 1, characterized in that, For the simple index set In the first Subjective weights are calculated Specifically: According to the reduced indicator set The risk probability transfer matrix is constructed according to the risk probability influence between different indicators , The number of indicators in the reduced indicator set ​ Then, the probability transition matrix is obtained by normalizing the column vectors in the matrix The probability transition matrix is obtained by normalizing the column vectors in the matrix The probability transition matrix is obtained by iterating ; is a fixed probability value; probability transition matrix After convergence, the value at the corresponding position of each element is the subjective weight of the corresponding indicator. The subjective weight of each indicator is ; Finally, the weight coefficient and is calculated by a correction function shown in the following equation. ； ； obtaining the most significant weight of the index .

4. The index system oriented screening optimization intelligent power grid security evaluation method according to claim 1, characterized in that, ​ First, the fuzzy evaluation set is defined representing very high, high, medium, low and very low, respectively; Then, the simple index set Reorganize the numerical data Each numerical data , using the isosceles triangle membership function to Convert into fuzzy evaluation vector: ； all the fuzzy evaluation vectors corresponding to the to-be-evaluated objects form a set ; for each set of matrix elements , using the weight corresponding to the element , the corresponding is calculated as follows where each position corresponds to the membership value of high, higher, medium, lower and very low security level respectively: ； Finally, the comprehensive safety level of the i-th object to be evaluated is calculated using the maximum membership principle. the maximum membership principle. Similarly, the following is obtained one constituting a set That is, the following is obtained a safety evaluation level of the object to be evaluated as a result.

Images