D-number-based human error event correlation analysis method for nuclear power plants

Abstract

The present application relates to the technical field of risk assessment, in particular to a nuclear power plant human error event correlation analysis method based on D number, which comprises the following steps: determining the factors influencing the correlation degree of human error events and the structural relationship thereof, determining the weight of each influencing factor, establishing a membership function and an anchor point for describing the correlation degree, converting the semantic evaluation of each influencing factor by multiple experts into D number, determining the difference between D numbers, correcting the reliability and obtaining the reliability weight coefficient, determining the combination order of multiple groups of D numbers to guide the combination of D numbers, integrating the D number set by comprehensively considering the weight of each influencing factor and the semantic evaluation thereof, and outputting the correlation degree and reliability of human operation in the nuclear power plant. The present application determines the reliability of D numbers by measuring the difference between D numbers, solves the problem that the combination order is difficult to determine due to the non-compliance with the commutative law of multiple groups of D numbers, has good effect in the nuclear power plant human reliability analysis, and has important theoretical significance and application value.

Description

Technical Field

[0001] This invention relates to the field of risk assessment technology, specifically to a method for correlation analysis of human error events in nuclear power plants based on the D-number. Background Technology

[0002] In the design and risk assessment of large and complex systems, especially in the operation of critical systems like nuclear power plants where human intervention is required, human factors play a crucial role in the safe operation of nuclear power plants, and human error is one of the main causes of nuclear power plant accidents. Human Factors Reliability Analysis (HRA), as a systematic framework, can effectively assess the system risks that may result from human intervention. Correlation analysis within HRA can assess the impact of human error-induced task failure on the probability of subsequent task failures when there is a correlation between tasks.

[0003] Human error event correlation analysis can intuitively present decision-makers with the relationship between the degree of correlation between events and the probability of human error, helping them to avoid such risks. If an accident does occur, correlation analysis can also more effectively identify the cause, thereby designing targeted accident prevention strategies. To conduct human error event correlation analysis in highly uncertain environments, it is necessary to consider the impact of various properties such as temporal proximity, task relevance, and behavioral similarity on the degree of correlation; its essence is a multi-attribute decision-making problem.

[0004] Dempster's evidence theory (DS) algorithm, as one of the effective methods for handling multi-attribute decision-making, can extend the basic event space in probability theory into a power set of basic events, also known as the identification frame. It establishes a basic probability assignment function on the identification frame and proposes the Dempster combination rule to fuse uncertain information in group decision-making. Existing methods combine DS evidence theory with the analytic hierarchy process (AHP) for relevance evaluation. However, when dealing with multi-attribute decision-making problems, it is important to note that the evidence theory framework is based on the assumption of mutual exclusivity of elements in the identification frame. This assumption limits the rationality of evidence theory in evaluating semantic information.

[0005] D-number theory is an extension of DS evidence theory, effectively expressing uncertain and incomplete information. It breaks the constraint of mutually exclusive elements in the identification framework of evidence theory and incorporates the open world concept from the initial stage of generating the basic probability assignment function. It allows for incomplete trust levels even with incomplete cognition and hesitation, a characteristic highly suitable for relevance analysis scenarios involving semantic evaluation, making it an effective means of handling such problems. Existing methods for relevance evaluation using D-number and a relevance analysis model based on D-number and the analytic hierarchy process—a decision-making trial and evaluation laboratory method—have also been proposed. However, the D-number combination rules do not satisfy the commutative and associative laws. These methods directly perform sequential combination when combining multi-expert evaluation information and use average weighting in the D-number construction process, ignoring the differences between evaluation levels and negatively impacting the results.

[0006] Compared to previous methods, this invention proposes the concept of D-number support based on Li distance, uses the prospect value function to correct the D-number confidence and uses this confidence ranking to determine the order of D-number combinations. After obtaining a reasonable and reliable combination order, D-number combination and integration are performed. Under the premise of effectively expressing and measuring uncertain information, this invention further improves the rationality and credibility of the human error event correlation assessment results. Summary of the Invention

[0007] The summary section of this invention provides a brief overview of the concepts, which will be described in detail in the detailed description section that follows. This summary section is not intended to identify key or essential features of the claimed invention, nor is it intended to limit the scope of the claimed invention.

[0008] To address the technical problem of low rationality and reliability in correlation analysis of human error events, this invention proposes a correlation analysis method for human error events in nuclear power plants based on the D-number.

[0009] This invention provides a method for correlation analysis of human error events in nuclear power plants based on the D-number, the method comprising:

[0010] Based on the pre-obtained set of influencing factors from the correlation analysis, the hierarchical structure relationship is determined;

[0011] Based on the semantic evaluations of each influencing factor in the set of influencing factors obtained in advance by multiple experts, the membership function of the descriptive relevance level and its anchor point are established, and the D number is constructed.

[0012] Determine the weight coefficient of each influencing factor in the set of influencing factors;

[0013] Based on hierarchical relationships, the constructed D numbers are combined;

[0014] Integrate the combined D numbers;

[0015] The degree of relevance of human intervention is determined based on the integrated D-number and the weight coefficients of each influencing factor in the set of influencing factors.

[0016] Furthermore, the construction of the D-number based on the semantic evaluations of each influencing factor in the influencing factor set by multiple experts obtained in advance, the membership function of the established descriptive relevance level, and its anchor points includes:

[0017] Obtain semantic evaluations of each influencing factor in the set of influencing factors from multiple experts. The discontinuous elements in the semantic evaluations can form a non-empty set Ω = {b1, b2, b3, ..., b...}. n}, where n is the number of semantic evaluation levels in set Ω, b n The semantic evaluation made by experts, that is, the evaluation results made by experts, in which, And when i≠j, b i ≠b j For any v i ≥0 and The number D is represented as D = {(b1, v1), (b2, v2), ..., (b...}. i v i ), ..., (b n v n Using the formula The semantic evaluations of various influencing factors by multiple experts were transformed into confidence D-scores and lack of confidence D-scores, where b c For the expert assessment corresponding to the confidence D number, v c Let m be the membership function corresponding to the confidence D number. y For expert E y The anchor point corresponding to the semantic evaluation of the influencing factors, |m y |for m y The modulus, α, is the reliability of the expert evaluation, and r y For expert E y The degree of hesitation in semantic evaluation of influencing factors is used to construct the confidence D-number, denoted as D. c =(b c v c Using formula b I =b c , Constructing a D-number of insecurity is called D. I =(b I v I ), where b I For the expert assessment corresponding to the lack of confidence in the D number, v I Let D be the membership function corresponding to the number of unconfident D.

[0018] Furthermore, a membership function describing the relevance level and its anchor points are established, including:

[0019] Establish a membership function that describes the relevance level;

[0020] Based on these correlation levels, the correlation level of the influencing factors is assessed and quantified, and is referred to as the anchor point, where the value range of the anchor point is [0, 1].

[0021] Furthermore, the combination of the constructed D numbers based on hierarchical relationships includes:

[0022] Through formula Calculate the i-th D number d i and the j-th D number d j The distance between them, Li, and There are two (2) N ×2 N ) order matrix, The elements in the formula are (A, B∈2) Ω The expression is obtained by finding that || is the modulo operation, and A and B are elements of the power set Ω. The elements in the formula We obtain, where i ≠ j, S1, S2 ∈ 2 Ω The variable i represents the degree of non-exclusivity E. ij The relative matrix formed The row number of the first element in set S1 is given by the matrix RM, while j represents the column number of the first element in the relative matrix S2. The degree of non-exclusivity is determined by the formula... Obtain, of which L i and L j Let i and j represent n language constants, and L represent the language constants. i and L j The area of ​​the intersection between them is S ij L i and L j The area of ​​the union between them is U ij ;

[0023] From the i-th D number d i The Li distance d between the j-th D number and the j-th D number D-number (d i d j Construct the dissimilarity matrix DM among the n D numbers. The dissimilarity matrix DM is:

[0024]

[0025] Similarity metric sim ijFrom the formula sim(d) i d j )=1-d D-number (d i d j The similarity measure matrix (SMM) is derived as follows:

[0026]

[0027] Through formula The support of the D-number is calculated as sup(i), where sim(d i d j ) represents the i-th D number d i The similarity between the j-th D number and the j-th D number;

[0028] The i-th D number d i Credibility c i According to formula c i =sup(i)-1 is calculated, and then the formula is used. and formula Obtain the i-th D number d i Local credibility CD i and global credibility CD g ;

[0029] Based on the obtained local and global credibility and the prospect value function, we obtain D. i Correction credibility ACD i The corresponding formula is:

[0030]

[0031] Where ξ and η represent the risk situation, and ω represents the loss characteristics;

[0032] Normalizing the corrected confidence level yields the formula for the normalized corrected confidence level:

[0033]

[0034] in, This is the normalized corrected confidence level obtained by normalizing the corrected confidence level.

[0035] Based on the i-th D number D i Normalized correction confidence Determine the i-th number D i weighting factor w i The corresponding formula is:

[0036]

[0037] Where e is the natural constant, and exp{} is an exponential function with the natural constant e as its base;

[0038] The D-numbers are fused based on their weighting factors.

[0039] Furthermore, the process of fusing the D-numbers based on their weighting factors includes:

[0040] The D-numbers are sorted according to their weight factors, and multiple D-numbers are merged. The merging order of the D-numbers is determined by their weights, with D-numbers with smaller weights being selected for merging first. The order of combining multiple D-numbers should be consistent with their weight order from smallest to largest. The formula corresponding to the final combination result of multiple D-numbers is:

[0041]

[0042] Among them, w i <w j <w k w i w j and w k D i D j and D k The weighting factor, the i-th D number D i With the j-th D number D j Combinations are denoted as The rule for combining D numbers can be denoted as: D(b) = v, C is:

[0043]

[0044] in,

[0045] Furthermore, the integration of the combined D-numbers includes:

[0046] Through formula Perform D-number integration to obtain the F values ​​of each influencing factor. X The aggregation operator I(D) of confidence level D and lack of confidence level D CX ) and I(D IX ).

[0047] Furthermore, determining the degree of relevance of human intervention based on the integrated D-number and the weight coefficients of each influencing factor in the set of influencing factors includes:

[0048] According to the formula P(D)=∑I(D) CX )·W X Calculate the correlation degree P(D) of human intervention, where I(D)CX ) is the influencing factor F X The confidence D-number aggregation operator for (X = 1, 2, ...), W X The corresponding weight coefficient is used to determine the confidence level α using the following formula:

[0049]

[0050] Among them, I(D) IX Influencing factor F X The degree of self-confidence D-number aggregation operator.

[0051] Furthermore, determining the weight coefficients of each influencing factor in the set of influencing factors includes:

[0052] Using the analytic hierarchy process (AHP), the weight coefficients W of each influencing factor are calculated, and (F1, ..., F2) are set. i F n If is a set of n influencing factors, then the comparison matrix obtained based on expert evaluation is M. n×n =[m ij ], where element m ij Represents the decision element F i Relative to F j The judgment of relative importance satisfies the following conditions:

[0053]

[0054] After constructing the judgment matrix, calculate M of the judgment matrix. n×n Maximum eigenvalue λ max and its corresponding eigenvectors And through the formula The consistency index CI and the consistency ratio CR are obtained through the formula. The formula is used to calculate RI, where RI is the random consistency index. The value of RI is related to the dimension x of the comparison matrix: x = 1, RI = 0; x = 2, RI = 0; x = 3, RI = 0.52; x = 4, RI = 0.89; x = 5, RI = 1.12; x = 6, RI = 1.26; x = 7, RI = 1.36.

[0055] If CR < 0.1, the matrix consistency test is considered passed; otherwise, it is not, and the comparison matrix needs to be reconstructed. After passing the consistency test, the judgment matrix M is determined. n×n The largest eigenvalue λ max Corresponding feature vector w in i That is, the priority value corresponding to the i-th scheme, where λ max satisfy

[0056] The present invention has the following beneficial effects:

[0057] The method proposed in this invention has a simple structure, a clear and straightforward process, and a reasonable design, making it convenient to implement and use. This invention performs fuzzy classification on multiple complex pieces of evidence, significantly reducing the confusion and complexity of the target information, resulting in more accurate and practically significant predictions. This invention scientifically explains the differences between pieces of evidence, which is logically sound. This invention uses the membership degree between pieces of evidence as a statistical characteristic for classifying known evidence, which is a prerequisite for fuzzy classification, providing a basis for fuzzy classification, reflecting the actual situation between pieces of evidence, and effectively suppressing the influence of conflicting evidence on the fusion results, resulting in better focusing effects. Furthermore, based on determining the structural relationship of the correlation degree of human error events affecting nuclear power plants, it can transform the semantic evaluations of different influencing factors by multiple experts into D-numbers that can effectively express uncertain information. By measuring the degree of difference between multiple sets of D-numbers and combining it with the prospective value function for correction, the credibility weight coefficients of multiple sets of D-numbers are obtained. By sorting multiple D-number combinations by credibility weight coefficients, the problem of difficulty in achieving multiple D-number combinations due to the lack of commutative property in traditional D-number combination rules is overcome. This reduces the ambiguity and subjectivity of expert semantic evaluation. The correlation degree and credibility of reasonable and reliable nuclear power plant human operations are obtained by using D-number integration. Attached Figure Description

[0058] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0059] Figure 1 The flowchart shows the correlation analysis method for human error events in nuclear power plants based on D-number according to the present invention.

[0060] Figure 2 This is a schematic diagram illustrating the structural relationship of the five influencing factors in this invention;

[0061] Figure 3 This is a detailed flowchart of the present invention. Detailed Implementation

[0062] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the specific implementation methods, structures, features, and effects of the technical solution proposed according to the present invention are described in detail below with reference to the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0063] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0064] This invention provides a method for correlation analysis of human error events in nuclear power plants based on the D-number. The method includes the following steps:

[0065] Based on the pre-obtained set of influencing factors from the correlation analysis, the hierarchical structure relationship is determined;

[0066] Based on the semantic evaluations of each influencing factor in the set of influencing factors obtained in advance by multiple experts, the membership function of the descriptive relevance level and its anchor point are established, and the D number is constructed.

[0067] Determine the weight coefficient of each influencing factor in the set of influencing factors;

[0068] Based on hierarchical relationships, the constructed D numbers are combined;

[0069] Integrate the combined D numbers;

[0070] The degree of relevance of human intervention is determined based on the integrated D-number and the weight coefficients of each influencing factor in the set of influencing factors.

[0071] The following is a detailed explanation of each of the above steps:

[0072] refer to Figure 1 The flowchart illustrates some embodiments of the method for correlation analysis of human error events in nuclear power plants based on the D-number according to the present invention. This method for correlation analysis of human error events in nuclear power plants based on the D-number includes the following steps:

[0073] Step 101: Determine the hierarchical structure relationship based on the set of influencing factors obtained in the pre-acquired influence correlation analysis.

[0074] In some embodiments, hierarchical relationships can be determined based on a pre-obtained set of influencing factors from a correlation analysis.

[0075] As an example, first, obtain the F values ​​of each factor that influences the correlation analysis. XThe influencing factors set includes all factors affecting the correlation analysis. Then, the influencing factors in the set are hierarchically structured to determine the relationships between them. The hierarchical relationship can be any relationship between the influencing factors.

[0076] For example, a working model for human error accidents in nuclear power plants can include five influencing factors: temporal proximity (F1), cue similarity (F2), goal similarity (F3), behavioral similarity (F4), and task relevance (F5). The structural relationships among these five influencing factors are as follows: Figure 2 As shown.

[0077] Step 102: Construct the D-number based on the semantic evaluations of each influencing factor in the influencing factor set by multiple experts obtained in advance, the membership function of the descriptive relevance level and its anchor point.

[0078] In some embodiments, confidence D-numbers and lack of confidence D-numbers can be constructed based on the semantic evaluations of each influencing factor in the set of influencing factors obtained in advance by multiple experts, i.e., the evaluation results made by experts on each influencing factor.

[0079] Among them, multiple experts can be at least two experts.

[0080] As an example, we obtain the semantic evaluations of each influencing factor in the set of influencing factors from multiple experts. The discontinuous elements in the semantic evaluations can form a non-empty set Ω = {b1, b2, b3, ..., b}. n}, where n is the number of semantic evaluation levels in the set Ω. n The semantic evaluation made by experts, that is, the evaluation results made by experts. Among them, And when i≠j, b i ≠b j For any v i ≥0 and The number D can be represented as D = {(b1, v1), (b2, v2), ..., (b...} i v i ), ..., (b n v n Using the formula The semantic evaluations of various influencing factors by multiple experts were transformed into confidence D-scores and lack of confidence D-scores, where b c For the expert assessment corresponding to the confidence D number, v c Let m be the membership function corresponding to the confidence D number. y For expert E y The anchor point corresponding to the semantic evaluation of the influencing factors, |m y |for m yThe modulus, α, is the reliability of the expert evaluation, and r y For expert E y The degree of hesitation in semantic evaluation of influencing factors is used to construct the confidence D-number, denoted as D. c =(b c v c Using formula b I =b c , Constructing a D-number of insecurity is called D. I =(b I v I ), where b I For the expert assessment corresponding to the lack of confidence in the D number, v I Let D be the membership function corresponding to the number of unconfident D.

[0081] For example, through formula By combining the relevance index with the semantic evaluations of each expert, a confidence score (D-score) is constructed. c Using formula b I =b c , Constructing a number of self-confidence D I .

[0082] In temporal proximity (F1), D can be constructed by expert E1 evaluation. c ={(0.500, 0.4)}, D I ={(0.500, 0.6)}; D can be constructed based on expert E2's evaluation. c ={(0.500, 0.4), (0.750, 0.1)}, D I ={(0.500, 0.1), (0.750, 0.4)}; D can be constructed based on expert E3's evaluation. c ={(0.625, 0.4)}, D I ={(0.625, 0.6)}.

[0083] In the cue relevance (F2), D can be constructed by expert E1 evaluation. c ={(0.375, 0.4)}, D I ={(0.375, 0.6)}; D can be constructed based on expert E2's evaluation. c ={(0.500, 0.2)}, D I ={(0.500, 0.8)}; D can be constructed based on expert E3's evaluation. c ={(0.375, 0.2)}, D I ={(0.375, 0.7)}.

[0084] In target similarity (F3), D can be constructed by expert E1 evaluation. c ={(0.250, 1.0)}, D I ={(0.250, 0)}; D can be constructed based on expert E2's evaluation. c ={(0.250, 1.0)}, D I ={(0.250, 0)}; D can be constructed based on expert E3's evaluation. c ={(0.875, 0.2)}, D I ={(0.250, 0)}.

[0085] In behavioral similarity (F4), D can be constructed by expert E1 evaluation. c ={(0.875, 0.2)}, D I ={(0.875, 0.8)}; D can be constructed based on expert E2 evaluation. c ={(0.750, 0.2), (1.0, 0.2)}, D I ={(0.750, 0.3), (1.0, 0.3)}; D can be constructed based on expert E3's evaluation. c ={(0.750, 0.2), (1.0, 0.2)}, D I ={(0.750, 0.4), (1.0, 0.1)}.

[0086] As another example, a membership function describing the level of relevance and its anchor points can be established.

[0087] For example, firstly, a membership function describing the correlation level is established. Then, based on these correlation levels, the correlation level of the influencing factors is evaluated and quantified, and this is called the anchor point. In this invention, the value range of the anchor point is [0, 1]. The closer the value is to 1, the higher the correlation level, and vice versa.

[0088] For example, anchor points can be constructed for each influencing factor. The relevance of influencing factors can be divided into five levels: no relevance [ZD], low relevance [LD], moderate relevance [MD], high relevance [HD], and perfect relevance [CD]. Based on these five relevance levels, the anchor point for each influencing factor is determined and quantified. The anchor point is the index corresponding to each relevance level. Table 1 can include the semantic evaluations of each influencing factor made by the first expert E1, the second expert E2, and the third expert E3, as well as their confidence and lack of confidence in their evaluations. Tables 2, 3, 4, and 5 show the relevance level of each influencing factor and its corresponding index.

[0089] The expert assessment of the relevance of human intervention can be shown in Table 1.

[0090] Table 1

[0091]

[0092] The anchor points for the influencing factor "time proximity" can be shown in Table 2.

[0093] Table 2

[0094]

[0095]

[0096] The anchor points for the influencing factor "task relevance" can be shown in Table 3.

[0097] Table 3

[0098]

[0099] The anchor points for the influencing factor "clue similarity" can be shown in Table 4.

[0100] Table 4

[0101]

[0102]

[0103] The anchor points for the influencing factor "target similarity" can be shown in Table 5.

[0104] Table 5

[0105]

[0106] Step 103: Determine the weight coefficient of each influencing factor in the set of influencing factors.

[0107] In some embodiments, the analytic hierarchy process (AHP) can be used to calculate the weight coefficient W of each influencing factor.

[0108] As an example, the analytic hierarchy process (AHP) can be used to calculate the weight coefficients W of each influencing factor. We can set (F1, ..., F...) i F n If is a set of n influencing factors, then the comparison matrix obtained based on expert evaluation is M. n×n =[m ij ], where element m ij Represents the decision element F i Relative to F j The judgment of relative importance satisfies the following conditions:

[0109]

[0110] After constructing the judgment matrix, calculate M of the judgment matrix.n×n Maximum eigenvalue λ max and its corresponding eigenvectors And through the formula The consistency index CI and the consistency ratio CR are obtained through the formula. Calculate RI, where RI is the random consistency index. The value of RI is related to the dimension x of the comparison matrix: x=1, RI=0; x=2, RI=0; x=3, RI=0.52; x=4, RI=0.89; x=5, RI=1.12; x=6, RI=1.26; x=7, RI=1.36.

[0111] Generally, if CR < 0.1, the matrix consistency test is considered passed; otherwise, it fails, and the comparison matrix needs to be reconstructed. After passing the consistency test, the judgment matrix M is determined. n×n The largest eigenvalue λ max Corresponding feature vector w in i That is, the priority value corresponding to the i-th scheme, where λ max satisfy

[0112] For example, the analytic hierarchy process (AHP) can be used to weight influencing factors. In the hierarchical structure of influencing factors, the bottom-level factors of cue similarity and target similarity are weighted first. Using the AHP, the weights of these two influencing factors can be calculated as follows:

[0113]

[0114] Where M1 is the comparison matrix between clue factor similarity and target similarity constructed by experts. W F2 W is the weight of clue relevance (F2). F3 The weights for target similarity (F3).

[0115] Similarly, the second level of the hierarchy measures factors influencing performance: time proximity, task relevance, and similarity. The weights of these three factors are calculated using the following formulas:

[0116]

[0117] Where M2 is a comparison matrix of temporal proximity, task relevance, and executor similarity. W F1 W F5 W F4 The weights are time proximity (F1), task relevance (F5), and behavioral similarity (F4), respectively. The consistency ratio (CR) of M2 is 0.003. Since 0.003 < 0.1, it meets the consistency test requirements.

[0118] Step 104: Based on the hierarchical structure relationship, combine the constructed D numbers.

[0119] In some embodiments, the constructed D-numbers can be combined based on hierarchical relationships.

[0120] As an example, this step may include the following steps:

[0121] The first step is to use the formula Calculate the i-th D number d i and the j-th D number d j The distance between them is Li. Where, and There are two (2) N ×2 N ) order matrix, The elements in the formula are (A, B∈2) Ω The expression is obtained by finding that || is the modulo operation, and A and B are elements of the power set Ω. The elements in the formula We obtain, where i ≠ j, S1, S2 ∈ 2 Ω The variable i represents the degree of non-exclusivity E. ij The relative matrix formed The row number of the first element in set S1 is given by the matrix RM, while j represents the column number of the first element in the relative matrix S2. The degree of non-exclusivity is determined by the formula... Obtain, of which L i and L j Let i and j represent n language constants. L i and L j The area of ​​the intersection between them is S ij L i and L j The area of ​​the union between them is U ij .

[0122] For example, the formula for determining the Li distance between the confidence D numbers of different influencing factors can be:

[0123]

[0124] Among them, E ij Let d = 0.2, i, j = 1, 2, 3, 4, 5. In the D-number constructed by multiple experts for temporal proximity (F1), d 11 =0,d 12 =0.071, d 13 =0.268, d 23 = 0.239. In the D-number constructed from cue relevance (F2), d11 =0,d 12 =0.213, d 13 =0.071, d 23 =0.153. In the D-number constructed based on behavioral similarity (F3), d 11 =0,d 12 =0.148, d 13 =0.164, d 23 =0.484.

[0125] The second step involves obtaining the i-th D number d from the first step included in this example. i The Li distance d between the j-th D number and the j-th D number D-number (d i d j Construct the dissimilarity matrix DM among the n D numbers. The dissimilarity matrix DM can be represented as:

[0126]

[0127] The greater the distance between two D numbers, the weaker their mutual support. When a D number significantly conflicts with other D numbers, it is considered to have a smaller impact on the final combination result. Similarity metric sim ij From the formula sim(d) i d j )=1-d D-number (d i d j The similarity measure matrix (SMM) is represented as follows:

[0128]

[0129] For example, in temporal proximity (F1), the dissimilarity matrix DM and the similarity metric matrix SMM can be:

[0130]

[0131] In cue relevance (F2), the dissimilarity matrix DM and the similarity measure matrix SMM can be:

[0132]

[0133] In behavioral similarity (F4), the dissimilarity matrix DM and the similarity metric matrix SMM can be:

[0134]

[0135] The third step can be done using formulas. The support of the D-number is calculated as sup(i), where sim(d id j ) represents the i-th D number d i The similarity between the j-th D number and the j-th D number.

[0136] For example, in temporal proximity (F1), sup(1) = 2.661, sup(2) = 2.69, sup(3) = 2.493.

[0137] In the cue relevance (F2), sup(1) = 2.716, sup(2) = 2.634, sup(3) = 2.776.

[0138] In behavioral similarity (F4), sup(1) = 2.688, sup(2) = 2.368, sup(3) = 2.352.

[0139] Step 4, the i-th D number d i Credibility c i According to formula c i =sup(i)-1 is calculated, and then the formula is used. and formula Obtain the i-th D number d i Local credibility CD i and global credibility CD g .

[0140] For example, in temporal proximity (F1), c1 = 1.661, c2 = 1.69, c3 = 1.493, CD1 = 0.3429, CD2 = 0.3489, CD3 = 0.3082, CD g =0.3333.

[0141] In the cue relevance (F2), c1 = 1.716, c2 = 1.634, c3 = 1.776, CD1 = 0.3348, CD2 = 0.3188, CD3 = 0.3464, CD g =0.3333.

[0142] In the behavioral similarity (F4) test, c1 = 1.688, c2 = 1.368, c3 = 1.352, CD1 = 0.3829, CD2 = 0.3103, CD3 = 0.3068, CD g =0.3333.

[0143] The fifth step is to combine the obtained local and global confidence levels with the prospect value function to obtain the corrected confidence level ACD of Di. i The corresponding formula can be:

[0144]

[0145] Where ξ and η represent the risk profile, both taken as 0.88. ω represents the loss characteristic, taken as 2.25.

[0146] For example, in temporal proximity (F1), ACD1 = 0.0168, ACD2 = 0.0257, and ACD3 = 0.0879.

[0147] In the cue relevance (F2), ACD1 = 0.0033, ACD2 = -0.0542, and ACD3 = 0.0131.

[0148] In behavioral similarity (F4), ACD1 = 0.0711, ACD2 = -0.0814, and ACD3 = -0.0922.

[0149] Step 6: Based on step 5 in this example, normalize the corrected confidence level to obtain the normalized corrected confidence level.

[0150] For example, the formula for obtaining the normalized correction confidence level can be:

[0151]

[0152] in, It is the normalized corrected confidence level obtained by normalizing the corrected confidence level.

[0153] For example, in temporal proximity (F1),

[0154] In cue relevance (F2),

[0155] In behavioral similarity (F4),

[0156] Step 7: Determine the weighting factor for each D number based on the normalized corrected confidence level of each D number.

[0157] For example, based on the i-th D number D i Normalized correction confidence Determine the i-th number D i weighting factor w i The corresponding formula can be:

[0158]

[0159] Where e is the natural constant, an irrational number approximately equal to 2.718281828459904523536... exp{} is an exponential function with the natural constant e as its base.

[0160] For example, in temporal proximity (F1), w1 = 0.4033, w2 = 0.4362, and w3 = 0.1605.

[0161] In the cue relevance (F2), w1 = 0.3873, w2 = 0.1648, and w3 = 0.4479.

[0162] In behavioral similarity (F4), w1 = 0.5679, w2 = 0.2232, and w3 = 0.2089.

[0163] Step 8: Merge the D numbers according to their weighting factors.

[0164] For example, multiple D-numbers are merged by sorting them according to their weight factors. The merging order of the D-numbers is determined by their weights; D-numbers with smaller weights are selected for merging first. The order in which multiple groups of D-numbers are combined should be consistent with their weight order from smallest to largest. Then, the final combination result of multiple D-numbers can be represented by the formula:

[0165]

[0166] Among them, w i <w j <w k w i w j and w k D i D j and D k The weighting factor. The i-th D number D i With the j-th D number D j Combinations are denoted as The rule for combining D numbers can be denoted as: Defined as D(b) = v, C can be:

[0167]

[0168] in,

[0169] For example, in temporal proximity (F1),

[0170] In cue relevance (F2),

[0171] In target similarity (F3),

[0172] In behavioral similarity (F4),

[0173] Step 105: Integrate the combined D numbers.

[0174] In some embodiments, the combined D-numbers can be integrated.

[0175] As an example, it can be done through the formula Perform D-number integration to obtain the F values ​​of each influencing factor. X The aggregation operator I(D) of confidence level D and lack of confidence level D CX ) and I(D IX ).

[0176] For example, in temporal proximity (F1), I(D) CX ) = 0.1013, I(D IX =0.1344.

[0177] In cue relevance (F2), I(D) CX ) = 0.0457, I(D IX = 0.1066.

[0178] In target similarity (F3), I(D) CX ) = 0.2500, I(D IX ) = 0.

[0179] In behavioral similarity (F4), I(D) CX )=0.0983,I(D IX = 0.4212.

[0180] Step 106: Determine the degree of relevance of human operation based on the integrated D number and the weight coefficients of each influencing factor in the set of influencing factors.

[0181] In some embodiments, the degree of relevance of human intervention can be determined based on the integrated D-number and the weight coefficients of each influencing factor in the set of influencing factors.

[0182] As an example, we can use the formula P(D)=∑I(D) CX )·W X Calculate the correlation degree P(D) of human intervention. Where, I(D) CX ) is the influencing factor F X The confidence D-number aggregation operator for (X = 1, 2, ...), W X This is its corresponding weighting coefficient. The formula for determining the credibility α can be:

[0183]

[0184] Among them, I(D) IX Influencing factor FX The degree of self-confidence D-number aggregation operator.

[0185] For example, P(D) = ∑I(D) CX ·W X = 1.0865. The formula for determining the confidence level α can be:

[0186]

[0187] The detailed flowchart of the present invention is as follows: Figure 3 As shown.

[0188] This invention features a simple structure, reasonable design, and clear objectives. Based on determining the structural relationships affecting the correlation of human error events in nuclear power plants, it transforms the semantic evaluations of different influencing factors by multiple experts into D-numbers that effectively express uncertain information. By measuring the differences between multiple sets of D-numbers and incorporating a prospective value function for correction, it obtains credibility weight coefficients for multiple sets of D-numbers. The order of combining multiple sets of D-numbers is obtained by ranking them using these credibility weight coefficients. This overcomes the problem of traditional D-number combination rules failing to satisfy the commutative law, which makes it difficult to achieve multiple D-number combinations. It also reduces the ambiguity and subjectivity of expert semantic evaluations. Through D-number integration, it obtains a reasonable and reliable correlation and credibility of human operations in nuclear power plants.

[0189] This invention effectively assesses and measures the correlation degree of human error in the human factors reliability analysis of nuclear power plants. It uses expert evaluations, which are inherently fuzzy and subjective, as input factors, and obtains a final, specific numerical value of the correlation degree through model calculation and quantification, providing a more intuitive understanding. When constructing the D-number, this invention considers the differences in correlation levels reflected by the membership function curves describing the correlation levels and incorporates these differences into the D-number construction, resulting in a D-number that better reflects the evaluation characteristics. This D-number is used to express the uncertainty of information and distinguish the importance of each level within a multi-level subset. Therefore, it more intuitively reflects the original differences in the data in the D-number, making it more suitable for problems involving linguistic evaluation processes. Furthermore, the fuzziness and subjectivity are reduced through the fusion calculation of the D-number, enhancing the relevance and specificity of the expert evaluation process. The results and comparisons using different combination rules are shown in Table 6. As shown in Table 6, for the relevance degree P(D) of human operation, the E-AHP method based on evidence theory and analytic hierarchy process yields the lowest result, while the D-AHP method based on D-number theory and analytic hierarchy process and the D-number and analytic hierarchy process-decision pilot and evaluation laboratory method yield the highest results. The result calculated by the present invention is in the middle. However, based on the same calculation example and expert evaluation, the calculation results of the three methods are basically consistent, which shows the rationality and effectiveness of the algorithm of the present invention in handling such problems.

[0190] The results and comparisons of using different combination rules are shown in Table 6.

[0191] Table 6

[0192]

[0193] Method 1 can be the E-AHP method. Method 2 can be the D-AHP method. Method 3 can be a laboratory method based on D-number and analytic hierarchy process (AHP) for decision pilot and evaluation.

[0194] In summary, the method proposed in this invention has a simple structure, a clear and straightforward process, and a reasonable design, making it convenient to implement and use. This invention performs fuzzy classification on multiple complex pieces of evidence, significantly reducing the confusion among the evidence and the complexity of the target information, resulting in more accurate and practically significant predictions. This invention scientifically explains the differences between pieces of evidence in a logically sound manner. By using the membership degree between pieces of evidence as a statistical characteristic for classifying known evidence, this invention provides a basis for fuzzy classification, reflects the actual situation between the evidence, and effectively suppresses the influence of conflicting evidence on the fusion results, resulting in better focusing effects.

[0195] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for D number-based human error event correlation analysis in a nuclear power plant, characterized by, The method comprises the following steps: According to the influence factor set obtained in advance for the influence correlation analysis, the hierarchical structure relationship is determined; According to the semantic evaluation of each influence factor in the influence factor set by multiple experts obtained in advance, the membership function and its anchor point for describing the correlation level are established, and the D number is constructed; The weight coefficient of each influence factor in the influence factor set is determined; The constructed D number is combined based on the hierarchical structure relationship; The combined D number is integrated; According to the integrated D number and the weight coefficient of each influence factor in the influence factor set, the correlation degree of human operation is determined; The constructed D number is combined based on the hierarchical structure relationship, comprising: Through formula Calculate the first Number of D and the Number of D The distance between them, Li, and There are two 1-th order matrix, The elements in the formula are , We obtain, where || is the modulo operation, and A and B are sets. Elements of the power set The elements in the formula Obtain, among which, , The variable i represents the degree of non-exclusivity. The relative matrix formed Middle set The row number of the first element, where j represents the relative position of the matrix RM. The column number of the first element, rather than the degree of exclusivity, is determined by the formula. Obtain, among which, and Let i and j represent n language constants. and The area of ​​the intersection between them is , and The area of ​​the union between them is ; The Li distance between the ith D number and the jth D number The difference degree matrix DM between the n D numbers is constructed, and the difference degree matrix DM is: similarity measure From the equation The similarity measure matrix SMM is derived as: The support of D number is calculated by formula wherein, is the similarity between the ith D number and the jth D number;​ the i-th D number the credibility of according to the formula is calculated, and then the i-th D number the local credibility of and the global credibility of are obtained according to the formula and the formula ; According to the obtained local credibility and global credibility combined with the foreground value function, the modified credibility of is obtained The corresponding formula is: wherein, and is a risk posture, is a loss characteristic; The normalized correction confidence is obtained by normalizing the correction confidence, and the formula corresponding to the normalized correction confidence is: wherein is a normalized modification confidence obtained by normalizing the modification confidence; According to the i-th D number of the normalized modified credibility , the weight factor of the i-th D number is determined The corresponding formula is: where e is the natural constant, is the exponential function with base the natural constant e; According to the weight factor of the D number, the D number is fused; According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: in, , , and for , and The weighting factor, the i-th D number With the j-th D number Combinations are denoted as The combination rule for D numbers can be written as: , , , C is: wherein , .

2. The method for analyzing the correlation of human error events in a nuclear power plant based on D number according to claim 1, characterized in that, According to the weight factor of the D number, the D number is fused, comprising: Obtain semantic evaluations of each influencing factor in the set of influencing factors from multiple experts. The discontinuous elements in the semantic evaluations can form a non-empty set. , where n is a set The number of semantic evaluation levels in Chinese. The semantic evaluation made by experts, that is, the evaluation results made by experts, in which, And when hour, For any and The D number is represented as Using the formula , The semantic evaluations of various influencing factors by multiple experts were transformed into confidence D-scores and lack of confidence D-scores, among which... For the expert assessment corresponding to the self-confidence D number, Let D be the membership function corresponding to the confidence D number. For experts The anchor point corresponding to the semantic evaluation of the influencing factors. for The modulus, To assess the reliability of expert evaluations. For experts The degree of hesitation in semantic evaluation of influencing factors is used to construct the confidence D-score. Using the formula , Constructing a D-number record of insecurity ,in, For experts who lack confidence in their D-score, Let D be the membership function corresponding to the number of unconfident D.

3. The method of claim 1, wherein, According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising:

4. The method of claim 1, wherein, According to the weight factor of the D number, the D number is fused, comprising: The D number set is integrated by formula The self-confidence D number and the non-self-confidence D number of each influencing factor are obtained by the aggregation operator and .

5. The method for D number based human error event correlation analysis of a nuclear power plant according to claim 4, characterized in that, According to the weight factor of the D number, the D number is fused, comprising: According to the formula The degree of correlation of human operation is calculated , wherein, is an influencing factor The self-confidence D number aggregation operator, is its corresponding weight coefficient, and the credibility The corresponding formula is: wherein, is an influencing factor of the degree of self-doubt D number aggregation operator.

6. The method of claim 1, wherein, According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor of the D number, the D number is fused, comprising: According to the weight factor The analytic hierarchy process is used to calculate the weight coefficient W of each influencing factor, and the weight coefficient W is set The set of n influencing factors is obtained according to the expert evaluation, and the comparison matrix is wherein the element represents the relative importance of the decision element relative to , and the judgment satisfies the following conditions: After constructing the judgment matrix, calculate the judgment matrix. Maximum eigenvalue and its corresponding eigenvectors And through the formula The consistency index CI and the consistency ratio CR are obtained through the formula. The formula is used to calculate RI, where RI is the random consistency index. The value of RI is related to the dimension x of the comparison matrix. For x=1, RI=0; for x=2, RI=0; for x=3, RI=0.52; for x=4, RI=0.89; for x=5, RI=1.12; for x=6, RI=1.26; for x=7, RI=1.

36. If the following condition is satisfied , it can be determined that the matrix consistency test is passed, otherwise, it is not passed, and the comparison matrix needs to be reconstructed; after passing the consistency test, the maximum eigenvalue of the matrix is determined, and the corresponding eigenvector is determined , that is, the priority value corresponding to the i th scheme, wherein satisfies .

Images