Trusted ship power engine intelligent fault diagnosis method
By screening important features in ship power engine fault diagnosis, building an interpretable initial model and using P-CMA-ES optimization strategy, the problem of difficult to balance accuracy and interpretability in the existing technology is solved, and high-precision and high-interpretability fault diagnosis is achieved.
Patent Information
- Application Number
- CN202411856247.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-17
- Publication Date
- 2025-05-16
AI Technical Summary
The existing ship-powered engine fault diagnosis methods are difficult to balance between accuracy and interpretability, resulting in insufficient accuracy and interpretability of the model, and lack of high credibility and high interpretability methods.
A trusted intelligent fault diagnosis method for ship power engines is adopted to ensure the high fault diagnosis accuracy and interpretability of the model by obtaining fault data, filtering important features, building an interpretable initial model, designing a balanced optimization strategy, and using an improved projection covariance matrix adaptive evolution strategy (P-CMA-ES).
While maintaining high fault diagnosis accuracy, the interpretability of the model is significantly enhanced, solving the balance between accuracy and interpretability, and improving the credibility and availability of fault diagnosis results.
Smart Images
Figure CN120011774A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of fault detection, and in particular relates to a reliable intelligent fault diagnosis method for a ship power engine. Background Art
[0002] Ship power engines operate under harsh conditions such as high humidity, high salinity, long-term continuous operation, and complex load changes in the marine environment. As the core power device of the ship, the operating status of the ship power engine is directly related to the reliability and safety of the entire ship. Therefore, it is very important to establish a reliable intelligent fault diagnosis method for ship power engines.
[0003] However, existing ship engine fault diagnosis methods have obvious deficiencies and limitations in terms of the balance between accuracy and interpretability. Existing technologies mostly rely on signal processing and machine learning techniques, such as spectrum analysis, wavelet transform, and vibration signal feature extraction, and combine support vector machine (SVM) and deep learning algorithms for fault classification and diagnosis. Although these methods have improved the accuracy of fault detection, they still face several key problems:
[0004] First, there are many indicators that affect the safety status of ship power engines. However, too many input attributes will lead to an explosion of BRB combination rules;
[0005] Secondly, the accuracy of the model can be significantly improved by optimizing the algorithm. However, the randomness of the optimization algorithm often destroys the interpretability of the model.
[0006] Finally, in rule-based systems, decisions depend on explicit sets of rules. Rules built based on expert knowledge can be understood, but their accuracy is average. Rules generated by learning from data have high accuracy but low interpretability. In the field of ship power engine fault detection, which requires high credibility and high interpretability, methods that lack interpretability lack practicality.
[0007] To build a reliable intelligent fault diagnosis method for ship power engines, two problems need to be solved: how to significantly enhance the interpretability of the model while maintaining high fault diagnosis accuracy, thereby solving the balance problem between accuracy and interpretability. The present invention proposes a reliable intelligent fault diagnosis method for ship power engines. This method can not only effectively extract fault features through interpretable and intelligent technologies and reduce dependence on manual experience, but also ensure interpretability while ensuring diagnostic accuracy through optimization algorithms, and well balance the accuracy and interpretability of the model, thereby improving the credibility and usability of fault diagnosis results. Therefore, it is of great practical significance to establish a reliable intelligent fault diagnosis method for ship power engines.
[0008] Based on this, the present invention designs a reliable intelligent fault diagnosis method for ship power engine to solve the above problems. Summary of the invention
[0009] The purpose of the present invention is to solve the obvious deficiencies and limitations of some ship power engine fault diagnosis methods in terms of the balance between accuracy and interpretability. The existing technologies mostly rely on signal processing and machine learning technologies, such as spectrum analysis, wavelet transform and vibration signal feature extraction, and combine support vector machine (SVM) and deep learning algorithms for fault classification and diagnosis. Although these methods have improved the accuracy of fault detection, they still face several key problems: First, there are many indicators that affect the safety status of ship power engines. However, too many input attributes will lead to an explosion of combination rules of BRB. Secondly, the accuracy of the model can be significantly improved by optimization algorithms. However, the randomness of the optimization algorithm often destroys the interpretability of the model. Finally, in a rule-based system, decisions depend on explicit rule sets. Rules constructed based on expert knowledge can be understood, but the accuracy is general. Rules generated by learning from data have high accuracy but low interpretability. In the field of ship power engine fault detection that requires high credibility and high interpretability, methods that lack interpretability lack practicality. A reliable ship power engine intelligent fault diagnosis method is proposed.
[0010] In order to achieve the above object, the present invention adopts the following technical solutions:
[0011] A reliable intelligent fault diagnosis method for a ship power engine comprises the following steps:
[0012] Step 1, obtaining ship power engine failure data;
[0013] Step 2: Screen important features of the fault data, extract fault features from the data set obtained in step 1, sort the feature importance based on the Spearman correlation coefficient, and select features with higher importance;
[0014] Step 3: Build an interpretable initial model and use the evidence reasoning algorithm to infer the confidence rule base;
[0015] Step 4: Design a balanced optimization strategy for both accuracy and interpretability.
[0016] Step 5: Optimize the model using the improved projected covariance matrix adaptive evolutionary strategy (P-CMA-ES) that can balance interpretability and accuracy.
[0017] Step six: case test, conduct case test on the optimized model.
[0018] As a further description of the above technical solution:
[0019] The data in step 1 are various monitoring indicators of the ship's power engine, including power, speed, torque, fuel consumption rate, thermal efficiency, emission standards, specific oil consumption rate, intake pressure and temperature, exhaust temperature, noise and vibration, reliability and maintenance interval, starting performance and lubricating oil consumption rate.
[0020] As a further description of the above technical solution:
[0021] In step 2, attribute selection is performed based on the absolute value of the Spearman correlation coefficient, which is calculated as follows:
[0022]
[0023] in, Indicates i th Input attribute data x i and the Spearman correlation coefficient between the result attribute data y, yes The absolute value of describes the magnitude of the correlation, p is the number of samples, and r*f(□) represents the rank of the attribute data, that is, the position of the data in descending order;
[0024] Then, the relevance of the attributes is sorted in order from largest to smallest, and the process is expressed as:
[0025]
[0026] Among them, q is the number of indicators in the data set that have the potential to be used as input attributes, x1,x2,…,x q are the sorted attributes, and x1 has the highest correlation with the result.
[0027] As a further description of the above technical solution:
[0028] In the step three, an interpretable initial confidence rule base model is constructed. The specific process is to select two features with the highest importance from the important features selected in the above step two, and input them into the model to create a confidence rule base model.
[0029] As a further description of the above technical solution:
[0030] In step 3, the confidence rule base is inferred by the evidence reasoning algorithm, and the reasoning process includes:
[0031] First, the matching degree between the input data and the reference value is calculated as follows:
[0032]
[0033] Among them, S(x i ) represents the input data x i The process of information transformation, Represents x i Belong to i th Attribute j th The matching degree of the reference value, J i Indicates the number of reference values, and Represents two adjacent reference values,
[0034] After obtaining the matching degree of the data in the rule, the activation weight is calculated as follows:
[0035]
[0036] Among them, ω k Indicates k th The activation weight of the rule, δ i Indicates the relative weight of the attribute;
[0037] Then, the activated rules are used as the reasoning basis, and the evidence reasoning algorithm is used to reason about the trust level of the result. The reasoning process is as follows:
[0038]
[0039] Among them, β n (n=1,2,…,N) represents n th Result Level D n degree of belief;
[0040] The output form of the confidence rule base is as follows:
[0041]
[0042] in, A set of data representing attributes The belief distribution obtained;
[0043] Finally, the inference results of the confidence rule base are calculated as follows:
[0044]
[0045] in, represents the final result of the confidence rule base, u(D n ) represents n th The utility value of the resulting level, Represents the prediction result of the confidence rule base.
[0046] As a further description of the above technical solution:
[0047] In step 4, in the designed interpretability strategy, inactive rules are not allowed to participate in the optimization, and the activation factor To determine the activation status of the rule, it is expressed as:
[0048]
[0049] η k =(w1,w2,…,w m ) (12)
[0050] k=1,2,…,L,m=1,2,…,M (13)
[0051] Each set of input data may activate some rules, when the activation weight w m ≠0, it means that the rule is activated by this set of input data, η k It means that after inputting M groups of data, k th The total activation state of the rule, if η k All 0, indicating k th If the rule is not activated, the activation factor Finally, keep the relevant parameters of the rule.
[0052] As a further description of the above technical solution:
[0053] In step 4, in the designed interpretability strategy, the interpretable step-length convergence strategy determines the initial step-length and controls the convergence speed by introducing the reliability of expert knowledge. The step-length convergence calculation is as follows:
[0054]
[0055] Among them, γ t+1 represents the step size, T represents the maximum number of iterations, ε represents the minimum threshold of the step size, E c represents the credibility of expert knowledge, and its calculation method is as follows:
[0056]
[0057] Among them, E c represents the credibility of expert knowledge, y(i) represents the estimated value obtained from the knowledge provided by the expert through the confidence rule base expert system, y0(i) represents the true state of the ship power engine, F represents the trust factor, and U represents the size of the training data set.
[0058] As a further description of the above technical solution:
[0059] In step 4, in the designed interpretability strategy, the process of the population selection strategy based on dual statistics guidance includes:
[0060] First, for v th The population calculates the mean square error (MSE) value and the Euclidean distance (Ed) value. The Ed value is the Euclidean distance between the population individuals and the expert knowledge, which indicates the degree of distance or proximity to the expert knowledge during the optimization process.
[0061]
[0062] Among them, MSE(Φ v:τ ) and Ed(Φ v:τ ) respectively represent the th MSE and Ed values calculated for subpopulations, τ represents the population size;
[0063] Then, these groups were sorted and numbered according to their MSE values, with the smallest value listed first;
[0064] No MSE =Sort[MSE(Φ v:τ )],v=1,2,…,τ (17)
[0065] Similarly, the rankings are numbered according to the Ed value;
[0066] No Ed =Sort[Ed(Φ v:τ )],v=1,2,…,τ (18)
[0067] Calculate the overall ranking and then add E c As a balancing factor, and select a superior subgroup based on the balancing factor;
[0068] No balance =Sort[E c *No MSE +(1-E c )*No Ed ] (19)
[0069]
[0070] in, represents the selected comprehensive excellent sub-population, It represents the size of the offspring population. Finally, according to the evolution of these excellent sub-populations, it iterates to the next generation population.
[0071] As a further description of the above technical solution:
[0072] In step 4, in the designed interpretability strategy, the rule confidence distribution should be subject to the following constraints:
[0073]
[0074] Among them, C β Represents the constraints on the regular confidence distribution.
[0075] As a further description of the above technical solution:
[0076] In the step 5, an improved projection covariance matrix adaptive evolution strategy (P-CMA-ES) optimization algorithm is used to balance interpretability and accuracy;
[0077] First, determine the objective function:
[0078]
[0079] Where Ω represents the objective function based on dual statistics;
[0080] Then, optimization is performed, and the process optimization process includes:
[0081] Step 1: Initialization, let the initial parameters be in Represents the initial parameter vector to be optimized, parameter set The definition is as follows:
[0082]
[0083] Step 2: Sampling, in order to generate the population, you need to do the following:
[0084]
[0085] in, is (t+1) th The generated v th Solution vector, o represents the population mean, ε is the normal distribution, M t t th The covariance matrix of the generation population, γ is the step size;
[0086] Step 3: Add constraints and introduce interpretable constraints. This step is performed as follows:
[0087] (1) Inactivated rules are not allowed to participate in optimization;
[0088] [η k =1]→[θ1,…,θ L ,β 1,1 ,…,β N,L ,δ1,…,δ N ] retain (25)
[0089] (2) The optimized rules are consistent with expert judgment;
[0090] [β1,1 ,…,β N,L ]~C β (26)
[0091] Among them, β 1,1 ,…,β N,L is the newly generated belief distribution;
[0092] Step 4: Projection. In order to make the solution vector satisfy the optimization constraints, this operation includes the following:
[0093]
[0094] Where d = 1, 2, ..., D represents the number of constrained variables, τ represents the number of constraints, and V = [1, 1, ..., 1] 1*L is the parameter vector;
[0095] Step 5: Select, based on the population selection method guided by dual statistics to select solutions and update subpopulations, calculate and rank the MSE value and Ed value for the population;
[0096]
[0097] Based on the overall ranking, the best subgroup is selected;
[0098] No balance =Sort[E c *No MSE +(1-E c )*No Ed ] (29)
[0099]
[0100] The optimal subgroup update is:
[0101]
[0102] Among them, q is the weight set;
[0103] Step 6: Adjust, perform adaptation operation to update the covariance matrix;
[0104]
[0105]
[0106] Among them, e1, e2, e s ,e c is the learning rate;
[0107] Step 7: Interpretable step size convergence, the step size γ can be expressed as follows:
[0108]
[0109] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:
[0110] 1. In the present invention, the interpretability of the model can be significantly enhanced while maintaining high fault diagnosis accuracy, thereby solving the balance problem between accuracy and interpretability, effectively enhancing the safety of the system, and effectively reducing the errors caused by human factors. The attribute selection method adopted can effectively screen the important features of the ship power engine fault data set and improve the diagnostic accuracy. At the same time, an interpretability strategy is designed to improve the interpretability in the optimization stage, ensure the interpretability of the model, and use the credibility of expert knowledge to balance the interpretability and accuracy of the model.
[0111] 2. In the present invention, the Spearman correlation coefficient method is used to screen important features of historical data of ship power engine failures, reduce data dimensions, improve model efficiency, reduce the risk of model overfitting, enhance the interpretability of the model and improve the performance of fault diagnosis.
[0112] 3. In the present invention, the balanced optimization strategy for the dual metrics of accuracy and interpretability improves the practicality of the model, reduces overfitting, enhances the generalization ability, and improves the transparency and trustworthiness of the model.
[0113] 4. In the present invention, the improved projection covariance matrix adaptive evolutionary strategy (P-CMA-ES) that can balance interpretability and accuracy is used for optimization, which can improve the optimization efficiency, reduce irrelevant or redundant variables in high-dimensional optimization problems, enhance the accuracy of the model and maintain strong interpretability, so that the model has better robustness. BRIEF DESCRIPTION OF THE DRAWINGS
[0114] Figure 1 A schematic diagram of a method flow of a reliable intelligent fault diagnosis method for a ship power engine proposed by the present invention;
[0115] Figure 2 A structural schematic diagram of model reasoning in a reliable intelligent fault diagnosis method for ship power engines proposed by the present invention;
[0116] Figure 3 A flow chart of the optimization algorithm in a reliable intelligent fault diagnosis method for a ship power engine proposed by the present invention. DETAILED DESCRIPTION
[0117] 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.
[0118] Please see attached Figure 1 -Attached Figure 3 The present invention provides a technical solution: a reliable intelligent fault diagnosis method for a ship power engine, comprising the following steps:
[0119] Step 1, obtaining ship power engine failure data;
[0120] Step 2: Screen important features of the fault data, extract fault features from the data set obtained in step 1, sort the feature importance based on the Spearman correlation coefficient, and select features with higher importance;
[0121] Step 3: Build an interpretable initial model and use the evidence reasoning algorithm to infer the confidence rule base;
[0122] Step 4: Design a balanced optimization strategy for both accuracy and interpretability.
[0123] Step 5: Optimize the model using the improved projected covariance matrix adaptive evolutionary strategy (P-CMA-ES) that can balance interpretability and accuracy.
[0124] Step six: case test, conduct case test on the optimized model.
[0125] Specifically, the data in step one are various monitoring indicators of the ship's power engine, including power, speed, torque, fuel consumption rate, thermal efficiency, emission standards, specific oil consumption rate, intake pressure and temperature, exhaust temperature, noise and vibration, reliability and maintenance interval, starting performance and lubricating oil consumption rate.
[0126] Specifically, in step 2, attribute selection is performed based on the absolute value of the Spearman correlation coefficient, which is calculated as follows:
[0127]
[0128] in, Indicates i th Input attribute data x i and the Spearman correlation coefficient between the result attribute data y, yes The absolute value of describes the magnitude of the correlation, p is the number of samples, and r*f(□) represents the rank of the attribute data, that is, the position of the data in descending order;
[0129] Then, the relevance of the attributes is sorted in order from largest to smallest, and the process is expressed as:
[0130]
[0131] Among them, q is the number of indicators in the data set that have the potential to be used as input attributes, x1,x2,…,x q are the sorted attributes, and x1 has the highest correlation with the result.
[0132] Specifically, in step three, an interpretable initial confidence rule base model is constructed. The specific process is to select two features with the highest importance from the important features selected in step two above, and input them into the model to create a confidence rule base model.
[0133] Specifically, in step 3, the confidence rule base is inferred by the evidence reasoning algorithm, and the reasoning process includes:
[0134] First, the matching degree between the input data and the reference value is calculated as follows:
[0135]
[0136] Among them, S(x i ) represents the input data x i The process of information transformation, Represents x i Belong to i th Attribute j th The matching degree of the reference value, J i Indicates the number of reference values, and Represents two adjacent reference values,
[0137] After obtaining the matching degree of the data in the rule, the activation weight is calculated as follows:
[0138]
[0139] Among them, ω k Indicates k th The activation weight of the rule, δ i Indicates the relative weight of the attribute;
[0140] Then, the activated rules are used as the reasoning basis, and the evidence reasoning algorithm is used to reason about the trust level of the result. The reasoning process is as follows:
[0141]
[0142] Among them, β n (n=1,2,…,N) represents n th Result Level D n degree of belief;
[0143] The output form of the confidence rule base is as follows:
[0144]
[0145] in, A set of data representing attributes The belief distribution obtained;
[0146] Finally, the inference results of the confidence rule base are calculated as follows:
[0147]
[0148] in, represents the final result of the confidence rule base, u(D n ) represents n th The utility value of the resulting level, Represents the prediction result of the confidence rule base.
[0149] Specifically, in step 4, in the designed interpretability strategy, inactive rules are not allowed to participate in the optimization, and the activation factor To determine the activation status of the rule, it is expressed as:
[0150]
[0151] η k =(w1,w2,…,w m ) (12)
[0152] k=1,2,…,L,m=1,2,…,M (13)
[0153] Each set of input data may activate some rules, when the activation weight w m ≠0, it means that the rule is activated by this set of input data, η k It means that after inputting M groups of data, k th The total activation state of the rule, if η k All 0, indicating k th If the rule is not activated, the activation factor Finally, keep the relevant parameters of the rule.
[0154] Specifically, in step 4, in the designed interpretability strategy, the interpretable step-length convergence strategy determines the initial step-length and controls the convergence speed by introducing the reliability of expert knowledge. The step-length convergence calculation is as follows:
[0155]
[0156] Among them, γ t+1 represents the step size, T represents the maximum number of iterations, ε represents the minimum threshold of the step size, E c represents the credibility of expert knowledge, and its calculation method is as follows:
[0157]
[0158] Among them, E c represents the credibility of expert knowledge, y(i) represents the estimated value obtained from the knowledge provided by the expert through the confidence rule base expert system, y0(i) represents the true state of the ship power engine, F represents the trust factor, and U represents the size of the training data set.
[0159] Specifically, in the step 4, in the designed interpretability strategy, the process of the population selection strategy based on dual statistics guidance includes:
[0160] First, for v th The population calculates the mean square error (MSE) value and the Euclidean distance (Ed) value. The Ed value is the Euclidean distance between the population individuals and the expert knowledge, which indicates the degree of distance or proximity to the expert knowledge during the optimization process.
[0161]
[0162] Among them, MSE(Φ v:τ ) and Ed(Φ v:τ ) respectively represent the th MSE and Ed values calculated for subpopulations, τ represents the population size;
[0163] Then, these groups were sorted and numbered according to their MSE values, with the smallest value listed first;
[0164] No MSE =Sort[MSE(Φ v:τ )],v=1,2,…,τ (17)
[0165] Similarly, the rankings are numbered according to the Ed value;
[0166] No Ed =Sort[Ed(Φ v:τ )],v=1,2,…,τ (18)
[0167] Calculate the overall ranking and then add E c As a balancing factor, and select a superior subgroup based on the balancing factor;
[0168] No balance =Sort[E c *No MSE +(1-E c )*No Ed ] (19)
[0169]
[0170] in, represents the selected comprehensive excellent sub-population, It represents the size of the offspring population. Finally, according to the evolution of these excellent sub-populations, it iterates to the next generation population.
[0171] Specifically, in step 4, in the designed interpretability strategy, the rule confidence distribution should be subject to the following constraints:
[0172]
[0173] Among them, C β Represents the constraints on the regular confidence distribution.
[0174] Specifically, in the step 5, an improved projection covariance matrix adaptive evolution strategy (P-CMA-ES) optimization algorithm is used to balance interpretability and accuracy;
[0175] First, determine the objective function:
[0176]
[0177] Where Ω represents the objective function based on dual statistics;
[0178] Then, optimization is performed, and the process optimization process includes:
[0179] Step 1: Initialization, let the initial parameters be in Represents the initial parameter vector to be optimized, parameter set The definition is as follows:
[0180]
[0181] Step 2: Sampling, in order to generate the population, you need to do the following:
[0182]
[0183] in, is (t+1)th The generated v th Solution vector, o represents the population mean, ε is the normal distribution, M t t th The covariance matrix of the generation population, γ is the step size;
[0184] Step 3: Add constraints and introduce interpretable constraints. This step is performed as follows:
[0185] (1) Inactivated rules are not allowed to participate in optimization;
[0186] [η k =1]→[θ1,…,θ L ,β 1,1 ,…,β N,L ,δ1,…,δ N ] retain (25)
[0187] (2) The optimized rules are consistent with expert judgment;
[0188] [β 1,1 ,…,β N,L ]~C β (26)
[0189] Among them, β 1,1 ,…,β N,L is the newly generated belief distribution;
[0190] Step 4: Projection. In order to make the solution vector satisfy the optimization constraints, this operation includes the following:
[0191]
[0192] Where d = 1, 2, ..., D represents the number of constrained variables, τ represents the number of constraints, and V = [1, 1, ..., 1] 1*L is the parameter vector;
[0193] Step 5: Select, based on the population selection method guided by dual statistics to select solutions and update subpopulations, calculate and rank the MSE value and Ed value for the population;
[0194]
[0195] Based on the overall ranking, the best subgroup is selected;
[0196] No balance =Sort[E c *No MSE +(1-E c )*No Ed ] (29)
[0197]
[0198] The optimal subgroup update is:
[0199]
[0200] Among them, q is the weight set;
[0201] Step 6: Adjust, perform adaptation operation to update the covariance matrix;
[0202]
[0203] Among them, e1, e2, e s ,e c is the learning rate;
[0204] Step 7: Interpretable step size convergence, the step size γ can be expressed as follows:
[0205]
[0206] 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. A reliable intelligent fault diagnosis method for ship power engine, characterized in that: The steps include: Step 1, obtaining ship power engine failure data; Step 2: Screen important features of the fault data, extract fault features from the data set obtained in step 1, sort the feature importance based on the Spearman correlation coefficient, and select features with higher importance; Step 3: Build an interpretable initial model and use the evidence reasoning algorithm to infer the confidence rule base; Step 4: Design a balanced optimization strategy for both accuracy and interpretability. Step 5: Optimize the model using the improved projected covariance matrix adaptive evolutionary strategy (P-CMA-ES) that can balance interpretability and accuracy. Step six: case test, conduct case test on the optimized model.
2. A reliable ship power engine intelligent fault diagnosis method according to claim 1, characterized in that: The data in step 1 are various monitoring indicators of the ship's power engine, including power, speed, torque, fuel consumption rate, thermal efficiency, emission standards, specific oil consumption rate, intake pressure and temperature, exhaust temperature, noise and vibration, reliability and maintenance interval, starting performance and lubricating oil consumption rate.
3. A reliable intelligent fault diagnosis method for ship power engine according to claim 1, characterized in that: In step 2, attribute selection is performed based on the absolute value of the Spearman correlation coefficient, which is calculated as follows: in, Indicates i th Input attribute data x i and the Spearman correlation coefficient between the result attribute data y, yes The absolute value of describes the magnitude of the correlation, p is the number of samples, and r*f(□) represents the rank of the attribute data, that is, the position of the data in descending order; Then, the relevance of the attributes is sorted in order from largest to smallest, and the process is expressed as: Among them, q is the number of indicators in the data set that have the potential to be used as input attributes, x1,x2,…,x q are the sorted attributes, and x1 has the highest correlation with the result.
4. A reliable intelligent fault diagnosis method for ship power engine according to claim 1, characterized in that: In the step three, an interpretable initial confidence rule base model is constructed. The specific process is to select two features with the highest importance from the important features selected in the above step two, and input them into the model to create a confidence rule base model.
5. A reliable intelligent fault diagnosis method for ship power engine according to claim 4, characterized in that: In step 3, the confidence rule base is inferred by the evidence reasoning algorithm, and the reasoning process includes: First, the matching degree between the input data and the reference value is calculated as follows: Among them, S(x i ) represents the input data x i The process of information transformation, Represents x i Belong to i th Attribute j th The matching degree of the reference value, J i represents the number of reference values, and Represents two adjacent reference values, After obtaining the matching degree of the data in the rule, the activation weight is calculated as follows: Among them, ω k Indicates k th The activation weight of the rule, δ i Indicates the relative weight of the attribute; Then, the activated rules are used as the reasoning basis, and the evidence reasoning algorithm is used to reason about the trust level of the result. The reasoning process is as follows: Among them, β n (n=1,2,…,N) represents n th Result Level D n degree of belief; The output form of the confidence rule base is as follows: in, A set of data representing attributes The belief distribution obtained; Finally, the inference results of the confidence rule base are calculated as follows: in, represents the final result of the confidence rule base, u(D n ) represents n th The utility value of the resulting level, Represents the prediction result of the confidence rule base.
6. A reliable intelligent fault diagnosis method for ship power engine according to claim 1, characterized in that: In step 4, in the designed interpretability strategy, inactive rules are not allowed to participate in the optimization, and the activation factor To determine the activation status of the rule, it is expressed as: the k =(w1,w2,…,w m ) (12) k=1,2,…,L,m=1,2,…,M (13) Each set of input data may activate some rules, when the activation weight w m ≠0, it means that the rule is activated by this set of input data, η k It means that after inputting M groups of data, k th The total activation state of the rule, if η k All 0, indicating k th If the rule is not activated, the activation factor Finally, keep the relevant parameters of the rule.
7. A reliable intelligent fault diagnosis method for ship power engine according to claim 6, characterized in that: In step 4, in the designed interpretability strategy, the interpretable step-length convergence strategy determines the initial step-length and controls the convergence speed by introducing the reliability of expert knowledge. The step-length convergence calculation is as follows: c t+1 =2*(1-E c )*(1-t / T) Ec+2 +e (14) Among them, γ t+1 represents the step size, T represents the maximum number of iterations, ε represents the minimum threshold of the step size, E c represents the credibility of expert knowledge, and its calculation method is as follows: Among them, E c represents the credibility of expert knowledge, y(i) represents the estimated value obtained from the knowledge provided by the expert through the confidence rule base expert system, y0(i) represents the true state of the ship power engine, F represents the trust factor, and U represents the size of the training data set.
8. A reliable intelligent fault diagnosis method for ship power engine according to claim 7, characterized in that: In step 4, in the designed interpretability strategy, the process of the population selection strategy based on dual statistics guidance includes: First, for v th The population calculates the mean square error (MSE) value and the Euclidean distance (Ed) value. The Ed value is the Euclidean distance between the population individuals and the expert knowledge, which indicates the degree of distance or proximity to the expert knowledge during the optimization process. Among them, MSE(Φ v:τ ) and Ed(Φ v:τ ) respectively represent the th MSE and Ed values calculated for subpopulations, τ represents the population size; Then, these groups were sorted and numbered according to their MSE values, with the smallest value listed first; No MSE =Sort[MSE(Φ v:τ )],v=1,2,…,τ (17) Similarly, the rankings are numbered according to the Ed value; No Ed =Sort[Ed(Φ v:τ )],v=1,2,…,τ (18) Calculate the overall ranking and then add E c As a balancing factor, and select a superior subgroup based on the balancing factor; No balance =Sort[E c *No MSE +(1-E c )*No Ed ] (19) in, represents the selected comprehensive excellent sub-population, It represents the size of the offspring population. Finally, according to the evolution of these excellent sub-populations, it iterates to the next generation population.
9. A reliable intelligent fault diagnosis method for ship power engine according to claim 1, characterized in that: In step 4, in the designed interpretability strategy, the rule confidence distribution should be subject to the following constraints: Among them, C β Represents the constraints on the regular confidence distribution.
10. A reliable intelligent fault diagnosis method for ship power engine according to claim 1, characterized in that: In the step 5, an improved projection covariance matrix adaptive evolution strategy (P-CMA-ES) optimization algorithm is used to balance interpretability and accuracy; First, determine the objective function: Where Ω represents the objective function based on dual statistics; Then, optimization is performed, and the process optimization process includes: Step 1: Initialization, let the initial parameters be in Represents the initial parameter vector to be optimized, parameter set The definition is as follows: Step 2: Sampling, in order to generate the population, you need to do the following: in, is (t+1) th The generated v th Solution vector, o represents the population mean, ε is the normal distribution, M t t th The covariance matrix of the generation population, γ is the step size; Step 3: Add constraints and introduce interpretable constraints. This step is performed as follows: (1) Inactivated rules are not allowed to participate in optimization; [or k =1]→[θ1,…,θ L ,b 1,1 ,…,b N,L ,δ1,…,δ N ] retain (25) (2) The optimized rules are consistent with expert judgment; [β 1,1 ,…,β N,L ]～C β (26) Among them, β 1,1 ,…,β N,L is the newly generated belief distribution; Step 4: Projection. In order to make the solution vector satisfy the optimization constraints, this operation includes the following: Where d = 1, 2, ..., D represents the number of constrained variables, τ represents the number of constraints, and V = [1, 1, ..., 1] 1*L is the parameter vector; Step 5: Select, based on the population selection method guided by dual statistics to select solutions and update subpopulations, calculate and rank the MSE value and Ed value for the population; Based on the overall ranking, the best subgroup is selected; No balance =Sort[E c *No MSE +(1-E c )*No Ed ] (29) The optimal subgroup update is: Among them, q is the weight set; Step 6: Adjust, perform adaptation operation to update the covariance matrix; Among them, e1, e2, e s ,e c is the learning rate; Step 7: Interpretable step size convergence, the step size γ can be expressed as follows: