Inplanatable double-optimization electromechanical equipment safety state evaluation method

Abstract

The invention discloses an interpretable double-optimization electromechanical equipment safety state evaluation method, which belongs to the technical field of detection, and comprises the following steps: 1, obtaining a data set of electromechanical equipment; 2, constructing a basic BRB model; step 3, carrying out primary optimization on the confidence coefficient; 4, providing four interpretable constraint strategies; 5, further optimizing by using an evolutionary strategy; step 6, acquiring a safety state evaluation result of the electromechanical equipment; according to the method, the confidence coefficient of the basic BRB model is subjected to double optimization, four interpretable constraint strategies are provided in the optimization process, meanwhile, the evidence reasoning analysis method adopted by the method not only considers subjective knowledge of expert knowledge for evaluation, but also fully considers objective control of engineering data, so that the method has the advantages of being high in reliability and high in reliability. Therefore, it is ensured that the evaluation result is more reliable, and the transparency, rationality and integrity of the evaluation process are improved on the premise that the accuracy of the characteristic indexes is ensured.

Description

Technical Field

[0001] The present invention belongs to the field of detection technology, and in particular relates to an interpretable dual-optimization electromechanical equipment safety status assessment method. Background Art

[0002] Electromechanical equipment is a type of equipment used in industrial and agricultural production. It combines mechanical and electrical systems to perform various tasks. With the development of technology, modern electromechanical equipment increasingly adopts microelectronics technology, computer technology and Internet of Things technology to improve its intelligence level and operating efficiency. Ensuring the safe state of electromechanical equipment is essential to prevent accidents, improve production efficiency, extend equipment life and reduce maintenance costs. Electromechanical equipment systems are composed of multiple components and subsystems, such as motors, control devices, sensors, etc., which work together to achieve specific functional goals. In view of the complexity and variability of the tasks undertaken by electromechanical equipment, the evaluation of its safety status should not only focus on the performance monitoring of each individual component, but also conduct a comprehensive evaluation from the level of the entire system. In addition, given that electromechanical equipment tasks are often accompanied by high risks, early detection and proper handling of potential failure problems are extremely important to ensure the smooth progress of tasks.

[0003] In existing BRB research, experts often provide parameters such as initial confidence. Although expert knowledge can provide broad and scientific guidance in one case, the applicability of such guidance to more specific data sets may be limited. In addition, randomized global optimization methods may undermine the interpretability of expert knowledge. Therefore, in order to overcome these challenges, a safety status assessment method for electromechanical equipment with an interpretable dual-optimization confidence rule base is urgently needed. Summary of the invention

[0004] The purpose of the present invention is to propose an explainable dual-optimization safety status assessment method for electromechanical equipment in order to solve the problems raised in the above background technology.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] An interpretable dual-optimization electromechanical equipment safety status assessment method includes the following steps:

[0007] Step 1: Obtain the data set of electromechanical equipment. Through the sensors installed on the equipment, the equipment parameters are monitored and collected in real time to obtain the electromechanical equipment data set.

[0008] Step 2: Construct a basic BRB model, determine the reference values ​​of the premise attributes and the reference values ​​corresponding to the results, and use expert knowledge and input data to create a basic BRB model;

[0009] Step 3: Perform initial optimization on the confidence. For the model obtained in step 2, a Gaussian membership function confidence optimization method based on Bayesian updating (GMF-B) is proposed to perform initial optimization on the confidence.

[0010] Step 4: Propose four interpretability constraint strategies to prevent the interpretability of expert knowledge from being destroyed during the optimization process;

[0011] Step 5: Use evolutionary strategy to further optimize and use the projection covariance matrix adaptive algorithm P-CMA-ES to further optimize the parameters in the model to further improve the model accuracy;

[0012] Step 6: Obtain the safety status assessment results of the electromechanical equipment. Apply the evidence reasoning and parsing algorithm ER to infer the model after parameter tuning in step 5 to obtain the safety status assessment results of the electromechanical equipment.

[0013] As a further description of the above technical solution: Step three is specifically: combining the reference value and the prior attribute of the data set, and then obtaining a more reasonable confidence distribution through Gaussian membership function and Bayesian update.

[0014] As a further description of the above technical solution: In step 3, the method is as follows:

[0015] 3.1 The reference values ​​of the attribute and expert knowledge results are A and H, respectively, which are used to calculate the subsequent membership. Then, according to the initial confidence defined by the expert, the prior probability distribution matrix is ​​generated, denoted as P;

[0016] β r =g(A,β e ,x) (1)

[0017] Where A represents the reference value of the previous attribute, β e represents the initial reference value determined by expert knowledge, x represents the input data, β r represents a more reasonable confidence after optimization, and g(·) represents the function of the GMF-B method;

[0018] 3.2 According to the calculation formula of Gaussian membership function, the membership of the previous attribute reference value is calculated. The calculation formula of Gaussian membership function is as follows:

[0019]

[0020] According to the calculation formula of Gaussian membership function, the membership of the reference value of the previous attribute result is calculated. The calculation formula of Gaussian membership function is as follows:

[0021]

[0022] 3.3 By multiplying the membership of the reference value of the previous attribute in the current rule with the membership of the result, the comprehensive membership corresponding to the result can be obtained, and the formula is as follows:

[0023]

[0024] in Indicates the comprehensive membership corresponding to the i-th result in the current rule, φ 1 A ,φ 2 A ,…,φ M A represents the membership degree of M prior attribute reference values, φ i H represents the membership degree of the i-th result;

[0025] 3.4 The above steps calculate the comprehensive membership of the results based on the input data and reference values. This membership can be used as the influence of data on confidence. Therefore, the initial expert knowledge is used as the prior probability distribution, and the above membership is used as new evidence for the Bayesian method to update the prior probability. This process can be expressed as follows:

[0026]

[0027] P(C i |H)=P(H|C i )(C i ) (6)

[0028] Where P(H|C i ) is the likelihood function expressed as the sum of comprehensive memberships, P(C i ) is the prior probability of the current combination, P(C i |H) is the posterior probability of the current combination;

[0029] 3.5 Repeat the first four steps, calculate the posterior probability of each set of input data and each combination, and normalize them to obtain the final posterior possibility distribution matrix, which is the confidence distribution of expert knowledge optimized by the data distribution.

[0030] As a further description of the above technical solution: In the step 4, specifically: four interpretability constraint strategies are proposed to constrain the second optimization process of the model. The following is a detailed introduction to these four strategies:

[0031] 4.1 Strategy 1: Ensure that activated rules participate in optimization and reasoning:

[0032] Ω={θ 1 ,...,θ k ,δ 1 ,...,δM ,β 1,k ,...,β N,k} (7)

[0033] The activation weight can be used to identify inactive rules. The activation of the kth rule can be expressed as follows:

[0034] Q k ={ω 1 ,ω 2 ,...,ω E},k=1,2,...,L (8)

[0035] Where E represents the number of input data, ω 1 ,ω 2 ,...,ω E Represents the active weight of the current rule for all input data. If all activation weights are 0, the current rule will never be activated. Therefore, for this rule, the parameters in Ω should be retained;

[0036] 4.2 Strategy 2: Set the optimization range of this parameter:

[0037] In order to maintain the integrity of expert knowledge during the optimization process, it is necessary to set a reasonable optimization range for each parameter. This method not only utilizes expert experience, but also can be effectively adjusted in the actual application of specific electromechanical equipment, so as to achieve a balanced optimization result. This process can be described as follows:

[0038] Ω low ≤Ω≤Ω up : (9)

[0039] {θ k,low ≤θ k ≤θ k,up ,k=1,2,...,L

[0040] δ m,low ≤δ m ≤δ m,up ,m=1,2,...,M

[0041] β n,k,low ≤β n,k ≤β n,k,up ,n=1,...,N}

[0042] Where Ω up and Ω low are the upper and lower bounds of the optimization range respectively;

[0043] 4.3 Strategy 3: Ensure the confidence distribution is reasonable:

[0044] A reasonable confidence distribution is either monotonic or convex. Therefore, the confidence distribution should be strictly constrained during the optimization process to make it conform to a reasonable distribution. The constraint on confidence can be expressed as:

[0045] β i ~K i (i=1,2,...,L) (10)

[0046] K i ∈{{β 1 ≤β 2 ≤...≤β n}

[0047] or{β 1 ≥β 2 ≥...≥β n}

[0048] or{β 1 ≤...≤max(β 1 ,β 2 ,...,β n )≥...≥β n}}

[0049] 4.4 Strategy 4: Punishing behaviors that exceed the limit:

[0050] Although boundary constraints are added during the optimization process, the step size setting of the optimization algorithm may not be applicable to all parameters and iterative processes; in the process of optimization algorithm exploration, it is still possible to exceed the boundary. Although the parameter can obtain higher accuracy, its out-of-bounds behavior has destroyed the original interpretability, so it needs to be punished. The principle of punishment is to add an additional value to this parameter so that it will be eliminated in the subsequent screening process. The calculation method of this value is as follows:

[0051]

[0052] Where Ω i Represents the initial population generated by the optimization algorithm.

[0053] As a further description of the above technical solution: In step five, specifically: use P-CMA-ES to optimize the confidence to further improve the accuracy of the model. At the same time, in this step, two important parameters that affect the model results - rule weight and attribute weight - are optimized for the first time.

[0054] As a further description of the above technical solution: In step 5, the combination of the interpretability constraint strategy improves the original P-CMA-ES method, and the enhanced P-CMA-ES optimization method is explained as follows:

[0055]

[0056] The modeling accuracy of DO-BRB-I is represented by the mean square error (MSE). In this optimization, rule weight, attribute weight and confidence are the optimization parameters, so MSE can be expressed as follows:

[0057]

[0058] As a further description of the above technical solution: in step 5, the method is as follows:

[0059] 5.1 Set the initial parameters as follows:

[0060] Ω 0 ={θ 1 ,...θ k ,δ 1 ,...,δ M ,β 1 ,...,β n} (14)

[0061] where w g =Ω 0 is the optimized parameter set;

[0062] 5.2 (Sampling operation): The initial population can be determined by:

[0063] Ω i g+1 ~ω g +ε g N(0,C g )i=1,...,λ (15)

[0064] Where Ω i g+1 represents the i-th solution of the g+1-th generation, ω represents the average value of the population, ε represents the step size, N represents the normal distribution, C g represents the covariance matrix of the g-th generation;

[0065] 5.3 (Constraint Operation): In this step, four constraint strategies are added to the optimization process of the model. These four constraint strategies not only retain the rule parameters that have never been activated, but also impose detailed constraints on them to ensure that expert knowledge is not compromised;

[0066] 5.4 (Projection operation): The solution produced by the sampling operation may not satisfy the constraints, so a projection operation is required to ensure that the constraints are followed:

[0067]

[0068] The hyperplane can be represented as A e Ω ig (1+n e ×(j-1):n e ×j)=1, where n e Represents the solution Ω i g The number of constrained variables in the equation, j = 1, ..., N + 1 represents the solution The number of constraints in the medium, A e =[1...1} 1×N represents the parameter vector;

[0069] 5.5 (Selection Operation): Update the average by performing a selection operation using the following formula:

[0070]

[0071] where h i represents the weight coefficient of the i-th equation, represents the i-th solution in the (g+1)-th generation, and τ represents the subgroup size;

[0072] 5.6 (Update operation): Update the covariance matrix through adaptive operation and determine the overall search range and direction. The calculation process is shown in the following formula:

[0073]

[0074] As a further description of the above technical solution: in step six, the reasoning process is mainly: using the ER reasoning engine to perform reasoning to obtain the expected utility value, thereby obtaining the output result of the BRB model.

[0075] As a further description of the above technical solution: In step 6, the specific process is as follows:

[0076] 6.1 Calculate the matching degree between the input sample information and the confidence rule, which indicates the flexibility of the rule. The matching degree of the Kth rule to the i-th input is calculated as follows:

[0077]

[0078] in, represents the matching degree, A represents the reference value of the premise attribute, and x represents the input data. This membership function ensures that at least one rule can be activated for each data input;

[0079] 6.2 Determine the activation weight by the following calculation:

[0080]

[0081] Among them, δ i(i=1,...,M) represents the attribute weight of the i-th evaluation index;

[0082] 6.3 Use the ER algorithm to perform reasoning and obtain the final confidence β n (n=1,...,N); the calculation method is as follows:

[0083]

[0084]

[0085] 6.4 Finally, the expected utility value is calculated and the final output result is as follows:

[0086] y={(H n ,β n ),n=1,…,N} (23)

[0087]

[0088] Where A′ is the actual input vector, μ(H n ) indicates H n The utility of the proposed method is , μ(S(A′)) is the final expected utility, and S(·) is the set of confidence distributions. This method combines the IF-THEN rule-based method with the utility-based method, which improves the reliability of the initial information and facilitates the logical adjustment of the confidence structure during the reasoning process.

[0089] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0090] 1. In the present invention, the confidence of the basic BRB model is doubly optimized, and four interpretability constraint strategies are proposed in the optimization process, which further enhances the interpretability of the model, thereby better completing the safety status assessment of electromechanical equipment; at the same time, the evidence reasoning and analysis method adopted by the present invention not only considers the subjective understanding of the expert knowledge on the assessment, but also fully considers the objective control of the engineering data, thereby ensuring that the assessment results are more reliable. On the premise of ensuring the accuracy of the characteristic indicators, this method improves the transparency, rationality and integrity of the assessment process. BRIEF DESCRIPTION OF THE DRAWINGS

[0091] Figure 1 A step diagram of an interpretable dual-optimization electromechanical equipment safety status assessment method proposed by the present invention;

[0092] Figure 2 A method flow chart of an interpretable dual-optimization electromechanical equipment safety status assessment method proposed by the present invention;

[0093] Figure 3A schematic diagram of a Gaussian membership function confidence optimization method based on Bayesian updating for an interpretable dual-optimization electromechanical equipment safety status assessment method proposed in the present invention;

[0094] Figure 4 A schematic diagram of a rule weight and attribute weight optimization method for an interpretable dual-optimization electromechanical equipment safety status assessment method proposed by the present invention;

[0095] Figure 5 Schematic diagram of the evidential reasoning method of the explainable dual-optimization electromechanical equipment safety status assessment method proposed in the present invention. DETAILED DESCRIPTION

[0096] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0097] Please refer to the attached Figure 1 -Attached Figure 5 The present invention provides a technical solution: an interpretable dual-optimization electromechanical equipment safety status assessment method, comprising the following steps:

[0098] Step 1: Obtain the data set of electromechanical equipment. Through the sensors installed on the equipment, the equipment parameters are monitored and collected in real time to obtain the electromechanical equipment data set.

[0099] Step 2: Construct a basic BRB model, determine the reference values ​​of the premise attributes and the reference values ​​corresponding to the results, and use expert knowledge and input data to create a basic BRB model;

[0100] Step 3: Perform initial optimization on the confidence. For the model obtained in step 2, a Gaussian membership function confidence optimization method based on Bayesian updating (GMF-B) is proposed to perform initial optimization on the confidence.

[0101] Step 4: Propose four interpretability constraint strategies to prevent the interpretability of expert knowledge from being destroyed during the optimization process;

[0102] Step 5: Use evolutionary strategy to further optimize and use the projection covariance matrix adaptive algorithm P-CMA-ES to further optimize the parameters in the model to further improve the model accuracy;

[0103] Step 6: Obtain the safety status assessment results of the electromechanical equipment. Apply the evidence reasoning and parsing algorithm ER to infer the model after parameter tuning in step 5 to obtain the safety status assessment results of the electromechanical equipment.

[0104] Step three is as follows: combine the reference value and the prior attribute of the data set, and then obtain a more reasonable confidence distribution through Gaussian membership function and Bayesian update.

[0105] In step three, the method is as follows:

[0106] 3.1 The reference values ​​of the attribute and expert knowledge results are A and H, respectively, which are used to calculate the subsequent membership. Then, according to the initial confidence defined by the expert, the prior probability distribution matrix is ​​generated, denoted as P;

[0107] β r =g(A,β e ,x) (1)

[0108] Where A represents the reference value of the previous attribute, β e represents the initial reference value determined by expert knowledge, x represents the input data, β r represents a more reasonable confidence after optimization, and g(·) represents the function of the GMF-B method;

[0109] 3.2 According to the calculation formula of Gaussian membership function, the membership of the previous attribute reference value is calculated. The calculation formula of Gaussian membership function is as follows:

[0110]

[0111] According to the calculation formula of Gaussian membership function, the membership of the reference value of the previous attribute result is calculated. The calculation formula of Gaussian membership function is as follows:

[0112]

[0113] 3.3 By multiplying the membership of the reference value of the previous attribute in the current rule with the membership of the result, the comprehensive membership corresponding to the result can be obtained, and the formula is as follows:

[0114]

[0115] in Indicates the comprehensive membership corresponding to the i-th result in the current rule, φ 1 A ,φ 2 A ,…,φ M A represents the membership degree of M prior attribute reference values, φ i H represents the membership degree of the i-th result;

[0116] 3.4 The above steps calculate the comprehensive membership of the results based on the input data and reference values. This membership can be used as the influence of data on confidence. Therefore, the initial expert knowledge is used as the prior probability distribution, and the above membership is used as new evidence for the Bayesian method to update the prior probability. This process can be expressed as follows:

[0117]

[0118] P(C i |H)=P(H|C i )(C i ) (6)

[0119] Where P(H|C i ) is the likelihood function expressed as the sum of comprehensive memberships, P(C i ) is the prior probability of the current combination, P(C i |H) is the posterior probability of the current combination;

[0120] 3.5 Repeat the first four steps, calculate the posterior probability of each set of input data and each combination, and normalize them to obtain the final posterior possibility distribution matrix, which is the confidence distribution of expert knowledge optimized by the data distribution.

[0121] In step 4, four interpretability constraint strategies are proposed to constrain the second optimization process of the model. The following is a detailed introduction to these four strategies:

[0122] 4.1 Strategy 1: Ensure that activated rules participate in optimization and reasoning:

[0123] Ω={θ 1 ,...,θ k ,δ 1 ,...,δ M ,β 1,k ,...,β N,k} (7)

[0124] The activation weight can be used to identify inactive rules. The activation of the kth rule can be expressed as follows:

[0125] Q k ={ω 1 ,ω 2 ,...,ω E},k=1,2,...,L (8)

[0126] Where E represents the number of input data, ω 1 ,ω 2 ,...,ω ERepresents the active weight of the current rule for all input data. If all activation weights are 0, the current rule will never be activated. Therefore, for this rule, the parameters in Ω should be retained;

[0127] 4.2 Strategy 2: Set the optimization range of this parameter:

[0128] In order to maintain the integrity of expert knowledge during the optimization process, it is necessary to set a reasonable optimization range for each parameter. This method not only utilizes expert experience, but also can be effectively adjusted in the actual application of specific electromechanical equipment, so as to achieve a balanced optimization result. This process can be described as follows:

[0129] Ω low ≤Ω≤Ω up : (9)

[0130] {θ k,low ≤θ k ≤θ k,up ,k=1,2,...,L

[0131] δ m,low ≤δ m ≤δ m,up ,m=1,2,...,M

[0132] β n,k,low ≤β n,k ≤β n,k,up ,n=1,...,N}

[0133] Where Ω up and Ω low are the upper and lower bounds of the optimization range respectively;

[0134] 4.3 Strategy 3: Ensure the confidence distribution is reasonable:

[0135] A reasonable confidence distribution is either monotonic or convex. Therefore, the confidence distribution should be strictly constrained during the optimization process to make it conform to a reasonable distribution. The constraint on confidence can be expressed as:

[0136] β i ~K i (i=1,2,...,L) (10)

[0137] K i ∈{{β 1 ≤β 2 ≤...≤β n}

[0138] or{β 1 ≥β 2 ≥...≥β n}

[0139] or{β 1 ≤...≤max(β 1 ,β 2 ,...,β n )≥...≥β n}}

[0140] 4.4 Strategy 4: Punishing behaviors that exceed the limit:

[0141] Although boundary constraints are added during the optimization process, the step size setting of the optimization algorithm may not be applicable to all parameters and iterative processes; in the process of optimization algorithm exploration, it is still possible to exceed the boundary. Although the parameter can obtain higher accuracy, its out-of-bounds behavior has destroyed the original interpretability, so it needs to be punished. The principle of punishment is to add an additional value to this parameter so that it will be eliminated in the subsequent screening process. The calculation method of this value is as follows:

[0142]

[0143] Where Ω i Represents the initial population generated by the optimization algorithm.

[0144] In step five, specifically: use P-CMA-ES to optimize the confidence to further improve the accuracy of the model. At the same time, in this step, two important parameters that affect the model results - rule weight and attribute weight - are optimized for the first time.

[0145] In step 5, the incorporation of the interpretability constraint strategy improves the original P-CMA-ES method. The enhanced P-CMA-ES optimization method is explained as follows:

[0146]

[0147] The modeling accuracy of DO-BRB-I is represented by the mean square error (MSE). In this optimization, rule weight, attribute weight and confidence are the optimization parameters, so MSE can be expressed as follows:

[0148]

[0149] In step five, the method is as follows:

[0150] 5.1 Set the initial parameters as follows:

[0151] Ω 0 ={θ 1 ,...θ k ,δ 1 ,...,δ M ,β 1 ,...,β n} (14)

[0152] where w g =Ω 0 is the optimized parameter set;

[0153] 5.2 (Sampling operation): The initial population can be determined by:

[0154] Ω i g+1 ~ω g +ε g N(0,C g )i=1,...,λ(15)

[0155] Where Ω i g+1 represents the i-th solution of the g+1-th generation, ω represents the average value of the population, ε represents the step size, N represents the normal distribution, C g represents the covariance matrix of the g-th generation;

[0156] 5.3 (Constraint Operation): In this step, four constraint strategies are added to the optimization process of the model. These four constraint strategies not only retain the rule parameters that have never been activated, but also impose detailed constraints on them to ensure that expert knowledge is not compromised;

[0157] 5.4 (Projection operation): The solution produced by the sampling operation may not satisfy the constraints, so a projection operation is required to ensure that the constraints are followed:

[0158]

[0159] The hyperplane can be represented as A e Ω i g (1+n e ×(j-1):n e ×j)=1, where n e Represents the solution Ω i g The number of constrained variables in the equation, j = 1, ..., N + 1 represents the solution The number of constraints in the medium, A e =[1...1} 1×N represents the parameter vector;

[0160] 5.5 (Selection Operation): Update the average by performing a selection operation using the following formula:

[0161]

[0162] where h i represents the weight coefficient of the i-th equation, represents the i-th solution in the (g+1)-th generation, and τ represents the subgroup size;

[0163] 5.6 (Update operation): Update the covariance matrix through adaptive operation and determine the overall search range and direction. The calculation process is shown in the following formula:

[0164]

[0165] In step six, the reasoning process is mainly: use the ER reasoning engine to perform reasoning, obtain the expected utility value, and thus obtain the output result of the BRB model.

[0166] In step six, the specific process is as follows:

[0167] 6.1 Calculate the matching degree between the input sample information and the confidence rule, which indicates the flexibility of the rule. The matching degree of the Kth rule to the i-th input is calculated as follows:

[0168]

[0169] in, represents the matching degree, A represents the reference value of the premise attribute, and x represents the input data. This membership function ensures that at least one rule can be activated for each data input;

[0170] 6.2 Determine the activation weight by the following calculation:

[0171]

[0172] Among them, δ i (i=1,...,M) represents the attribute weight of the i-th evaluation index;

[0173] 6.3 Use the ER algorithm to perform reasoning and obtain the final confidence β n (n=1,...,N); the calculation method is as follows:

[0174]

[0175]

[0176] 6.4 Finally, the expected utility value is calculated and the final output result is as follows:

[0177] y={(H n ,β n ),n=1,…,N} (23)

[0178]

[0179] Where A′ is the actual input vector, μ(H n ) indicates Hn The utility of the proposed method is , μ(S(A′)) is the final expected utility, and S(·) is the set of confidence distributions. This method combines the IF-THEN rule-based method with the utility-based method, which improves the reliability of the initial information and facilitates the logical adjustment of the confidence structure during the reasoning process.

[0180] Based on the above, the advantage of the present invention is that when the present invention is used, the basic BRB model is first created using expert knowledge and input data; the level of the result is determined by the expert, and on the premise of having the premise attribute reference value and the result reference value, the Gaussian membership function confidence optimization method (GMF-B) based on Bayesian update is adopted; this method can optimize the confidence according to the data distribution while maintaining the original expert knowledge; then the projection covariance matrix adaptive evolution strategy (P-CMA-ES) is used to further optimize other parameters such as confidence, forming a double optimization; finally, the constructed model is inferred by the evidence reasoning analysis algorithm to obtain the final result; in addition, in view of the randomness of the optimization algorithm, according to the characteristics of electromechanical equipment, four interpretability constraint strategies are proposed to constrain the interpretability; under the premise of ensuring interpretability, the present invention implements a double improvement strategy for the safety status assessment of electromechanical equipment; in this improvement process, by reasonably adjusting the parameters, not only its interpretability is maintained, but also the accuracy of the assessment is significantly improved.

[0181] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. An interpretable dual-optimization electromechanical equipment safety status assessment method, characterized in that: The following steps are involved: Step 1: Obtain the data set of electromechanical equipment. Through the sensors installed on the equipment, the equipment parameters are monitored and collected in real time to obtain the electromechanical equipment data set. Step 2: Construct a basic BRB model, determine the reference values ​​of the premise attributes and the reference values ​​corresponding to the results, and use expert knowledge and input data to create a basic BRB model; Step 3: Perform initial optimization on the confidence. For the model obtained in step 2, a Gaussian membership function confidence optimization method based on Bayesian updating (GMF-B) is proposed to perform initial optimization on the confidence. Step 4: Propose four interpretability constraint strategies to prevent the interpretability of expert knowledge from being destroyed during the optimization process; Step 5: Use evolutionary strategy to further optimize and use the projection covariance matrix adaptive algorithm P-CMA-ES to further optimize the parameters in the model to further improve the model accuracy; Step 6: Obtain the safety status assessment results of the electromechanical equipment. Apply the evidence reasoning and parsing algorithm ER to infer the model after parameter tuning in step 5 to obtain the safety status assessment results of the electromechanical equipment.

2. According to claim 1, an interpretable dual-optimization electromechanical equipment safety status assessment method is characterized in that: The step three is specifically: combining the reference value and the prior attribute of the data set, and then obtaining a more reasonable confidence distribution through Gaussian membership function and Bayesian updating.

3. According to claim 2, an interpretable dual-optimization electromechanical equipment safety status assessment method is characterized in that in the step three, the method is as follows: 3.1 The reference values ​​of the attribute and expert knowledge results are A and H, respectively, which are used to calculate the subsequent membership. Then, according to the initial confidence defined by the expert, the prior probability distribution matrix is ​​generated, denoted as P; b r =g(A,β e ,x) (1) Where A represents the reference value of the previous attribute, β e represents the initial reference value determined by expert knowledge, x represents the input data, β r represents a more reasonable confidence after optimization, and g(·) represents the function of the GMF-B method; 3.2 According to the calculation formula of Gaussian membership function, the membership of the previous attribute reference value is calculated. The calculation formula of Gaussian membership function is as follows: According to the calculation formula of Gaussian membership function, the membership of the reference value of the previous attribute result is calculated. The calculation formula of Gaussian membership function is as follows: 3.3 By multiplying the membership of the reference value of the previous attribute in the current rule with the membership of the result, the comprehensive membership corresponding to the result can be obtained, and the formula is as follows: in Indicates the comprehensive membership corresponding to the i-th result in the current rule, φ1 A ,φ2 A ,…,φ M A represents the membership degree of M prior attribute reference values, φ i H represents the membership degree of the i-th result; 3.4 The above steps calculate the comprehensive membership of the results based on the input data and reference values. This membership can be used as the influence of data on confidence. Therefore, the initial expert knowledge is used as the prior probability distribution, and the above membership is used as new evidence for the Bayesian method to update the prior probability. This process can be expressed as follows: P(C i |H)=P(H|C i )(C i ) (6) Where P(H|C i ) is the likelihood function expressed as the sum of comprehensive memberships, P(C i ) is the prior probability of the current combination, P(C i |H) is the posterior probability of the current combination; 3.5 Repeat the first four steps, calculate the posterior probability of each set of input data and each combination, and normalize them to obtain the final posterior possibility distribution matrix, which is the confidence distribution of expert knowledge optimized by the data distribution.

4. According to claim 1, an interpretable dual-optimization electromechanical equipment safety status assessment method is characterized in that, in the step 4, four interpretable constraint strategies are proposed to constrain the second optimization process of the model. The following is a detailed introduction to these four strategies: 4.1 Strategy 1: Ensure that activated rules participate in optimization and reasoning: Ω={θ1,...,θ k ,δ1,...,δ M ,b 1,k ,...,b N,k } (7) The activation weight can be used to identify inactive rules. The activation of the kth rule can be expressed as follows: Q k ={ω1,ω2,...,ω E },k=1,2,...,L (8) Where E represents the number of input data, ω1,ω2,...,ω E Represents the active weight of the current rule for all input data. If all activation weights are 0, the current rule will never be activated. Therefore, for this rule, the parameters in Ω should be retained; 4.2 Strategy 2: Set the optimization range of this parameter: In order to maintain the integrity of expert knowledge during the optimization process, it is necessary to set a reasonable optimization range for each parameter. This method not only utilizes expert experience, but also can be effectively adjusted in the actual application of specific electromechanical equipment, so as to achieve a balanced optimization result. This process can be described as follows: Oh low ≤Ω≤Ω up : (9) {i} k,low ≤θ k ≤θ k,up ,k=1,2,...,L d m,low ≤δ m ≤δ m,up ,m=1,2,...,M b n,k,low ≤β n,k ≤β n,k,up ,n=1,...,N} Where Ω up and Ω low are the upper and lower bounds of the optimization range respectively; 4.3 Strategy 3: Ensure the confidence distribution is reasonable: A reasonable confidence distribution is either monotonic or convex. Therefore, the confidence distribution should be strictly constrained during the optimization process to make it conform to a reasonable distribution. The constraint on confidence can be expressed as: b i ~K i (i=1,2,...,L) (10) K i ∈{{β1≤β2≤...≤β n} or{β1≥β2≥...≥β n} or{β1≤...≤max(β1,β2,...,β n )≥...≥β n }} 4.4 Strategy 4: Punishing behaviors that exceed the limit: Although boundary constraints are added during the optimization process, the step size setting of the optimization algorithm may not be applicable to all parameters and iterative processes; in the process of optimization algorithm exploration, it is still possible to exceed the boundary. Although the parameter can obtain higher accuracy, its out-of-bounds behavior has destroyed the original interpretability, so it needs to be punished. The principle of punishment is to add an additional value to this parameter so that it will be eliminated in the subsequent screening process. The calculation method of this value is as follows: Where Ω i Represents the initial population generated by the optimization algorithm.

5. According to claim 4, an interpretable dual-optimization electromechanical equipment safety status assessment method is characterized in that: In the step five, specifically: use P-CMA-ES to optimize the confidence to further improve the accuracy of the model. At the same time, in this step, two important parameters that affect the model results, rule weight and attribute weight, are optimized for the first time.

6. The interpretable dual-optimization electromechanical equipment safety status assessment method according to claim 5 is characterized in that: In step 5, the combination of the interpretability constraint strategy improves the original P-CMA-ES method. The enhanced P-CMA-ES optimization method is explained as follows: The modeling accuracy of DO-BRB-I is represented by the mean square error (MSE). In this optimization, rule weight, attribute weight and confidence are the optimization parameters, so MSE can be expressed as follows:

7. The interpretable dual-optimization electromechanical equipment safety status assessment method according to claim 6 is characterized in that: In the step 5, the method is as follows: 5.1 Set the initial parameters as follows: Oh 0 ={θ1,...θ k ,δ1,...,δ M ,β1,...,β n } (14) where w g =Ω 0 is the optimized parameter set; 5.2 (Sampling operation): The initial population can be determined by: Oh i g+1 ~ω g +e g N(0,C g )i=1,...,λ (15) Where Ω i g+1 represents the i-th solution of the g+1-th generation, ω represents the average value of the population, ε represents the step size, N represents the normal distribution, C g represents the covariance matrix of the g-th generation; 5.3 (Constraint Operation): In this step, four constraint strategies are added to the optimization process of the model. These four constraint strategies not only retain the rule parameters that have never been activated, but also impose detailed constraints on them to ensure that expert knowledge is not compromised; 5.4 (Projection operation): The solution produced by the sampling operation may not satisfy the constraints, so a projection operation is required to ensure that the constraints are followed: The hyperplane can be represented as A e Ω i g (1+n e ×(j-1):n e ×j)=1, where n e Represents the solution Ω i g The number of constrained variables in the equation, j = 1, ..., N + 1 represents the solution The number of constraints in the medium, A e =[1...1} 1×N represents the parameter vector; 5.5 (Selection Operation): Update the average by performing a selection operation using the following formula: where h i represents the weight coefficient of the i-th equation, represents the i-th solution in the (g+1)-th generation, and τ represents the subgroup size; 5.6 (Update operation): Update the covariance matrix through adaptive operation and determine the overall search range and direction. The calculation process is shown in the following formula:

8. The interpretable dual-optimization electromechanical equipment safety status assessment method according to claim 1 is characterized in that: In step six, the reasoning process mainly includes: using the ER reasoning engine to perform reasoning to obtain the expected utility value, thereby obtaining the output result of the BRB model.

9. The interpretable dual-optimization electromechanical equipment safety status assessment method according to claim 8 is characterized in that: In step six, the specific process is as follows: 6.1 Calculate the matching degree between the input sample information and the confidence rule, which indicates the flexibility of the rule. The matching degree of the Kth rule to the i-th input is calculated as follows: in, represents the matching degree, A represents the reference value of the premise attribute, and x represents the input data. This membership function ensures that at least one rule can be activated for each data input; 6.2 Determine the activation weight by the following calculation: Among them, δ i (i=1,...,M) represents the attribute weight of the i-th evaluation index; 6.3 Use the ER algorithm to perform reasoning and obtain the final confidence β n (n=1,...,N); the calculation method is as follows: 6.4 Finally, the expected utility value is calculated and the final output result is as follows: y={(H n ,β n ),n=1,…,N} (23) Where A′ is the actual input vector, μ(H n ) indicates H n The utility of the proposed method is , μ(S(A′)) is the final expected utility, and S(·) is the set of confidence distributions. This method combines the IF-THEN rule-based method with the utility-based method, which improves the reliability of the initial information and facilitates the logical adjustment of the confidence structure during the reasoning process.

Images