Performance comprehensive evaluation method of data-driven fault diagnosis algorithm, medium and equipment

Abstract

The invention discloses a performance comprehensive evaluation method for a data-driven fault diagnosis algorithm, a medium and equipment, and the method comprises the steps: selecting a key performance parameter as an evaluation index, and constructing a comprehensive evaluation index system; calculating a combination weight by adopting a game theory thought; a fault diagnosis algorithm performance comprehensive evaluation model based on the combination of statistics and a neural network is constructed, an improved Topsis method is combined with a grey correlation analysis method to generate a comprehensive evaluation data set, a BP neural network used for genetic algorithm optimization scores a fault diagnosis algorithm, and quantitative performance evaluation is achieved. According to the method, the one-sidedness problem of single-index evaluation can be solved, the BP neural network model optimized through the genetic algorithm can be effectively used for subsequent algorithm performance evaluation, the defect that a traditional evaluation method is time-consuming and labor-consuming is overcome, and the efficiency is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of fault diagnosis algorithm performance evaluation, and specifically relates to a comprehensive performance evaluation method, medium and equipment for a data-driven fault diagnosis algorithm. Background Art

[0002] Nowadays, fault diagnosis algorithms are widely used in the industrial field. How to judge whether a fault diagnosis algorithm is feasible and how to compare and evaluate the performance of different fault diagnosis algorithms are problems that need to be solved. At present, the effectiveness evaluation of fault diagnosis algorithms mostly relies on a single indicator, which makes the evaluation process one-sided and too subjective. In addition, the same steps need to be repeated every time the performance evaluation is performed, which is time-consuming and labor-intensive, resulting in low efficiency.

[0003] Therefore, an evaluation method is needed to evaluate the performance of data-driven fault diagnosis algorithms. This method can quantitatively evaluate the algorithm and assist algorithm developers in debugging the fault diagnosis algorithm and further improve the algorithm performance. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention provides a comprehensive performance evaluation method, medium and device for a data-driven fault diagnosis algorithm, which solves the one-sided problem of traditional evaluation methods that rely on a single indicator.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] A comprehensive performance evaluation method for a data-driven fault diagnosis algorithm, characterized in that it comprises the following steps:

[0007] Select the key performance parameters of the fault diagnosis algorithm as evaluation indicators;

[0008] Based on the selected evaluation indicators, the game theory is used to calculate the combined weight composed of subjective weight and objective weight, which is used to evaluate the importance of the evaluation indicators of the fault diagnosis algorithm;

[0009] The Topsis method and grey correlation analysis method are used in combination with combined weights to construct a comprehensive evaluation data set. A comprehensive performance evaluation model of the fault diagnosis algorithm is constructed based on the BP neural network optimized by the genetic algorithm. The comprehensive evaluation data set is used to train and test the comprehensive performance evaluation model. The trained comprehensive performance evaluation model is used to evaluate the performance of the fault diagnosis algorithm.

[0010] To optimize the above technical solutions, the specific measures taken also include:

[0011] Furthermore, the evaluation indicators include cost-capability evaluation indicators and algorithm performance evaluation indicators; the cost-capability evaluation indicators include CPU running time, algorithm development cost and number of algorithm iterations; the algorithm performance evaluation indicators include accuracy, fault detection rate, fault false alarm rate and fault omission rate.

[0012] Furthermore, the calculation process of the subjective weight includes:

[0013] 1) Collect the relative importance scores of the evaluation indicators and construct the judgment matrix A according to the relative scores of each evaluation indicator:

[0014]

[0015] In the formula, A ij It represents the importance score of indicator i relative to indicator j, and n is the number of indicators;

[0016] 2) Calculate the subjective weight W1 according to the following formula:

[0017]

[0018] In the formula, λ max is the maximum eigenvalue of the judgment matrix A, λ is the eigenvalue of A, and I is the unit matrix.

[0019] Furthermore, the calculation process of the subjective weight also includes:

[0020] The reliability of the judgment matrix is ​​tested as follows:

[0021]

[0022] In the formula, CI is the consistency index value, and RI is the average random consistency index;

[0023] When CR<0.1, it indicates that the judgment matrix is ​​consistent and the consistency test passes; otherwise, the judgment matrix is ​​reconstructed until the consistency test passes.

[0024] Furthermore, the objective weight calculation process includes:

[0025] 1) Construct the following original decision matrix Y:

[0026]

[0027] Where Y aj is the value of each indicator, m is the number of algorithms, and n is the number of indicators;

[0028] 2) Use the “min-max method” to standardize the original decision matrix;

[0029] 3) Solve the ratio P of each evaluation index under each algorithm aj :

[0030]

[0031] Construct the corresponding ratio matrix P:

[0032]

[0033] Calculate the information entropy of each evaluation index:

[0034]

[0035] In the formula, e j is the information entropy of the jth indicator under all m algorithms, if p aj =0 or p aj =1, then define p aj 1np aj =0;

[0036] 4) Calculate the objective weight w of each evaluation index according to the following formula: i :

[0037]

[0038] In the formula, α j is the index difference coefficient, α j =1-e j ; The objective weight w of the calculated n indicators j Construct objective weight vector W2 = [w1, w2, ..., w n ].

[0039] Furthermore, the calculation of the combined weight composed of the subjective weight and the objective weight using the game theory idea specifically includes:

[0040] 1) Construct the basic weight vector set W3:

[0041] W3={W1,W2}=[W 31 , W 32 , …, W 3u , …, W 3v ] T ;

[0042] Where, 0 <u≤v∈R,W 3u The combined weight vector calculated for the linear combination of W1 and W2 is:

[0043]

[0044] Where η k is the weight coefficient, W kThe values ​​are W1 and W2;

[0045] 2) Find the weight vector set W3 that matches W k The weight vector W that minimizes the deviation * :

[0046]

[0047] The optimized first-order derivative condition is:

[0048]

[0049] Solve (η1, η2) and normalize it to get the normalized weight coefficient

[0050]

[0051] 3) Solve the combined weight vector W* as:

[0052]

[0053] In the formula, is the element in the combined weight vector W* after solving.

[0054] Furthermore, the Topsis method and the grey correlation analysis method are used to construct a comprehensive evaluation data set in combination with the combined weights, specifically including:

[0055] 1) By Get the normalized matrix R:

[0056]

[0057] In the formula, R aj Represents Y aj The normalized value of

[0058] 2) Construct absolute positive and negative ideal solutions, where the absolute positive ideal solution is The absolute negative ideal solution is

[0059] 3) Calculate the weighted Euclidean distance of each algorithm to the absolute positive and negative ideal solutions, where the weighted Euclidean distance to the absolute positive ideal solution is The weighted Euclidean distance to the absolute negative ideal solution is

[0060] 4) Calculate the grey correlation matrix:

[0061]

[0062] Where θ is the resolution coefficient, θ∈[0,1), L + With L- Represent the grey incidence matrix with absolute positive and negative ideal solutions respectively;

[0063] 5) Calculate the weighted grey correlation degree of each algorithm with the absolute positive and negative ideal solutions, where the weighted grey correlation degree with the absolute positive ideal solution is The weighted grey correlation degree with the absolute negative ideal solution is

[0064] 6) Combine weighted Euclidean distance with weighted grey correlation:

[0065]

[0066] Where β and γ are the values ​​reflecting the evaluator's preference for position and shape, β+γ=1, β, γ∈[0,1];

[0067] 7) Calculate the score:

[0068]

[0069] In the formula, Score a Represents the score of the ath algorithm, calculates the scores of m algorithms and constructs a comprehensive evaluation data set by combining the evaluation index values ​​of each algorithm.

[0070] Furthermore, the comprehensive evaluation data set is used to train and test the comprehensive performance evaluation model, specifically:

[0071] The obtained comprehensive evaluation data set is shuffled, and the processed comprehensive evaluation data set is divided into a training set and a test set, and the performance comprehensive evaluation model is trained and tested.

[0072] Accordingly, the present invention proposes a computer-readable storage medium storing a computer program, characterized in that the computer program enables a computer to execute the comprehensive performance evaluation method of the data-driven fault diagnosis algorithm as described above.

[0073] Accordingly, the present invention proposes an electronic device, characterized in that it includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the comprehensive performance evaluation method of the data-driven fault diagnosis algorithm as described above is implemented.

[0074] The beneficial effects of the present invention are as follows: the present invention establishes a multi-index evaluation system for a data-driven fault diagnosis algorithm to solve the problem of one-sidedness of a traditional single index; the game theory idea is used to realize the combination of subjective and objective weights, to solve the disadvantage of excessive subjectivity of the traditional subjective and objective weight combination method, and to realize the balance and coordination of different weighting methods. The present invention realizes quantitative scoring and sorting of the performance of the fault diagnosis algorithm by combining the improved Topsis method with the gray correlation analysis method, realizes effective processing of the reverse order problem existing in the traditional method, and provides a comprehensive evaluation data set for the neural network; at the same time, the BP neural network model optimized by the genetic algorithm is used to construct a comprehensive performance evaluation model of the fault diagnosis algorithm, and a database is established based on a large amount of historical data. When evaluating, only the data of various indicators need to be input to obtain the algorithm performance score, which helps to solve the disadvantages of the traditional method that is time-consuming and labor-intensive, and improves efficiency. In summary, the comprehensive performance evaluation method of the data-driven fault diagnosis algorithm proposed by the present invention solves the problems of strong subjectivity, time-consuming and labor-intensive, and low efficiency of the traditional fault diagnosis algorithm performance evaluation method, realizes the intelligent evaluation of the performance of the fault diagnosis algorithm, and is conducive to the developers to realize the debugging and optimization of the fault diagnosis algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Figure 1 This is a flow chart of the comprehensive performance evaluation method of the data-driven fault diagnosis algorithm of the present invention.

[0076] Figure 2 The figure is a flow chart of the method for generating a comprehensive evaluation data set according to the present invention.

[0077] Figure 3 This is the flow chart of the genetic algorithm optimizing the BP neural network of the present invention.

[0078] Figure 4 This is a comparison chart of the errors of the BP neural network training set optimized by the genetic algorithm of the present invention.

[0079] Figure 5 This is a comparison chart of the error of the BP neural network test set optimized by the genetic algorithm of the present invention. DETAILED DESCRIPTION

[0080] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0081] In one embodiment, the present invention proposes a comprehensive performance evaluation method for a data-driven fault diagnosis algorithm, such as Figure 1 As shown, the specific steps include:

[0082] S1. Select key performance parameters as evaluation indicators and build a comprehensive evaluation indicator system.

[0083] Evaluation indicators are considered in two categories: cost-capability evaluation indicators and algorithm performance evaluation indicators; cost-capability evaluation indicators include but are not limited to CPU running time, algorithm development cost, and number of algorithm iterations; fault diagnosis algorithm performance indicators include but are not limited to accuracy, fault detection rate, fault false alarm rate, and fault missed alarm rate.

[0084] S2. Use game theory to calculate the combination weights.

[0085] S21. Calculate subjective weights using the analytic hierarchy process.

[0086] 1) Collect relevant experts and scholars to score the relative importance of indicators according to the 1-9 scale method, and construct a judgment matrix A=(A ij ) n×n :

[0087]

[0088] In the formula, A ij It represents the importance score of indicator i relative to indicator j, and n is the number of indicators.

[0089] In this embodiment, the relative importance of the indicators is scored according to the 1-9 scale method, as shown in Table 1. According to S1, the number of indicators n=4 is selected, which are accuracy, fault detection rate, fault false alarm rate, and CPU running time. The judgment matrix is ​​constructed according to the relative score value of each indicator as follows:

[0090]

[0091] Table 11-9 Scaling method

[0092] Scale meaning 1 Indicates that the two indicators are equally important. 3 Indicates that one indicator is slightly more important than the other indicator. 5 Indicates that one indicator is significantly more important than the other. 7 Indicates that one indicator is more important than the other. 9 Indicates that one indicator is extremely important than the other indicator. 2、4、6、8 The median of the two adjacent judgments above reciprocal If the scale of A compared to B is 3, then B is 1 / 3 compared to A.

[0093] 2) Hierarchical single sorting: The corresponding indicators at the same level, i.e., the corresponding evaluation indicators selected from S1, are sorted according to the weights calculated based on the relative importance of the corresponding target at the previous level, i.e., the performance of the fault diagnosis algorithm.

[0094] The subjective weight W1 is calculated according to the following formula:

[0095]

[0096] Among them, A is the judgment matrix, λ max is the maximum eigenvalue, that is, the maximum value of the eigenvalue λ obtained by eigendecomposing the matrix A, I is the unit matrix, W1 is λ max The corresponding eigenvector and the vector after normalization.

[0097] In this embodiment, the obtained subjective weights are shown in Table 2, that is, the weight vector W1 = [0.5471, 0.2408, 0.1618, 0.0504].

[0098] Table 2 Subjective weights

[0099] Accuracy Fault detection rate Fault false alarm rate cpu running time Weight 0.5471 0.2408 0.1618 0.0504

[0100] 3) Whether the judgment matrix composed of the relative importance scores of the indicators by relevant experts and scholars is accurate and reliable needs to be tested for consistency:

[0101]

[0102] Among them, CI is the consistency index value, and RI is the average random consistency index; when the consistency ratio CR<0.1, it indicates that the judgment matrix is ​​consistent; otherwise, the experts re-judge and reconstruct the judgment matrix until the consistency test passes.

[0103] In this embodiment, the RI values ​​are shown in Table 3.

[0104] Table 3 RI values

[0105] n 1 2 3 4 5 6 7 8 RI 0 0 0.52 0.89 1.12 1.26 1.36 1.41

[0106] The calculated results are CI=0.0493, CR=0.0554, which satisfies CR<0.1, indicating that the judgment matrix is ​​consistent and the obtained weight vector W1 can be adopted.

[0107] S22. Calculate the objective weight using the entropy weight method.

[0108] 1) Original decision matrix construction:

[0109]

[0110] Among them, Y aj is the value of each indicator, m is the number of algorithms, and n is the number of indicators.

[0111] 2) Use the “min-max method” to standardize the original decision matrix:

[0112] Processing of positive indicators:

[0113] Dealing with negative indicators:

[0114] In this embodiment, j=1, 2, 3, 4.

[0115] 3) Calculation of indicator information entropy:

[0116] First, solve the ratio of each indicator under each algorithm:

[0117]

[0118] Get the corresponding ratio matrix P:

[0119]

[0120] Secondly, calculate the information entropy of each indicator:

[0121]

[0122] Among them, e j is the information entropy of the jth indicator under all m algorithms, if p aj =0 or p aj =1, then define p aj lnp aj =0.

[0123] In this embodiment, m = 50. The information entropy of each indicator is calculated to be e1 = 0.9378, e2 = 0.9481, e3 = 0.9440, and e4 = 0.9465. According to formula α j =1-e j It can be calculated that the coefficient of difference of each indicator is α1=0.0622, α2=0.0519, α3=0.0560, α4=0.0535.

[0124] 4) Objective weight calculation:

[0125]

[0126] Among them, α j is the index difference coefficient, α j =1-e j ; The objective weight w of the calculated n indicators j Construct objective weight vector W2 = [w1, w2, ..., w n ].

[0127] In this embodiment, the calculated objective weights are shown in Table 4, that is, the weight vector W2 = [0.2782, 0.2321, 0.2505, 0.2393].

[0128] Table 4 Objective weights

[0129] Accuracy Fault detection rate Fault false alarm rate cpu running time Weight 0.2782 0.2321 0.2505 0.2393

[0130] S23. Use game theory to calculate the combination weight.

[0131] 1) Construct a basic weight vector set:

[0132] W3={W1,W2}=[W 31 , W 32 ,…,W3u,…,W 3v ] T ;

[0133] Among them, W 3u is the combined weight vector calculated by the linear combination of W1 and W2, 0 <u≤v∈R;

[0134] Any linear combination of these two weight vectors can form a combined weight vector:

[0135]

[0136] Among them, η k is the weight coefficient, W k The values ​​are W1 and W2, representing the weights calculated by S21 and S22 respectively.

[0137] 2) Solving the weight coefficient:

[0138] Find the most reasonable weight vector W in the weight vector set W3 * , is to make W * With W k Minimize the deviation of:

[0139]

[0140] The optimized first-order derivative condition is:

[0141]

[0142] Solve to get (η1, η2), and normalize it:

[0143]

[0144] In this embodiment, the weight coefficients are η1=1.1303, η2=-0.1303.

[0145] 3) Combined weight solution:

[0146]

[0147] in, The combined weight vector W after solving * The elements in .

[0148] In this embodiment, the calculated combined weights are shown in Table 5, that is, the combined weights W * =[0.5821,0.2419,O.1502,0.0258].

[0149] Table 5 Combination weights

[0150] Accuracy Fault detection rate Fault false alarm rate cpu running time Weight 0.5821 0.2419 0.1502 0.0258

[0151] S3. Use statistics and neural networks to construct a comprehensive performance evaluation model for fault diagnosis algorithms.

[0152] S31, using improved Topsis method and grey correlation analysis method to construct comprehensive evaluation data set, such as Figure 2 shown.

[0153] 1) By Get the normalized matrix R:

[0154]

[0155] Among them, R aj Represents Y aj The normalized value of .

[0156] 2) Construct absolute positive and negative ideal solutions:

[0157] The absolute positive ideal solution and the absolute negative ideal solution can be set by the evaluator himself according to the properties of each indicator, or can be determined by relevant experts based on experience.

[0158] Absolutely positive ideal solution:

[0159] Absolute negative ideal solution:

[0160] In this embodiment, the calculated absolute positive ideal solution V + =(1,1,1,1), absolute negative ideal solution V - =(0,0,0,0).

[0161] 3) Calculate the weighted Euclidean distance of each algorithm to the absolute positive and negative ideal solutions:

[0162] Weighted Euclidean distance to the absolutely positive ideal solution:

[0163] Weighted Euclidean distance to the absolute negative ideal solution:

[0164] In this embodiment, the weighted Euclidean distances between 50 fault diagnosis algorithms and the absolute positive ideal solution are calculated as:

[0165]

[0166] The weighted Euclidean distance between 50 fault diagnosis algorithms and the absolute negative ideal solution is calculated as:

[0167]

[0168] 4) Calculate the grey correlation matrix:

[0169]

[0170]

[0171] Among them, θ is the resolution coefficient, θ∈[0,1), which is used to reduce the influence of extreme values ​​on its calculation. Generally, θ=0.5 is more appropriate. + With L - Represent the grey incidence matrices with absolute positive and negative ideal solutions respectively.

[0172] 5) Calculate the weighted grey correlation between each algorithm and the absolute positive and negative ideal solutions:

[0173] Weighted grey correlation degree with the absolute positive ideal solution:

[0174] Weighted grey correlation degree with absolute negative ideal solution:

[0175] In this embodiment, the weighted grey correlation between 50 fault diagnosis algorithms and the absolute positive ideal solution is calculated as:

[0176]

[0177] The weighted grey correlation between 50 fault diagnosis algorithms and the absolute negative ideal solution is calculated as:

[0178]

[0179] 6) Combination of weighted Euclidean distance and weighted grey correlation:

[0180]

[0181] Among them, β and γ reflect the evaluator's preference for position and shape, and the evaluator can set their values ​​according to his or her own preference, but it is necessary to satisfy β+γ=1, β, γ∈[0,1]. In this embodiment, β=γ=0.5 is selected.

[0182] 7) Calculate the score:

[0183]

[0184] Among them, Score a Represents the score of the ath algorithm, calculates the scores of m algorithms and combines the index values ​​of each algorithm to construct a comprehensive evaluation data set.

[0185] In this embodiment, the performance scores of 50 fault diagnosis algorithms are calculated as follows:

[0186]

[0187] S32, based on the BP neural network optimized by genetic algorithm, the performance comprehensive evaluation model of data-driven fault diagnosis algorithm is constructed, such as Figure 3 shown.

[0188] 1) Division of training set and test set:

[0189] The comprehensive evaluation data set obtained in S31 is processed in a disordered manner, and the processed data set is divided into a training set and a test set, which are applied to the training and testing of the BP neural network optimized by the genetic algorithm. In this embodiment, 80% of the data in the processed data set is divided into a training set, and 20% of the data is divided into a test set.

[0190] 2) Genetic algorithm optimization of BP neural network:

[0191] By using genetic algorithms to optimize the optimal initial weights and thresholds required by the BP neural network, the problems of slow convergence and falling into local optimal solutions that may occur in the BP neural network are solved. The initial BP neural network weight threshold length is encoded as the initial population. After the initial population is generated, it evolves from generation to generation and generates an approximate solution. In the new offspring, new individuals are selected according to the size of individual fitness, and the crossover and mutation of genetic operators are completed at the same time, thereby generating a new population. This process continues from generation to generation until the optimal individual obtained at the end is decoded, that is, the optimal approximate solution, which is used as the optimal initial weight and threshold required by the BP neural network.

[0192] The chromosome fitness value selects the training error of the neural network, and the formula is as follows:

[0193]

[0194] Among them, K is the coefficient, N is the number of nodes in the output layer, and y i is the expected result of the jth output layer node, is the actual result output by the i-th output node.

[0195] The selection operator uses the roulette algorithm:

[0196]

[0197] Among them, F i is the fitness value of the chromosome individual, k is the coefficient, and N is the total number of chromosomes.

[0198] The crossover operator uses uniform crossover:

[0199]

[0200] Among them, b is a random number in the interval [0, 1] and is the crossover coefficient.

[0201] The mutation operator adopts non-uniform mutation and introduces a mutation rate reduction function, so that the mutation rate gradually decreases with the increase of iteration times.

[0202] 3) Training and testing of BP neural network optimized by genetic algorithm:

[0203] The indicator parameters in the training set data are used as the input of the BP neural network, and the algorithm score is used as the output of the BP neural network. Therefore, the number of neural units in the input layer is set to 4, the number of neural units in the output layer is set to 1, and the hidden layer is set to 1. Set the relevant training parameters, set the training batch to 1000, the training error target to 0.00001, and the learning rate to 0.0001. Compare the output of the BP neural network with the true value, and calculate the error between the predicted value of the algorithm score and the true value of the algorithm score. When the error is less than the set training error target, the network training is completed, such as Figure 4 As shown. Input the index parameters in the test set data into the trained BP neural network to obtain the corresponding score. Compare it with the actual score, as shown Figure 5 As shown in the figure, the error between the predicted value of the test set and the true value is small. The effectiveness of the BP neural network is tested and the performance evaluation model of the fault diagnosis algorithm is obtained.

[0204] In another embodiment, the present invention provides a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the comprehensive performance evaluation method of the data-driven fault diagnosis algorithm as described in the first embodiment.

[0205] In another embodiment, the present invention proposes an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the comprehensive performance evaluation method of the data-driven fault diagnosis algorithm as described in Example 1 is implemented.

[0206] In the embodiments disclosed in the present application, the computer storage medium can be a tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. The computer storage medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the above. More specific examples of computer storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above.

[0207] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0208] The above are only preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should be regarded as the protection scope of the present invention.

Claims

1. A comprehensive performance evaluation method for a data-driven fault diagnosis algorithm, characterized in that: The following steps are involved: Select the key performance parameters of the fault diagnosis algorithm as evaluation indicators; Based on the selected evaluation indicators, the game theory is used to calculate the combined weight composed of subjective weight and objective weight, which is used to evaluate the importance of the evaluation indicators of the fault diagnosis algorithm; The Topsis method and grey correlation analysis method are used in combination with combined weights to construct a comprehensive evaluation data set. A comprehensive performance evaluation model of the fault diagnosis algorithm is constructed based on the BP neural network optimized by the genetic algorithm. The comprehensive evaluation data set is used to train and test the comprehensive performance evaluation model. The trained comprehensive performance evaluation model is used to evaluate the performance of the fault diagnosis algorithm.

2. A comprehensive performance evaluation method for a data-driven fault diagnosis algorithm as claimed in claim 1, characterized in that: The evaluation indicators include cost capability evaluation indicators and algorithm performance evaluation indicators; the cost capability evaluation indicators include CPU running time, algorithm development cost and number of algorithm iterations; the algorithm performance evaluation indicators include accuracy, fault detection rate, fault false alarm rate and fault omission rate.

3. The comprehensive performance evaluation method of a data-driven fault diagnosis algorithm according to claim 1, characterized in that: The calculation process of the subjective weight includes: 1) Collect the relative importance scores of the evaluation indicators and construct the judgment matrix A according to the relative scores of each evaluation indicator: In the formula, A ij It represents the importance score of indicator i relative to indicator j, and n is the number of indicators; 2) Calculate the subjective weight W1 according to the following formula: In the formula, λ max is the maximum eigenvalue of the judgment matrix A, λ is the eigenvalue of A, and I is the unit matrix.

4. A comprehensive performance evaluation method for a data-driven fault diagnosis algorithm as claimed in claim 3, characterized in that: The calculation process of the subjective weight also includes: The reliability of the judgment matrix is ​​tested as follows: In the formula, CI is the consistency index value, and RI is the average random consistency index; When CR<0.1, it indicates that the judgment matrix is ​​consistent and the consistency test passes; otherwise, the judgment matrix is ​​reconstructed until the consistency test passes.

5. The method for comprehensive performance evaluation of a data-driven fault diagnosis algorithm according to claim 3, characterized in that: The calculation process of the objective weight includes: 1) Construct the following original decision matrix Y: Where Y aj is the value of each indicator, m is the number of algorithms, and n is the number of indicators; 2) Use the "min-max method" to standardize the original decision matrix; 3) Solve the ratio P of each evaluation index under each algorithm aj : Construct the corresponding ratio matrix P: Calculate the information entropy of each evaluation index: In the formula, e j is the information entropy of the jth indicator under all m algorithms, if p aj =0 or p aj =1, then define p aj ln aj =0; 4) Calculate the objective weight w of each evaluation index according to the following formula: j : In the formula, α j is the index difference coefficient, α j =1-e j ; The objective weight w of the calculated n indicators j Construct objective weight vector W2 = [w1, w2, ..., w n ].

6. A comprehensive performance evaluation method for a data-driven fault diagnosis algorithm as claimed in claim 5, characterized in that: The method of using game theory to calculate the combined weight composed of subjective weight and objective weight specifically includes: 1) Construct the basic weight vector set W3: W3={W1,W2}=[W 31 ,W 32 ,…,W 3u ,...,W 3v ] T ; Where, 0 <u≤v∈R,W 3u The combined weight vector calculated for the linear combination of W1 and W2 is: Where η k is the weight coefficient, W k The values ​​are W1 and W2; 2) Find the weight vector set W3 that matches W k The weight vector W that minimizes the deviation * : The optimized first-order derivative condition is: Solve (η1, η2) and normalize it to get the normalized weight coefficient 3) Solve the combined weight vector W* as: In the formula, is the element in the combined weight vector W* after solution.

7. A comprehensive performance evaluation method for a data-driven fault diagnosis algorithm as claimed in claim 6, characterized in that: The Topsis method and the grey correlation analysis method are used to construct a comprehensive evaluation data set in combination with the combined weights, specifically including: 1) By Get the normalized matrix R: In the formula, R aj Represents Y aj The normalized value of 2) Construct absolute positive and negative ideal solutions, where the absolute positive ideal solution is The absolute negative ideal solution is 3) Calculate the weighted Euclidean distance of each algorithm to the absolute positive and negative ideal solutions, where the weighted Euclidean distance to the absolute positive ideal solution is The weighted Euclidean distance to the absolute negative ideal solution is 4) Calculate the grey correlation matrix: Where θ is the resolution coefficient, θ∈[0,1), L + With L - Represent the grey incidence matrix with absolute positive and negative ideal solutions respectively; 5) Calculate the weighted grey correlation degree of each algorithm with the absolute positive and negative ideal solutions, where the weighted grey correlation degree with the absolute positive ideal solution is The weighted grey correlation degree with the absolute negative ideal solution is 6) Combine weighted Euclidean distance with weighted grey correlation: Where β and γ are the values ​​reflecting the evaluator's preference for position and shape, β+γ=1, β, γ∈[0,1]; 7) Calculate the score: In the formula, Score a Represents the score of the ath algorithm, calculates the scores of m algorithms and constructs a comprehensive evaluation data set by combining the evaluation index values ​​of each algorithm.

8. The method for comprehensive performance evaluation of a data-driven fault diagnosis algorithm according to claim 1, characterized in that: The comprehensive evaluation data set is used to train and test the comprehensive performance evaluation model, specifically: The obtained comprehensive evaluation data set is shuffled, and the processed comprehensive evaluation data set is divided into a training set and a test set, and the performance comprehensive evaluation model is trained and tested.

9. A computer-readable storage medium storing a computer program, characterized in that: The computer program enables a computer to execute the comprehensive performance evaluation method of a data-driven fault diagnosis algorithm as described in any one of claims 1-8.

10. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the comprehensive performance evaluation method for the data-driven fault diagnosis algorithm as described in any one of claims 1 to 8 is implemented.

Images