Thermal power generating unit health state assessment method based on improved weight fusion

Abstract

The invention discloses a thermal power generating unit health state evaluation method based on improved weight fusion. The method comprises the following steps: S1, constructing an index evaluation system by using an analytic hierarchy process; s2, determining the state level and relative deterioration degree numerical value division of the thermal power generating unit; s3, determining the weight distributed to each index; s4, calculating the membership degree of each index corresponding to each state grade as a comprehensive evaluation matrix in the fuzzy comprehensive evaluation method; s5, calculating evaluation sets of different parts in the index layer by using a fuzzy comprehensive evaluation method; s6, judging whether the evidence belongs to high-conflict evidence or not according to a synthesis rule, and S7, obtaining a final evaluation opinion of the thermal power generating unit by adopting an evidence theory improved by an SPD-DS method, the method starts from three key dimensions of conflict degree, similarity degree and uncertainty, calculates a correction coefficient of the evidence, corrects an original evidence body, effectively reduces conflicts among the evidences, and improves the evaluation accuracy of the thermal power generating unit. The corrected evidence synthesis result is more reasonable and reliable, and the health state of the thermal power generating unit is accurately reflected.

Description

Technical Field

[0001] The present invention relates to the technical field of thermal power unit health status assessment, and in particular to a thermal power unit health status assessment method based on improved weight fusion. Background Art

[0002] As the core equipment for power production, the health status of thermal power units is directly related to the stable power supply and efficient operation of the power system. Accurately assessing the health status of thermal power units can not only predict potential equipment failures in advance, reduce maintenance costs, and extend the service life of the units, but also effectively avoid power supply interruptions caused by unit failures, thereby ensuring the reliability and stability of the power system.

[0003] Given the importance of health status assessment of thermal power units, a clear and simple classification system has been formed for the weight calculation method of complex system evaluation indicators, which mainly includes three categories: subjective weighting method, objective weighting method and subjective and objective weighting method.

[0004] The subjective weighting method is a decision-making matrix constructed based on expert opinions, in which the weights of indicators rely entirely on the subjective judgment and experience of the experts. The hierarchical analysis method is the most commonly used one of this method; however, the subjective weighting method has certain limitations. Its judgment matrix is difficult to accurately reflect the ambiguity and hesitation of experts in the decision-making process. At the same time, the important information contained in the indicator operation data is not fully considered in the weight calculation process, which to some extent affects the accuracy and objectivity of the evaluation results.

[0005] Objective weighting methods primarily determine weights by analyzing the changing trends of an indicator's operational data over time. Common methods include entropy weighting, principal component analysis, and TOPSIS. However, relying solely on objective weighting methods can sometimes lead to irrational weighting. For example, in some cases, certain key indicators should be assigned higher weights, but due to significant fluctuations in their operational data, the final calculated weights are significantly lower. This situation may not truly reflect the indicator's importance and impact on system performance.

[0006] The subjective and objective weighting method can cleverly integrate the subjective opinions of experts with the objective information of indicator data, thereby more comprehensively reflecting the actual situation of the assessment object. This method is currently the most commonly used method for complex system status assessment. However, existing technologies do not consider the degree of conflict, similarity, and uncertainty between the various indicators in the system assessment, and cannot obtain a more accurate health status of thermal power units. Summary of the Invention

[0007] In order to overcome the above problems, the purpose of the present invention is to provide a thermal power unit health status assessment method based on improved weight fusion. This method calculates the correction coefficient of evidence from three key dimensions: degree of conflict, degree of similarity and uncertainty. By correcting the original evidence body, the conflict between evidence is effectively reduced, making the corrected evidence synthesis result more reasonable and reliable, and able to more accurately reflect the health status of the thermal power unit.

[0008] The technical solution adopted in the present invention is:

[0009] The health status assessment method of thermal power units based on improved weight fusion includes the following steps:

[0010] S1: Use the analytic hierarchy process to select indicators according to the system construction principles to establish the indicator system used for evaluation and collect indicator data;

[0011] S2: Determine the status level of the thermal power unit and the corresponding relative degradation numerical classification;

[0012] S3: Determine the weight assigned to each indicator;

[0013] S4: Calculate the membership degree of each indicator corresponding to each state level as the comprehensive evaluation matrix in the fuzzy comprehensive evaluation method;

[0014] S5: Use the fuzzy comprehensive evaluation method to calculate the evaluation sets of different parts in the indicator layer. These evaluation sets are different evidences of the DS evidence method, forming a matrix of basic probability distribution;

[0015] S6: Use synthesis rules to determine whether it is high-conflict evidence. If the conflict is small, synthesize directly. If it exceeds the threshold or even approaches 1, proceed to step 7.

[0016] S7: The evidence theory improved by the SPD-DS method is used to fuse the evidence to obtain the final evaluation opinion of the thermal power unit. The reliability of each of the three evidences is measured from three aspects: the degree of conflict, the degree of similarity and the degree of uncertainty. A very low reliability indicates that there is a great conflict with the remaining evidence and a greater degree of correction is required. Otherwise, basically no correction is required. After correction, the synthesis rule is used to fuse them to obtain the final evaluation result.

[0017] As a further description of the present invention, the specific method of determining the weight assigned to each indicator in S3 is:

[0018] S31: Subjective weights: The subjective weights of the indicators are calculated using the fuzzy analytic hierarchy process, where a fuzzy complementary judgment matrix is constructed based on the experts’ scores on the importance of the indicators.

[0019] S32: The objective weights are calculated using the CRITIC method after order parameter optimization using the collected sample data;

[0020] S33: Using game theory to calculate the optimized combination weights as the importance fuzzy sets in the fuzzy comprehensive evaluation. The optimized combination weights need to be given for different equipment groups.

[0021] As a further description of the present invention, the specific steps of the fuzzy analytic hierarchy process in S31 are:

[0022] S311: Establishing the fuzzy complementary judgment matrix,

[0023] S312: Calculate weights based on the judgment matrix.

[0024] S313: consistency check, verifying whether the weights are reasonable;

[0025] The specific steps of determining the objective weight using the CRITIC method after the order parameter optimization in S32 are as follows:

[0026] S321: Normalize the raw data,

[0027] S322: Calculate the standard deviation,

[0028] S323: Calculate the correlation coefficient and conflict coefficient between two factors to measure the degree of correlation and conflict between the factors.

[0029] S324: Calculate order parameters;

[0030] S325: Calculate the objective weight and obtain the objective weight vector;

[0031] The specific steps of using game theory to calculate the optimized combination weight in S33 are:

[0032] S331: Introducing subjective weights and objective weight Combination function ;

[0033] S332: Seeking the coefficient of subjective weight with the goal of minimizing deviation and the coefficient of objective weight The optimal solution of

[0034] S333: The coefficients in the obtained optimal allocation solution and Perform normalization:

[0035] S334: Obtain the final optimized combination weight.

[0036] As a further description of the present invention, the specific calculation process of S4 is:

[0037] The cloud model with optimized entropy value is used to construct three characteristic values corresponding to each state level: expectation , entropy , super entropy , substitute it into the indicator data to calculate the membership degree corresponding to each state and construct a comprehensive evaluation matrix.

[0038] As a further description of the present invention, the specific steps of the fuzzy comprehensive evaluation method in S5 are:

[0039] set up The set of various indicators that affect the status of thermal power units is the factor set ,set up evaluation statement, which gives the evaluation level of each state of the thermal power unit, namely the evaluation set , the specific steps consist of the following five steps:

[0040] S51: Single factor evaluation,

[0041] S52: Construction of comprehensive evaluation matrix,

[0042] S53: Establish a fuzzy set of indicator importance,

[0043] S54: Obtain fuzzy comprehensive evaluation set,

[0044] S55: Adopt the principle of maximum membership to determine the comprehensive evaluation opinion.

[0045] As a further description of the present invention, the threshold in S6 is 0.7.

[0046] As a further description of the present invention, the specific steps of the SPD-DS method in S7 to improve the evidence theory are:

[0047] S71: Measure the degree of conflict, using the modified Spearman correlation coefficient to measure the conflict of evidence;

[0048] S72: Measure the similarity. The distance between evidences is used to measure the similarity of evidences. The distance between evidences is determined by the support probability function.

[0049] S73: Measure uncertainty. Uncertainty is mainly described by the evidence itself. The degree of conflict and similarity describe the relationship between the evidence. Focus index is used to describe uncertainty.

[0050] S74: Calculate support coefficient;

[0051] S75: Calculate correction coefficient;

[0052] S76: Correction of evidence

[0053] Beneficial effects of the present invention:

[0054] The present invention is based on a thermal power unit health status assessment method based on improved weight fusion. The method improves the fuzzy comprehensive evaluation method based on order optimization weighting and optimized cloud entropy, and applies it to the thermal power unit health status assessment. Independent evaluation is given for indicators in different aspects. In the weight determination link, the CRITIC method is improved by introducing order parameters, so that the objective weight can be adjusted more flexibly, thereby more accurately reflecting the objective information contained in each indicator. At the same time, the fuzzy hierarchical analysis method is used to determine the subjective weight. This method fully combines expert experience and fuzzy mathematics theory, and provides a weight basis with a subjective perspective for the assessment. Afterwards, game theory is used to combine and optimize the subjective weight and the objective weight, realizing the organic integration of subjective and objective information. The results are combined to obtain more scientific and reasonable comprehensive weights. Regarding membership, a cloud model based on nonlinear decision optimization entropy is used. This model cleverly balances the clear classification capabilities of typical samples with the fuzzy processing capabilities of boundary samples. For samples in a typical state, the membership can be determined based on clear and unambiguous classification rules. For samples in a boundary state, the fuzzy characteristics are fully considered to provide a reasonable membership description, thereby more comprehensively and accurately characterizing the state characteristics of various indicators of the thermal power unit. During the comprehensive assessment phase, when synthesizing the SPD-DS evidence theory based on the results obtained from the evaluation of different indicators, high conflict between the evidence may lead to unreliable synthesis results. Therefore, a specially designed SPD-DS method for evidence body correction is designed. This method calculates the evidence correction coefficient based on three key dimensions: conflict degree, similarity degree, and uncertainty. By correcting the original evidence body, the conflict between the evidence is effectively reduced, making the corrected evidence synthesis results more reasonable and reliable, and more accurately reflecting the health status of the thermal power unit. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 This is a flow chart of the thermal power unit health status assessment method based on improved weight fusion proposed by the present invention;

[0056] Figure 2 This is a flowchart of the specific steps of the thermal power unit health status assessment method S3 based on improved weight fusion proposed by the present invention;

[0057] Figure 3 This is a flowchart of the specific steps of the thermal power unit health status assessment method S7 based on improved weight fusion proposed by the present invention. DETAILED DESCRIPTION

[0058] The specific implementation of the present invention is described below with reference to the accompanying drawings and embodiments:

[0059] It should be noted that the structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification for people familiar with this technology to understand and read, and are not used to limit the conditions under which the present invention can be implemented. Any structural modification, change in proportional relationship or adjustment of size should still fall within the scope of the technical content disclosed in the present invention without affecting the efficacy and purpose that can be achieved by the present invention.

[0060] At the same time, the terms such as "upper", "lower", "left", "right", "middle" and "one" quoted in this specification are only for the convenience of description and are not used to limit the scope of implementation of the present invention. Changes or adjustments to their relative relationships should be regarded as the scope of implementation of the present invention without substantially changing the technical content.

[0061] like Figures 1 to 3 As shown, it shows a specific embodiment of the present invention:

[0062] Example 1

[0063] The health status assessment method of thermal power units based on improved weight fusion includes the following steps:

[0064] S1: Use the analytic hierarchy process to select indicators according to the system construction principles to establish the indicator system used for evaluation and collect indicator data;

[0065] S2: Determine the status level of the thermal power unit and the corresponding relative degradation numerical classification;

[0066] S3: Determine the weight assigned to each indicator;

[0067] S4: Calculate the membership degree of each indicator corresponding to each state level as the comprehensive evaluation matrix in the fuzzy comprehensive evaluation method;

[0068] S5: Use the fuzzy comprehensive evaluation method to calculate the evaluation sets of different parts in the indicator layer. These evaluation sets are different evidences of the DS evidence method, forming a matrix of basic probability distribution;

[0069] S6: Use synthesis rules to determine whether it is high-conflict evidence. If the conflict is small, synthesize directly. If it exceeds the threshold or even approaches 1, proceed to step 7.

[0070] S7: The evidence theory improved by the SPD-DS method is used to fuse the evidence to obtain the final evaluation opinion of the thermal power unit. The reliability of each of the three evidences is measured from three aspects: the degree of conflict, the degree of similarity and the degree of uncertainty. A very low reliability indicates that there is a great conflict with the remaining evidence and a greater degree of correction is required. Otherwise, basically no correction is required. After correction, the synthesis rule is used to fuse them to obtain the final evaluation result.

[0071] In this embodiment, Figure 1 As shown in the figure, this method improves the fuzzy comprehensive evaluation method based on order optimization weighting and optimized cloud entropy, and applies it to the health status assessment of thermal power units. Independent evaluations are given for indicators in different aspects. In the weight determination link, the CRITIC method is improved by introducing order parameters, so that it can adjust the objective weights more flexibly, thereby more accurately reflecting the objective information contained in each indicator. At the same time, the fuzzy hierarchical analysis method is used to determine the subjective weights. This method fully combines expert experience and fuzzy mathematics theory, and provides a weight basis with a subjective perspective for the evaluation. Afterwards, game theory is used to combine and optimize the subjective weights with the objective weights, realizing the organic integration of subjective and objective information and obtaining a more scientific and reasonable comprehensive weight. In terms of membership acquisition, with the help of the cloud model of nonlinear decision optimization entropy value, this model cleverly balances the clear classification ability of typical samples and The relationship between the fuzzy processing capabilities of boundary samples is as follows: for samples in a typical state, their membership can be determined based on clear and definite classification rules; for samples in a boundary state, their fuzzy characteristics are fully considered and a reasonable membership description is given, thereby more comprehensively and accurately characterizing the state characteristics of each indicator of the thermal power unit. In the comprehensive evaluation stage, when the SPD-DS evidence theory is synthesized based on the results obtained from the evaluation of indicators from different aspects, the synthesis results may be unreliable due to high conflicts between the evidence. A specially designed SPD-DS method for evidence body correction is designed. This method calculates the correction coefficient of evidence based on three key dimensions: conflict degree, similarity degree and uncertainty. By correcting the original evidence body, the conflict between the evidence is effectively reduced, making the corrected evidence synthesis results more reasonable and reliable, and able to more accurately reflect the health status of the thermal power unit.

[0072] Example 2

[0073] In this embodiment, steps S1 and S2 fully collect and analyze thermal power unit design data and historical operating data, selecting key status parameters to construct a thermal power unit status evaluation index system. The specific architecture is shown in Table 1. The thermal power unit status evaluation system adopts a layered architecture consisting of a target layer, a project layer, and an indicator layer. The clear and logically coherent layers facilitate a systematic assessment of the health status of thermal power units.

[0074] Table 1 Thermal power unit health status evaluation index system

[0075]

[0076]

[0077]

[0078] In this embodiment, the status evaluation results of the thermal power unit are divided into four levels, namely {normal, concern, abnormal, critical}, and the corresponding comment sets are represented as .

[0079] in, "Normal" means the thermal power unit is operating well, all performance indicators are within normal range, and it can generate electricity stably and efficiently. "Attention" means that some indicators of the thermal power unit have experienced slight fluctuations. Although this has not yet had a significant impact on the unit's operation, it is necessary to closely monitor its changing trends to prevent potential problems from escalating. "Abnormal" means that the thermal power unit has obvious operating abnormalities, which may affect the stability and power generation efficiency of the unit and need to be promptly checked and handled with appropriate measures; It stands for "critical", indicating that the thermal power unit has a serious fault or major safety hazard, which has posed a direct threat to the safe operation of the unit and requires immediate shutdown and maintenance to avoid more serious consequences.

[0080] In this embodiment, the different indicator data have different dimensions, making it impossible to directly analyze the degradation degree of the indicators from the monitoring data. To facilitate evaluation and analysis, dimensionless processing is required to convert the indicator data values into relative degradation degrees. This embodiment uses the Min-Max normalization method to ultimately normalize the data to [0, 1]. The closer the relative degradation degree is to 1, the better the corresponding equipment is in condition, while the closer it is to 0, the more serious the problem. The relationship between the various status categories of thermal power unit equipment and the relative degradation degree is shown in Table 2.

[0081] Table 2 Correspondence between status categories and relative degradation degrees

[0082]

[0083] For some indicators, the larger the value, the better the status of the thermal power unit, which is a positive indicator. For other indicators, the smaller the value, the better the status of the thermal power unit, which is a negative indicator. The positive indicator processing method is as follows:

[0084] ,

[0085] The reverse indicator processing method is as follows:

[0086] ,

[0087] in Refers to each indicator quantity, is the minimum value of the indicator for a reasonable threshold, It is the maximum value of the threshold corresponding to the indicator.

[0088] Example 3

[0089] Specifically, the specific method of determining the weight assigned to each indicator in S3 is:

[0090] S31: Subjective weight The subjective weight of the indicator is calculated according to the fuzzy analytic hierarchy process, in which a fuzzy complementary judgment matrix is constructed according to the experts' scores on the importance of the indicator.

[0091] In this embodiment, the specific steps of the fuzzy analytic hierarchy process are:

[0092] S311: Establish a fuzzy complementary judgment matrix.

[0093] The establishment of the matrix requires the analysis of the influencing factors. Each two factors are divided into a group to compare which one is more important. The relative values of their importance are used to form the matrix. The judgment matrix of the influencing factors needs to be constructed , each value in the judgment matrix represent Compare the importance of.

[0094] The basis for analysis and judgment is usually derived from the importance scores of various factors assigned by multiple experts. The standard measurement of relative importance is shown in Table 3.

[0095] Table 3 Relative importance of influencing factors Measurement

[0096]

[0097] The conditions for satisfying the fuzzy complementary judgment matrix are:

[0098] ,

[0099] in and , so once Compare More importantly, Between 0.6 and 0.9, Between 0.4 and 0.1, it means Compare gradually increases in unimportance.

[0100] According to the above measurement method The factors are compared with each other, and the matrix formed is as follows:

[0101] ,

[0102] S312: Calculate weights according to the judgment matrix.

[0103] From the judgment matrix Solution The subjective weight of each factor .

[0104] ,

[0105] in .

[0106] S313: Consistency check.

[0107] The purpose of checking consistency is to verify whether the weights are reasonable. The consistency is verified by the compatibility of the judgment matrix. If the obtained consistency parameter is too large, it means that the weights are unreasonable and cannot be used for subsequent decision-making. It is necessary to go back to the first step to adjust the judgment matrix. The two fuzzy judgment matrices and Compatibility index The calculation formula is:

[0108] ,

[0109] Assume that the fuzzy judgment matrix The weight vector is obtained , meeting the conditions and ,set up ,in , then the judgment matrix The characteristic matrix of is:

[0110] ,

[0111] Given parameters Represent the decision-making party's opinion and meet the conditions Explain the fuzzy judgment matrix When the decision maker strictly requires the consistency test of the judgment matrix, the consistency will be reduced as much as possible. In most cases, scholars .

[0112] In actual problems, more than one expert's opinion is collected. Each expert's score will form a fuzzy judgment matrix, and the weight vector can be calculated according to the formula to complete the consistency test. If the judgment matrix obtained from each expert's experience meets the consistency test and the compatibility index of the two judgment matrices in this series of judgment matrices also meets the consistency test, the final subjective weight vector Each It can be derived from the following formula:

[0113] ,

[0114] in Represents the number of experts, which is also the number of judgment matrices and corresponding weight matrices. Representative Experts give their opinions on the The weight of the factors, .

[0115] S32: The objective weights are calculated using the CRITIC method after order parameter optimization using the collected sample data.

[0116] In this embodiment, the specific steps of determining the objective weight using the CRITIC method after optimizing the order parameter are as follows:

[0117] S321: Normalize the original data, set samples, each containing The data corresponding to each factor are all in the unified sample value Within the range, the sample matrix is formed after processing As shown below:

[0118] ;

[0119] S322: Calculate the standard deviation. Before calculating the standard deviation, you need to calculate the arithmetic mean of each factor, as shown in the following formula:

[0120] ,

[0121] in , Indicates the The arithmetic mean of the factors is The standard deviation of the factor The calculation is as follows:

[0122] ;

[0123] S323: Calculate the correlation coefficient and conflict coefficient between the two factors to measure the degree of correlation and conflict between the factors. The calculation formula of the correlation coefficient is as follows:

[0124] ,

[0125] ,

[0126] in , represent and The covariance of represents the number of samples, Representative and Correlation coefficient of factors;

[0127] The conflict coefficient correlation coefficient is obtained as shown in the following formula:

[0128] ,

[0129] The weight determination method is indicators, It is and The correlation coefficient of the factors, is the normalized conflict coefficient, ;

[0130] Since the correlation coefficient and conflict coefficient are determined by the degree of association between the indicator data, even if the traditional CRTIC method has normalized the original indicator data, normalization in this step is still necessary. The normalized conflict coefficient changes the numerical scale of the original correlation coefficient, which can avoid the disadvantage of excessively large or small conflict values affecting the final objective weight calculation. When an indicator is highly correlated with other indicators, it means that the information redundancy of the indicator is large, and the conflict coefficient It is relatively small and has low independence in the evaluation index system.

[0131] S324: Calculate order parameters and conflict coefficients Re-sort by value to get an ordered sequence ,in Smaller ones are ranked in the front, and the order parameters are calculated as follows:

[0132] ,

[0133] ,

[0134] in is the coefficient of variation, which affects the order parameter , Represents sorting information, the top ones right and The degree of influence will be relatively small. When the index conflict coefficient is large, it will have a greater impact on the order parameter at the end of the ordered sequence. If the difference in the conflict coefficients of various indicators is not large after sorting, it means that the correlation between the indicators is relatively similar.

[0135] S325: Calculate objective weight , as shown below:

[0136] ,

[0137] in , It is The standard deviation of an indicator represents the amount of information conveyed by the indicator. Since the data samples have been normalized in traditional methods, the standard deviation does not need to be normalized again. It can be seen that As an order parameter, it is crucial to adjust the final objective weight of the indicator. When the information conveyed by the indicator ranking is relatively strong and the order parameter is large, the conflict coefficient plays a more important role in the allocation of objective weights. When the indicator ranking is more uniform, the ranking provides less information and the order parameter is small, and the standard deviation of the indicator is more important when allocating objective weights.

[0138] When the conflict coefficient Relatively large, The individual indicators are relatively independent and have little correlation with other indicators. The greater the contribution, the larger the standard deviation of the indicator is. When more information is conveyed, the final weight distribution will be higher, which is in line with the idea of weight distribution of the CRITIC method.

[0139] After obtaining the objective weights, when using game theory to calculate the combined weights, there may sometimes be a mismatch between the scales of the subjective and objective weights, which can affect the solution of the combined coefficient. In particular, a weight with a very large data distribution span will completely dominate the combined weight, and even cause the coefficient of another weight to be negative. The solution formula for the game theory combined coefficient is converted to another solution form, as shown below:

[0140] ,

[0141] If the subjective weight and objective weight The difference in the distribution of the two sequence data is too great, which will lead to the coefficient matrix The condition number is large, making the coefficient matrix The closer it is to a singular matrix, the less reasonable the weight coefficients are. In this case, the matrix stability can be enhanced by adding regularization parameters to adjust the coefficient matrix, as shown in the following formula:

[0142] ,

[0143] in is the regularization parameter, which is usually , is the unit matrix, and the addition of regularization parameters ensures that the weight coefficients are all within Within a reasonable range, one of the subjective and objective weights can be prevented from being ignored in the combined weight due to mismatch in numerical scales.

[0144] S33: Using game theory to calculate the optimized combination weights as the importance fuzzy sets in the fuzzy comprehensive evaluation. The optimized combination weights need to be given for different equipment groups.

[0145] The specific steps of using game theory to calculate the optimized combination weights are:

[0146] S331: Introducing subjective weights and objective weight The combination function is shown below:

[0147] .

[0148] S332: Seeking the coefficient of subjective weight with the goal of minimizing deviation and the coefficient of objective weight The optimal solution is to adjust the coefficients of both parties, that is, to adjust the proportion of subjective and objective weights. According to the differential formula theory, it is converted into the solution of the equation system, as shown in the following formula:

[0149] .

[0150] S333: The coefficients in the obtained optimal allocation solution and Perform normalization:

[0151] ,

[0152] .

[0153] S334: Get the final optimized combination weight:

[0154] .

[0155] Example 4

[0156] Specifically, the specific calculation process of S4 is:

[0157] The cloud model with optimized entropy value is used to construct three characteristic values corresponding to each state level: expectation , entropy , super entropy , substitute it into the indicator data to calculate the membership degree corresponding to each state and construct a comprehensive evaluation matrix.

[0158] In this embodiment, the concept used in cloud theory to characterize the conversion from quantitative data to qualitative expression is membership, and the domain is set ,in is a set, and there are fuzzy sets in the domain , each right There is a corresponding stable tendency, using membership The size of the value describes the degree of tendency. It is essentially a random number, its range is .

[0159] In the cloud model, each cloud droplet represents a piece of data. The specific distribution of cloud droplets is determined by the three eigenvalues that define the cloud model: expectation, entropy, and superentropy. When the expectation and superentropy are equal, a higher entropy value indicates a wider distribution of cloud droplets and greater fuzziness. When the expectation and entropy are equal, a higher superentropy value indicates a more dispersed distribution of cloud droplets and a thicker cloud. These three eigenvalues combine fuzziness and randomness to complete the mapping process from quantitative to qualitative.

[0160] The two core components of the cloud model are the forward cloud generator and the reverse cloud generator, which can be implemented in software or hardware, completing the transition between qualitative and quantitative analysis. The forward cloud generator uses the three existing eigenvalues to generate corresponding cloud droplets, simulate the distribution of cloud droplets in different states, visualize the distribution characteristics, and ultimately determine the membership degree of each cloud droplet to each state.

[0161] The normal cloud model is formed by generating cloud droplets through a forward cloud generator, which has the characteristic of asymmetric thickness. The random number generation function is used to obtain the coordinate distribution and membership of the cloud droplets. Specifically,

[0162] ,

[0163] ,

[0164] ,

[0165] in Indicates is the expected value, is a normal random number with variance; Indicates For expectations, is a normal random number with variance; It is the membership degree of the cloud droplet to a certain state level. It can be seen that the generation of each cloud droplet has a certain randomness, and the randomness of the cloud model composed of cloud droplets is reflected. As the number of cloud droplets increases to a relatively large number, the characteristics of the entire cloud become relatively stable. At this time, the slight change of a single cloud droplet has no effect on the overall cloud. Therefore, the cloud model also has a certain degree of certainty. It is meaningless to discuss a certain cloud droplet alone. The cloud model focuses more on the characteristics of the cloud as a whole.

[0166] The reverse cloud generator derives the expectation, entropy and super entropy from known data samples based on statistical theory. In this way, in actual research, the distribution law standards that this type of data should have under various states can be established through the various state labels corresponding to historical data, avoiding the traditional method's reliance on expert experience and can be used for feature extraction.

[0167] Obtain the expected value in the cloud model based on the cloud droplet distribution coordinates , entropy and super entropy The process is as follows:

[0168] ,

[0169] ,

[0170] ,

[0171] in, Known In the sample indivual, , is the first-order sample absolute central moment, is the sample variance. When the sampled cloud droplet amount is relatively small, the error of the obtained eigenvalue will be large. At this time, the least squares method can be used to obtain higher accuracy, but the process is more complicated. As the number of cloud droplets gradually increases, the accuracy of the eigenvalue will increase accordingly.

[0172] In this embodiment, the cloud model with optimized entropy value is used to set the processing index data Corresponding to the membership of each state, the state is divided into Level, then the status The corresponding cloud model parameters are , . The entropy value of the rule is , the entropy value calculated by the 50% association rule is . Introducing the state The maximum deviation value of membership Calculated as:

[0173] ,

[0174] in Represents the membership status calculated using the 50% association rule The maximum membership degree of For status Adopted The rule calculates the minimum membership value, Is the state calculated by the optimized entropy value Membership value. and The average value is obtained after multiple calculations to avoid the error caused by the randomness added in a single calculation. The entropy value of the normal cloud model is continuously adjusted to meet the goal of the decision model, that is, to make The sum of the maximum deviation values of the membership of each level state is minimized to optimize the entropy value.

[0175] The 50% correlation rule is used to calculate the maximum membership value in the construction of the decision objective function. This takes into account the fact that the rule focuses more on the correlation degree of the state boundary. It is more reasonable to measure the maximum membership value by comprehensively considering the adjacent states, thereby improving the trustworthiness of the state membership value after optimization at the boundary. The rule calculation of the minimum membership takes into account the fact that the rules strictly define the entropy range. Measuring the minimum membership is more suitable, thereby appropriately reducing the overlap between states and making the distinction between states more obvious. Combining the two considerations can make the state boundary more scientific and the membership obtained more accurate. After optimization, the membership sequence of this indicator corresponding to all states is .

[0176] After entropy optimization, new cloud model eigenvalues of different states are obtained. According to the above formula , , A cloud map can be generated, where the horizontal axis of the cloud map is the relative degradation value obtained after preprocessing the sample data. The calculation formula will use Gaussian random perturbations to adjust the entropy value , While there is randomness, the uncertainty of the cloud model is also added. Each cloud droplet is randomly perturbed when it is calculated:

[0177] ,

[0178] All cloud droplets at each state level are sorted from small to large according to their relative degradation. For each state level in the cloud map, the relative degradation degree obtained after preprocessing requires selecting five surrounding cloud droplets and taking the average value as the membership degree for each state. For indicator data with a clear tendency towards state level, the membership calculated by the cloud model with optimized entropy can be centrally assigned to the corresponding state level, avoiding ambiguity in the distribution of membership across several states and ensuring clear classification of such indicator data. For indicator data with overlapping state characteristics, the optimized cloud model fully preserves the sample's membership to each state level, avoiding crude classification into a single state. It can reflect the transition and uncertainty of such indicators in different states, ensuring a more comprehensive and reasonable membership degree.

[0179] Example 4

[0180] Specifically, the specific steps of the fuzzy comprehensive evaluation method in S5 are:

[0181] set up The set of various indicators that affect the status of thermal power units is the factor set ,set up evaluation statement, which gives the evaluation level of each state of the thermal power unit, namely the evaluation set , the specific steps consist of the following five steps:

[0182] S51: Single factor evaluation, for each factor in the factor set All need to be evaluated, and the impact of this factor on each evaluation in the evaluation set should be clarified. Membership , we can know the single factor Evaluation set , single factor evaluation set It is based on the evaluation set fuzzy sets on ;

[0183] S52: Construction of comprehensive evaluation matrix: The single factor evaluation set is constructed into a comprehensive evaluation matrix by row, as shown below:

[0184] ,

[0185] in and ;

[0186] S53: Establishing the fuzzy set of indicator importance: From the factor set Constructing fuzzy sets to represent the importance of each factor considered when making the final decision, that is, weight, fuzzy set ,in It is The weight of the factors, ;

[0187] S54: Obtain fuzzy comprehensive evaluation set: Select appropriate fuzzy operation according to actual situation, and use Represented by the weight set and comprehensive evaluation matrix The fuzzy comprehensive evaluation set obtained by the operation is as follows:

[0188] ,

[0189] The fuzzy comprehensive evaluation set and evaluation set One-to-one correspondence, you can choose the commonly used There are six operation modes: , , , , , , The calculation is shown in formula 1. The calculation is shown in formula 2. The calculation is shown in formula 3. The calculation is shown in formula 4.

[0190] , Formula 1

[0191] , Equation 2

[0192] , Equation 3

[0193] , Equation 4

[0194] in , It is The weight of each factor is the most commonly used in current research. .

[0195] S55: Determine comprehensive evaluation opinions: Adopt the principle of maximum membership, Select the maximum value , corresponding to the evaluation set The evaluation statement is the final comprehensive evaluation opinion.

[0196] The above process describes a first-level fuzzy comprehensive evaluation. In more scenarios, influencing factors will be incorporated into the Analytic Hierarchy Process (AHP), dividing factors into layers, with multiple factors at the next level determining those at the previous level. Multi-level fuzzy comprehensive evaluation is suitable for this situation, addressing the problem of first-level fuzzy comprehensive evaluation failing to produce accurate results. The specific approach is to conduct a comprehensive evaluation from the lowest level of factors to the highest level.

[0197] Example 5

[0198] Specifically, the threshold in S6 is 0.7.

[0199] Example 6

[0200] Specifically, the specific steps of improving the evidence theory by the SPD-DS method in S7 are:

[0201] S71: Measure the degree of conflict and use the modified Spearman correlation coefficient to measure the conflict of evidence.

[0202] Compared with the traditional method of calculating the degree of conflict, the Spearman correlation coefficient reduces the impact of outliers on the synthesis effect, and avoids the limitation of the Pearson coefficient that can only process normally distributed samples. The original Spearman correlation coefficient requires that the sample cannot have repeated data, and the correlation coefficient itself does not distinguish and 0. But the two are very different in terms of evidence. The coefficient is 0, which means there is no correlation between the two pieces of evidence. and Each focal elements, and the sequences are and , sort the sequences in order of size to get the rank and ,For example After sorting, the fifth smallest data is in the rank middle , Similarly, sorting from small to large or from large to small does not affect the final coefficient calculation results. The Spearman correlation coefficient focuses on keeping the relative ranks consistent.

[0203] evidence and The original Spearman correlation coefficient can be expressed as , which is equivalent to calculating the Pearson coefficient of two ranks. The specific calculation is as follows:

[0204] ,

[0205] in and They are and In the evidence and The size ranking value in the sequence, and Is their respective rank and The mean of the series, .at this time The closer it is to 1, the less conflict there is between the two pieces of evidence. A value of 1 indicates that the order of the two pieces of evidence ranks is completely consistent, both monotonically increasing. A value of -1 indicates that the ranks are completely opposite, with one piece of evidence monotonically increasing while the other monotonically decreasing.

[0206] When the basic probability distribution function, denoted as BPAF, cannot be satisfied by the original Spearman correlation coefficient, an adjustment is made, and 1 is added to the above formula to make , in order to ensure that the final value is satisfied Between, then , the adjusted calculation is as follows:

[0207] ,

[0208] in ,so The improved correlation coefficient can be used to and 0, and meet the value range of BPAF. According to the default condition that the Spearman correlation coefficient cannot have repeated samples, since there is a high possibility that two or more focal elements in one evidence have the same BPAF, the rank average is used to deal with repeated values. For example, evidence Specifically , the original rank is , the last three ranks are updated to .evidence The degree of conflict with other evidence is:

[0209] ,

[0210] in Indicates the total amount of evidence, , The larger the value, the less conflict there is with other evidence. So far, the improved Spearman correlation coefficient can be used to represent the degree of conflict between evidence.

[0211] S72: Measure the degree of similarity. The distance between evidence is used to measure the degree of similarity. The distance between evidence is determined by the support probability function. The SPFE function defines the degree of support for the focal element in the evidence by other focal elements, as shown in the following formula:

[0212] ,

[0213] in It is the identification framework, Jiao Yuan , express The number of elements contained in the collection, Similarly, Overall representative of the focal element and Common parts Support level, last passed Normalization processing can simplify the calculation for a single-point focal element, as shown in the following formula:

[0214] ,

[0215] If multiple BPAFs contain single-point focal elements in larger focal elements , then the SPFE calculated from the single-point focal element will also be relatively large, proving that the evidence where the focal element is located strongly supports the single-point focal element, and thus the evidence is extended. and The SPFE function between is shown as follows:

[0216]

[0217] in , first calculate the evidence by formula and The SPFE value of each focal element in the , and then select the largest absolute difference as the SPFE value of the evidence body, when the total number of evidence is Proof in the case The calculation of the degree of similarity with other evidence is as follows:

[0218] ,

[0219] in , The closer it is to 1, the more similar it is to other evidence.

[0220] S73: Measure uncertainty. Uncertainty mainly describes the evidence itself. The degree of conflict and similarity describe the relationship between the evidence. The focus index can reasonably describe uncertainty. The specific calculation is shown in the following formula:

[0221] ,

[0222] in , which is the identification framework The number of focal elements, It's evidence Focus Element BPAF value, is the BPAF value of each focal element when all focal elements are evenly distributed, If the focus value Small, explain the evidence Supporting many focal elements means a greater degree of uncertainty.

[0223] Considering the above three perspectives, The larger it is, the less conflict there is with other evidence. The larger it is, the more similar it is to other evidence. The bigger the evidence The smaller the uncertainty itself, the more reliable the evidence is, so the support coefficient is defined as Greater weight is given to evidence with higher reliability. The support coefficient is calculated as follows:

[0224] ,

[0225] in , Adding 1 to the denominator can ensure that the calculated correction factor is at a stable level. 、 、 Both are very small. Not adding the denominator may cause the subsequent normalization calculation to be too large in scaling, resulting in a large error in the calculation. A value that is too large will affect the relative ratio of the original numerator to 1, for example , we can find that if the numerator drops too much, it will affect the accuracy of the correction. If the denominator is too small, it is equivalent to 1 and cannot stabilize extreme numbers. The coefficient used to correct the evidence body is calculated as follows:

[0226] ,

[0227] in , is the minimum support coefficient among all evidences. It is the maximum value among all the evidence support coefficients. The correction coefficient is calculated using the minimum-maximum normalization method. In the above formula, 0.1 gives extremely low-reliability evidence a very small weight to ensure that it is not discarded during evidence synthesis. The 0.9 adjusts the normalized value so that the evidence with extremely high credibility basically maintains the original basic distribution probability and is almost not corrected. It also controls the correction coefficient to not exceed 1. The final evidence correction is shown in the following formula:

[0228] ,

[0229] in It's evidence Focus Element BPAF value, It is the corrected evidence For the first focal element BPAF value, is the BPAF value adjusted for the uncertainty of the evidence.

[0230] The preferred embodiments of the present invention are described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Various changes can be made within the knowledge of ordinary technicians in this field without departing from the scope of the present invention.

[0231] Many other changes and modifications can be made without departing from the spirit and scope of the present invention. It should be understood that the present invention is not limited to the specific embodiments, and the scope of the present invention is defined by the appended claims.

Claims

1. A thermal power unit health status assessment method based on improved weight fusion, characterized in that: The following steps are involved: S1: Use the analytic hierarchy process to select indicators according to the system construction principles to establish the indicator system used for evaluation and collect indicator data; S2: Determine the status level of the thermal power unit and the corresponding relative degradation numerical classification; S3: Determine the weight assigned to each indicator; S4: Calculate the membership degree of each indicator corresponding to each state level as the comprehensive evaluation matrix in the fuzzy comprehensive evaluation method; S5: Use the fuzzy comprehensive evaluation method to calculate the evaluation sets of different parts in the indicator layer. These evaluation sets are different evidences of the DS evidence method, forming a matrix of basic probability distribution; S6: Use synthesis rules to determine whether it is high-conflict evidence. If the conflict is small, synthesize directly. If it exceeds the threshold or even approaches 1, proceed to step 7. S7: The evidence theory improved by the SPD-DS method is used to fuse the evidence to obtain the final evaluation opinion of the thermal power unit. The reliability of each of the three evidences is measured from three aspects: the degree of conflict, the degree of similarity and the degree of uncertainty. A very low reliability indicates that there is a great conflict with the remaining evidence and a greater degree of correction is required. Otherwise, basically no correction is required. After correction, the synthesis rule is used to fuse them to obtain the final evaluation result.

2. The method for health status assessment of thermal power units based on improved weight fusion according to claim 1, characterized in that: The specific method of S3 to determine the weight assigned to each indicator is: S31: Subjective weights: The subjective weights of the indicators are calculated using the fuzzy analytic hierarchy process, where a fuzzy complementary judgment matrix is constructed based on the experts’ scores on the importance of the indicators. S32: The objective weights are calculated using the CRITIC method after order parameter optimization using the collected sample data; S33: Using game theory to calculate the optimized combination weights as the importance fuzzy sets in the fuzzy comprehensive evaluation. The optimized combination weights need to be given for different equipment groups.

3. The thermal power unit health status assessment method based on improved weight fusion according to claim 2 is characterized in that: The specific steps of the fuzzy analytic hierarchy process in S31 are: S311: Establishing the fuzzy complementary judgment matrix, S312: Calculate weights based on the judgment matrix. S313: consistency check, verifying whether the weights are reasonable; The specific steps of determining the objective weight using the CRITIC method after the order parameter optimization in S32 are as follows: S321: Normalize the raw data, S322: Calculate the standard deviation, S323: Calculate the correlation coefficient and conflict coefficient between two factors to measure the degree of correlation and conflict between the factors. S324: Calculate order parameters; S325: Calculate the objective weight and obtain the objective weight vector; The specific steps of using game theory to calculate the optimized combination weight in S33 are: S331: Introducing subjective weights and objective weight Combination function ; S332: Seeking the coefficient of subjective weight with the goal of minimizing deviation and the coefficient of objective weight The optimal solution of S333: The coefficients in the obtained optimal allocation solution and Perform normalization: S334: Obtain the final optimized combination weight.

4. The method for health status assessment of thermal power units based on improved weight fusion according to claim 1, characterized in that: The specific calculation process of S4 is: The cloud model with optimized entropy value is used to construct three characteristic values corresponding to each state level: expectation , entropy , super entropy , substitute it into the indicator data to calculate the membership degree corresponding to each state and construct a comprehensive evaluation matrix.

5. The thermal power unit health status assessment method based on improved weight fusion according to claim 1 is characterized in that: The specific steps of the fuzzy comprehensive evaluation method in S5 are: set up The set of various indicators that affect the status of thermal power units is the factor set ,set up evaluation statement, which gives the evaluation level of each state of the thermal power unit, namely the evaluation set , the specific steps consist of the following five steps: S51: Single factor evaluation, S52: Construction of comprehensive evaluation matrix, S53: Establish a fuzzy set of indicator importance, S54: Obtain fuzzy comprehensive evaluation set, S55: Adopt the principle of maximum membership to determine the comprehensive evaluation opinion.

6. The method for health status assessment of thermal power units based on improved weight fusion according to claim 2, characterized in that: The threshold in S6 is 0.

7.

7. The thermal power unit health status assessment method based on improved weight fusion according to claim 1 is characterized in that: The specific steps of the SPD-DS method in S7 to improve the evidence theory are: S71: Measure the degree of conflict, using the modified Spearman correlation coefficient to measure the conflict of evidence; S72: Measure the similarity. The distance between evidences is used to measure the similarity of evidences. The distance between evidences is determined by the support probability function. S73: Measure uncertainty. Uncertainty is mainly described by the evidence itself. The degree of conflict and similarity describe the relationship between the evidence. Focus index is used to describe uncertainty. S74: Calculate support coefficient; S75: Calculate correction coefficient; S76: Correction of evidence.

Images