A device evaluation method

Abstract

The present invention provides a device evaluation method, which mainly includes the training and testing of a probabilistic neural network and the specific evaluation process. Multiple different probabilistic neural networks are trained and tested based on the index data of the equipment. 85% of the data is selected as the training sample set and 15% as the test sample set. The output of the probabilistic neural network is transformed into evidence by improving the evidence theory, and a reliability factor is designed to correct each probabilistic neural network. The basic probability assignment decision method is used to determine the final evaluation result of the equipment to be evaluated. The present invention uses a probabilistic neural network to solve the evaluation result, avoiding the influence of subjective factors. Compared with other types of neural networks, the learning process of the probabilistic neural network is simple, the convergence speed is fast, and the output layer is replaced with a softmax layer to perform classification output in the form of probability, ensuring the intuitiveness of the output result. After each network is corrected by the evidence reliability factor, the reliability of the evaluation result is greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of performance evaluation of lasers, and specifically provides a device evaluation method. Background Art

[0002] Equipment evaluation is a process of analyzing, processing, and comparing relevant data of various equipment to assist in making decisions. The above data includes various test data, design reviews, software and hardware tests, modeling and simulation, usage (including maintenance, storage, etc.) data, and historical data. Equipment evaluation runs through the entire life cycle of equipment and is a necessary basis for assessing equipment performance, evaluating equipment quality, and determining equipment development and procurement strategies.

[0003] Currently, the existing device evaluation methods mainly include:

[0004] 1. Expert scoring method: An evaluation method that emerged earlier and is widely used. Its advantage is that it is simple and intuitive to use, and can make quantitative evaluations in the absence of sufficient statistical data and original materials. The disadvantage is that it lacks theoretical and systematicness, and it is difficult to ensure the objectivity and accuracy of the evaluation results.

[0005] 2. Analytic hierarchy process: A practical multi-criteria decision-making method that represents a complex problem as an ordered hierarchical structure and ranks the advantages and disadvantages of decision-making methods through human judgment. The advantage is that it requires less data and takes less time for decision-making. The disadvantage is that when solving problems with numerous factors and large scales, it will lead to decision-making results that significantly deviate from objective laws.

[0006] 3. Fuzzy comprehensive evaluation method: Applying the principle of fuzzy relation synthesis to comprehensively evaluate the membership level of the evaluated object from multiple factors. The advantage is that it can comprehensively evaluate objects involving fuzzy factors. The disadvantage is that this method cannot solve the problem of duplicate evaluation information caused by the correlation between evaluation indicators, and there is no systematic and effective method for determining the membership function.

[0007] 4. Data envelopment analysis method: Evaluating the relative efficiency and benefits of the same type of decision-making units based on input and output data. The advantage is that the evaluation is completely based on the objective information of indicator data and can handle large-scale systems with multiple inputs and outputs. The disadvantage is that it can only show the relative development indicators of evaluation units and cannot show the actual development level.

[0008] 5. Grey comprehensive evaluation method: A comprehensive evaluation method that combines qualitative and quantitative methods. The advantages are simple calculation process, no need for data normalization, and strong reliability. The disadvantage is that it requires sample data to have time series characteristics, and it only discriminates the advantages and disadvantages of the evaluation object and cannot reflect the absolute level.

[0009] In addition, the artificial neural network evaluation method has been widely pursued in recent years. Currently, it mainly evaluates indicators by training a BP neural network. However, the evaluation method using a BP neural network has problems such as slow learning speed, easy occurrence of local minimization, and non-intuitive output results. When faced with a large amount of training data, it is prone to non-convergence, resulting in a very slow network training process. Moreover, using only one network for judgment may lead to large result deviations due to network performance issues.

[0010] Therefore, there is an urgent need for an equipment evaluation method with high evaluation accuracy, short training time, and high scalability. Summary of the Invention

[0011] To solve the above problems, the present invention organically combines a probabilistic neural network and an improved evidence theory, fully leveraging the advantages of both and overcoming their disadvantages, effectively improving the accuracy of the evaluation results and shortening the training time and evaluation processing time.

[0012] An equipment evaluation method provided by the present invention includes:

[0013] Training, testing, and correcting a probabilistic neural network, including the following steps:

[0014] Establish an index system and a grade system for equipment evaluation;

[0015] Collect sample data according to the index system, and randomly select 85% of the sample data as the training sample set and 15% as the test sample set;

[0016] Select no less than 3 probabilistic neural networks, with different activation functions, expansion speeds, and learning algorithms for each probabilistic neural network, and replace the output layer of the probabilistic neural network with a softmax layer. The number of nodes in the softmax layer is the same as the number of grades in the grade system;

[0017] Complete the training of the probabilistic neural network using the training sample set;

[0018] Complete the testing of the probabilistic neural network using the test sample set. Convert the output of the probabilistic neural network into evidence through the improved evidence theory, convert the basic probability assignment of the evidence to the Pignistic probability of each single element, and select the element with the largest value as the evaluation result;

[0019] Establish an indicator function to judge whether the evaluation result conforms to the true result, and calculate the reliability factor of the evidence output by each probabilistic neural network according to the indicator function to form a correction set;

[0020] Evaluate the equipment, including the following steps:

[0021] The equipment data is divided according to the index system to obtain equipment index data, and the equipment index data is respectively input into the probabilistic neural network, and the evidence output by different probabilistic neural networks is corrected according to the correction set;

[0022] The corrected evidence is combined pairwise using the Dempster combination rule to obtain a combination result;

[0023] Based on the combined result, the final evaluation result of the equipment to be evaluated is determined using the basic probability assignment decision method.

[0024] Preferably, the establishment process of the index system and the grade system is as follows:

[0025] Set the index system as U = {T1, T2,..., T M}, where T i (i = 1, 2,..., M) represents the i-th index in the index system, and M represents the total number of indexes;

[0026] The recognition framework of the evaluation result of the equipment is Θ = {A1, A2,..., A N}), where the evaluation result of the equipment has N grades, namely V1, V2,..., V N , and the proposition A i (i = 1, 2,..., N) represents that the grade of the equipment to be evaluated currently is V i .

[0027] Preferably, the number of nodes in the input layer of the probabilistic neural network is the same as the number of indexes in the index system, the number of nodes in the pattern layer of the probabilistic neural network is the same as the number of samples in the test sample set, and the number of nodes in the summation layer of the probabilistic neural network is the same as the number of grades in the grade system.

[0028] Preferably, after the output layer of the probabilistic neural network is replaced by a softmax layer, the calculation formula of the output result is as follows:

[0029]

[0030] Among them, the number of probabilistic neural networks C i (i = 1, 2,..., L) is L, and the output result of the summation layer of C i (i = 1, 2,..., L) is (S i1 , S i2 ,..., S iN ), S ij is the output result of the j-th (j = 1, 2,..., N) node of the summation layer, and the output result of the softmax layer is (R i1 , R i2 ,..., R iN ), R ijis the output result of the j-th (j = 1, 2,..., N) node of the softmax layer.

[0031] Preferably, the evidence generated by the output transformation of the probability neural network is:

[0032] E i =(m i (A1), m i (A2),..., m i (A N ), m i );

[0033] Among them, m i (A j ) represents the basic probability assignment of the focal element A j in the evidence Ei.

[0034] Preferably, the calculation process of the reliability factor of the evidence is as follows:

[0035] Set the test sample set as O = {o1, o2,..., o T}), O k (k = 1, 2,..., T) represents any test sample in the test sample set, and the reliability factor of the evidence corresponding to the probability neural network C i (i = 1, 2,..., L) is

[0036] Calculate the Pignistic probability function value BetP{o k}(k = 1, 2,..., T) corresponding to the test sample O i for the focal element A j in the evidence E k}(A j ) as:

[0037]

[0038] Among them, X represents a set;

[0039] Set the indicator function as δ jk (j = 1, 2,..., N; k = 1, 2,..., T), when the evaluation result is consistent with the true result, δ jk = 1; when the evaluation result is inconsistent with the true result, δ jk = 0;

[0040] Calculate the reliability factor i of the evidence corresponding to the probability neural network C as follows:

[0041]

[0042] Among them, α i is a variable, p jk = BetP{o k}(A j ).

[0043] Preferably, the evidence output by different probability neural networks is corrected according to the correction set, and then the corrected evidence E i the basic probability assignment m' j of the focal element A i (A j ) is:

[0044]

[0045] Preferably, the basic probability assignments m' i of the focal element A j in the corrected evidence E i (A j ) are combined pairwise to obtain the combination result as F = (m f (A1), m f (A2),..., m f (A N ), m f (Θ)), where m f (A j ) represents the combination result of the basic probability assignment m' i of the focal element A j in the corrected evidence E i (A j ).

[0046] Preferably, the number of probability neural networks is set to an odd number not less than three.

[0047] Compared with the prior art, the present invention can achieve the following beneficial effects:

[0048] 1. The present invention uses a probabilistic neural network to solve the evaluation results, avoiding the influence of subjective factors. Compared with other types of neural networks, the advantages of the probabilistic neural network are as follows: the network learning process is simple, the convergence speed is fast, and new category pattern samples can be added and reduced without long-term training and learning; its connection weights of hidden units can be directly obtained according to the sample features and expected outputs without repeatedly training the network; it has good fault tolerance and strong pattern classification ability; its pattern layer adopts a radial basis nonlinear mapping function, considering the influence generated by the intersection of different category samples, and the obtained decision boundary between different categories satisfies the Bayesian optimal solution, and the output layer is replaced by a softmax layer to perform classification output in the form of probability, ensuring the intuitiveness of the output results.

[0049] 2. Due to the diversity and complexity of the evaluated equipment, there are many equipment evaluation indicators and a large amount of data. If a single neural network is used for evaluation, the network structure is complex and difficult to determine. In addition, it is difficult to obtain the samples for network training, and there are inevitably contradictions and randomness, resulting in poor generalization ability of the neural network. To solve the problem of low accuracy of a single neural network, the evidence theory is introduced to accumulate evidence to narrow the hypothesis set, improving the accuracy of the evaluation results.

[0050] 3. The present invention takes each independent probabilistic neural network as the evidence source of the evidence theory. At the same time, the concept of evidence reliability factor is introduced to measure the reliability of the evidence given by different probabilistic neural networks. The output values of the probabilistic neural network are corrected by the evidence reliability factor and then used as the basic credibility of the propositions on the final result identification framework. Through the re-fusion of the evidence theory, the information of the evidence source can be fully utilized, greatly improving the recognition accuracy and eliminating the incompleteness and ambiguity of the information contained in a single evidence source. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a flowchart of the probabilistic neural network training and equipment evaluation provided according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0052] In the following, embodiments of the present invention will be described with reference to the drawings. In the following description, the same modules are denoted by the same reference numerals. In the case of the same reference numerals, their names and functions are also the same. Therefore, their detailed descriptions will not be repeated.

[0053] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and do not constitute a limitation to the present invention.

[0054] Figure 1Illustrates the training of a probabilistic neural network and the evaluation of equipment according to an embodiment of the present invention.

[0055] As Figure 1 shown, the present invention provides an equipment evaluation method, which is mainly applied to the evaluation process of optical equipment, such as lasers. Its evaluation indicators include service life, actual power, beam quality, etc. The specific process is as follows:

[0056] S1. Train, test, and correct the probabilistic neural network, including:

[0057] S101. Set the index space of the equipment to be evaluated as U = {T1, T2, …, T M}, that is, the index system, where T i (i = 1, 2, …, M) represents the i-th index in the index system U. For example, laser energy consumption, laser life, beam quality, the difference between actual power and theoretical power, etc. M represents the total number of indexes, and i is a subscript parameter without practical significance.

[0058] Set the evaluation result recognition framework of the equipment as Θ = {A1, A2, …, A N}, where the evaluation result of the equipment is N levels, that is, V1, V2, …, V N , and the proposition A i (i = 1, 2, …, N) represents that the current equipment to be evaluated is at the level of V i , and the proposition A i (i = 1, 2, …, N) can specifically express the evaluation level of the equipment. For example, by outputting the proposition through a display or an output system, the evaluation level can be clearly defined.

[0059] S102. In the existing neural network training and testing process, usually 70% of the sample data is selected as the training sample set and 30% as the test sample set. The embodiment of the present invention combines the characteristics of the probabilistic neural network to optimize the allocation ratio, selects 85% of the sample data as the training sample set and 15% as the test sample set. Through statistical testing, this ratio effectively improves the accuracy of the probabilistic neural network on the premise of ensuring the network test accuracy. In addition, a comparison network and a comparator can be designed to sort the data in the training sample set, and the data that seriously deviates from the data mode can be eliminated. The deviation threshold can be set to 50%. This process can eliminate problem data and avoid the problem that the accuracy of the network decreases due to the influence of problem data on individual output nodes during network training.

[0060] S103. Select no less than 3 probabilistic neural networks, and the activation function, expansion speed, and learning algorithm of each probabilistic neural network are different. In this embodiment, the number of probabilistic neural networks is determined as L, and L represents any integer greater than or equal to 3.

[0061] The topological structures of the probability neural networks are as follows:

[0062] Input layer: 1 layer, with the number of nodes being M, which is the same as the number of indicators in the indicator system.

[0063] Pattern layer: 1 layer, with the number of nodes being the same as the number of samples in the test sample set. The pattern layer adopts a radial basis non - linear mapping function.

[0064] Summation layer: 1 layer, with the number of nodes being N, which is the same as the number of levels in the level system.

[0065] Output layer: The output layer of the probability neural network is replaced by a softmax layer. The number of nodes in the softmax layer is the same as the number of levels in the level system. The softmax layer performs a conversion on the output result, and the output result is presented in the form of probability, ensuring that the output result of the probability neural network is visually displayed in the form of probability. Taking a node in the original output layer as an example, this node is a competitive neuron. It receives various probability density patterns output from the summation layer and selects a neuron with the maximum posterior probability density from the estimated probability densities of each classification pattern as the output.

[0066] After the output layer of the probability neural network is replaced by a softmax layer, the calculation formula for the output result is as follows:

[0067]

[0068] Among them, the number of probability neural networks C i (i = 1, 2,..., L) is L, and the output result of the summation layer of C i (i = 1, 2,..., L) is (S i1 , S i2 ,..., S iN ), S ij is the output result of the j - th (j = 1, 2,..., N) node of the summation layer, and the output result of the softmax layer is (R i1 , R i2 ,..., R iN ), R ij is the output result of the j - th (j = 1, 2,..., N) node of the softmax layer.

[0069] S104. Use the training sample set to complete the individual training of L probability neural networks. The training processes of each probability neural network do not interfere with each other. The number of probability neural networks can be increased or decreased at will, but the total number of probability neural networks should be no less than 3 in the end.

[0070] S105. Complete the test of the probabilistic neural network using the test sample set. Transform the output of the probabilistic neural network into evidence by improving the evidence theory, convert the basic probability assignment of the evidence to the Pignistic probability of each single element, and select the element with the largest value as the evaluation result. Among them, the evidence generated by transforming the output of the probabilistic neural network is:

[0071] E i =(m i (A1),m i (A2),...,m i (A N ),m i (Θ));

[0072] Among them, m i (A j ) represents the basic probability assignment of the focal element A i in the evidence E j .

[0073] It should be noted that a proposition is a set of elements under the recognition framework. A proposition represents a set with only one element, while a focal element is a set of elements under the evidence framework. A focal element can represent a set with one or more elements.

[0074] Test each probabilistic neural network separately, calculate the reliability factor to form a correction set, and correct each probabilistic neural network. The specific process is as follows:

[0075] Set the test sample set as O = {o1, o2,..., o T},O k (k = 1, 2,..., T) represents any test sample in the test sample set. The reliability factor of the evidence corresponding to the probabilistic neural network C i (i = 1, 2,..., L) is

[0076] Input the equipment evaluation index data of the test sample O k (k = 1, 2,..., T) into the probabilistic neural network C i (i = 1, 2,…, L). According to the output results (R i (i = 1, 2,…, L) of the probabilistic neural network C i1 , R i2 ,…, R iN ), generate the evidence E i . Make a decision on the evidence E i (i = 1, 2,…, L) according to the Pignistic probability function to obtain the test sample O k(k = 1, 2, ..., T) evaluation results, calculate the test sample O k (k = 1, 2, ..., T) corresponding evidence E i in the focal element A j Pignistic probability function value BetP{o k}(A j ) is:

[0077]

[0078] Among them, X represents a set;

[0079] And select the element with the largest value as the evaluation result of the test sample O k (k = 1, 2, ..., T).

[0080] Set the indicator function as δ jk (j = 1, 2, ..., N; k = 1, 2, ..., T), judge whether the evaluation result conforms to the true result. When the evaluation result is consistent with the true result, δ jk = 1; when the evaluation result is inconsistent with the true result, δ jk = 0; among them, k, j are similar to i, all of which are subscript parameters and have no actual meaning.

[0081] After all test samples complete the above processing, calculate the reliability factor of the evidence corresponding to the probabilistic neural network C i (i = 1, 2, ..., L) as follows: As follows:

[0082]

[0083] Among them, α i is a variable, p jk = BetP{o k}(A j ).

[0084] Form a correction set with the obtained reliability factors for recording. When a new network is added, supplement the reliability factor of the new network into the correction set.

[0085] S2. Use the probabilistic neural network that has completed the above training, testing, and correction to evaluate the equipment, including:

[0086] S201. Divide the equipment data according to the index system to obtain equipment index data. Input the equipment index data into L probabilistic neural networks respectively, and correct the evidence output by different probabilistic neural networks according to the reliability factors in the correction set. Then the corrected evidence E i in the focal element A jThe basic probability assignment m′ i (A j ) is as follows:

[0087]

[0088] As a preferred embodiment, the number L of probability neural networks is set to an odd number not less than three. During the evaluation process of the equipment, the reliability factor with the lowest corrected central value is assigned 0, and the corresponding probability neural network does not participate in this evaluation, further reducing the evaluation deviation of the probability neural network for the equipment evaluation index.

[0089] S202. Use the Dempster combination rule to pairwise combine the corrected evidence to obtain a combination result. The specific process is as follows:

[0090] Set the trust functions of the evaluation result recognition framework Θ = {A1, A2,..., A N} to be BEL1 and BEL2, m′1 and m'2 are the corresponding corrected basic probability assignments, B1, B2,..., B k and C1, C2,..., C r are their corresponding focal elements. Then the Dempster combination rule is as follows:

[0091]

[0092] Among them, H is the combined data set.

[0093] The combination result obtained by combining each evidence using the Dempster combination rule is F = (m f (A1), m f (A2), …, m f (A N ), m f (Θ)).

[0094] Based on the combination result, use the basic probability assignment decision method to determine the final evaluation result G of the equipment to be evaluated.

[0095] Let the final equipment evaluation result be V i , that is, the decision result is that the proposition A i (i = 1, 2, …, N) is true. Then it should satisfy the following conditions:

[0096] (1) A i (i = 1, 2, …, N) has the largest basic probability assignment, and this assignment is greater than the set probability threshold.

[0097] (2) The uncertainty m f (Θ) is less than the set precision threshold.

[0098] (3)A i (i = 1, 2, …, N) has a difference greater than the proposition output threshold from the basic probability assignments of other propositions.

[0099] The above probability threshold, precision threshold, and proposition output threshold all need to be set in combination with the specific probability neural network and precision requirements.

[0100] Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

[0101] The above specific implementation manners of the present invention do not constitute a limitation on the protection scope of the present invention. Any other corresponding changes and deformations made according to the technical concept of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. An equipment evaluation method, characterized in that, Specifically applied to the performance evaluation of lasers, including: Carrying out the training, testing and correction of the probabilistic neural network, including the following steps: Establishing an index system and a grade system for equipment evaluation, and the establishment processes of the index system and the grade system are as follows: Set the index system as , where represents the th index in the index system, M represents the total number of indices, and the indices include: laser energy consumption, laser life, beam quality, and the difference between the actual power and the theoretical power; The evaluation result recognition framework of the equipment is , where the evaluation result of the equipment is levels, namely , , …, , and the proposition indicates that the level of the equipment to be evaluated currently is ; Collecting sample data according to the index system, and randomly selecting 85% of the sample data as the training sample set and 15% as the test sample set; Selecting no less than 3 probabilistic neural networks, with different activation functions, expansion speeds and learning algorithms for each probabilistic neural network, and replacing the output layer of each probabilistic neural network with a softmax layer, and the number of nodes in the softmax layer is the same as the number of grades in the grade system; Completing the training of the probabilistic neural network using the training sample set; Completing the testing of the probabilistic neural network using the test sample set, converting the output of the probabilistic neural network into evidence through the improved evidence theory, converting the basic probability assignment of the evidence to the Pignistic probability of each single element, and selecting the element with the largest value as the evaluation result. After replacing the output layer of the probabilistic neural network with a softmax layer, the calculation formula of the output result is as follows: ； Among them, the probabilistic neural network has a number of L , , The output result of the summation layer of is where is the output result of the th node of the summation layer, , the output result of the softmax layer is where is the output result of the th node of the softmax layer; Establishing an indicator function to judge whether the evaluation result conforms to the true result, and calculating the reliability factor of the evidence output by each probabilistic neural network according to the indicator function to form a correction set; Evaluating the equipment, including the following steps: Dividing the equipment data according to the index system to obtain equipment index data, inputting the equipment index data into the probabilistic neural network respectively, and correcting the evidence output by different probabilistic neural networks according to the correction set; Using the Dempster combination rule to combine the corrected evidence pairwise to obtain a combination result; Based on the combination result, using the basic probability assignment decision method to determine the final evaluation result of the equipment to be evaluated.

2. The equipment evaluation method according to claim 1, characterized in that The number of nodes in the input layer of the probabilistic neural network is the same as the number of indicators in the index system, the number of nodes in the pattern layer of the probabilistic neural network is the same as the number of samples in the test sample set, the pattern layer adopts a radial basis nonlinear mapping function, and the number of nodes in the summation layer of the probabilistic neural network is the same as the number of grades in the grade system.

3. The equipment evaluation method according to claim 2, wherein The evidence generated by the output conversion of the probabilistic neural network is: ； Among them, , denotes the evidence and the basic probability assignment of the focal element in it.

4. The equipment evaluation method according to claim 3, characterized in that, The calculation process of the reliability factor of the evidence is as follows: Set the test sample set as , represents any test sample in the test sample set, and the reliability factor of the evidence corresponding to the probability neural network is ; Calculate the test sample The corresponding evidence The focal element The Pignistic probability function value of Is as follows: ； Among them, X represents a set; Set the indication function as , when the evaluation result is consistent with the true result, ; when the evaluation result is inconsistent with the true result, ; Calculate the probability neural network The reliability factor of the corresponding evidence is as follows: ； Among them, is a variable, , .

5. The equipment evaluation method according to claim 4, characterized in that, The evidence output by different probability neural networks is corrected according to the correction set, and the corrected evidence in the focal element basic probability assignment is as follows: 。 6. The equipment evaluation method according to claim 5, wherein, Use the Dempster combination rule to combine the basic probability assignments of the focal elements in the revised evidence pairwise to obtain the combination result as in the middle basic probability assignment of the focal element pairwise combination, and the combination result is , where represents the basic probability assignment of the focal element in the revised evidence basic probability assignment of combination result.

7. The equipment evaluation method according to claim 1, wherein The number of the probabilistic neural networks is set to an odd number not less than three.

Images