Data-real fusion test evaluation method based on result fusion

Abstract

The invention discloses a data-real fusion test evaluation method based on result fusion. The method comprises the steps that a digital test result and a real installation test result are fused into a whole through a data-real test result data fusion module; 2, constructing a multi-target test demand model of the equipment through a data-real fusion test demand dynamic adaptation module; and step 3, executing the construction of an adaptive improved Bayesian neural network model through a number-real fusion test result evaluation module to realize the deep evaluation of a number-real fusion test result, and optimizing the performance of the Bayesian neural network model by using a back propagation mechanism and the iterative training of the model. And it is ensured that the Bayesian neural network model can dynamically adapt to the constructed multi-target test demand model. According to the method, data of the two models are fused through redundancy removal and vacancy filling, and the fusion degree of a multi-target test demand model and an index factor model is constructed by establishing a test demand dynamic adaptation module to adjust the real-number test result, so that the method has better fusion advantages of the two models.

Description

Technical Field

[0001] The present invention belongs to the fields of control engineering and computer science, and specifically relates to a digital-real fusion test evaluation method based on result fusion. Background Art

[0002] When conducting tests on systems or equipment such as complex industrial manufacturing production lines and special operation vehicles, the traditional method is generally actual installation testing. With the development of the times and technology, digital testing has emerged and developed rapidly. Compared with actual installation testing, digital testing has the advantages of comprehensive test scenarios, rich test conditions, and abundant test samples, but it has always been low in credibility due to technical reasons. Although the test conditions of actual installation testing are limited and the cost of iterative testing is high, it has the advantages of strong accuracy and high credibility of test results. The experimental results of the two have their own advantages and disadvantages. How to merge the two sets of test results data with limitations and differences so that it takes into account the advantages and disadvantages of both tests and has high credibility and rich test conditions is an inevitable problem in the near future. Summary of the invention

[0003] The technical problem to be solved by the present invention is: to provide a digital-real fusion test evaluation method based on result fusion, which includes a digital-real test result data fusion module, a digital-real fusion test demand dynamic adaptation module and a digital-real fusion test result evaluation module, which can better absorb the advantages of rich digital test result samples and high credibility of actual installation test results, weaken the inherent disadvantages of the two, and realize reliable and effective fusion evaluation of digital-real test results. The method first designs a digital-real test result data fusion module, which integrates digital testing and actual installation testing into one, and handles the problem of redundant and missing results of the two result data; secondly, a digital-real fusion test demand dynamic adaptation module is designed, which can dynamically adjust the target orientation of the test evaluation; finally, a digital-real fusion test result evaluation module is designed, and a digital-real fusion test result evaluation model is formed by adaptively improving the Bayesian neural network and iteratively training the model based on the back-propagation mechanism, so as to realize reliable and effective fusion evaluation of digital-real test results.

[0004] The present invention provides a digital-real fusion test evaluation method based on result fusion, comprising:

[0005] Step 1: The digital test results and the actual test results are integrated through the digital test result data fusion module, and the missing test data are supplemented and the redundant and inefficient data are eliminated;

[0006] Step 2: Build a multi-objective test requirement model for the equipment through the dynamic adaptation module of digital-real fusion test requirements, and then iteratively optimize and verify the model based on the key primary test requirement data under the ideal working conditions of the equipment and the applicable compliance requirement data under the general working conditions;

[0007] Step 3: Perform the following operations through the digital-real fusion test result evaluation module:

[0008] a. Construct an adaptively improved Bayesian neural network model to achieve in-depth evaluation of the logarithmic real fusion test results. The Bayesian neural network model includes an input layer, a hidden layer, and an output layer. Receive the output of the fusion of digital test results and actual test results in step 1:

[0009] ;

[0010] The hidden layer uses variational inference method to optimize network parameters. is the weight, is the bias, is the activation function:

[0011] ;

[0012] The output layer is responsible for generating the final evaluation results. is the weight, is the output bias, is the output activation function:

[0013] ;

[0014] b. Use the back-propagation mechanism and iterative training of the model to optimize the performance of the Bayesian neural network model, ensuring that the Bayesian neural network model can dynamically adapt to the multi-objective test requirement model constructed in step 2.

[0015] This method is suitable for evaluating the results of multi-stage and multi-round iterative testing of equipment or systems.

[0016] The beneficial effects of the present invention compared with the prior art are:

[0017] (1) When the equipment is in a complex system, such as one with multiple sub-components and requires multiple rounds of iterative testing, it takes a huge amount of manpower and material resources to conduct limited actual installation testing, while digital testing cannot provide good credibility. For these two test methods and the evaluation methods of test results, the traditional methods have a low level of understanding of digital testing and a poor understanding of the degree of integration and fusion of digital and real fusion testing. The present invention integrates the data of the two by removing redundancies and supplementing deficiencies, and then establishes a dynamic adaptation module for test requirements, constructs a multi-objective test requirement model and an indicator factor model to adjust the degree of fusion of digital and real test results, so that it has better fusion advantages of the two.

[0018] (2) During the test process, the test results of the sub-links / sub-steps are quite different from the test results of the equipment or system as a whole, and the mature conditions for assembly testing are often not able to be carried out on a large scale due to various reasons. Traditional methods generally establish complex feature models and combine historical data to predict the performance of the equipment or system to be tested. However, due to the low cost of actual installation test data, it is often difficult to obtain effective results. This method establishes a Bayesian complex model by conducting real tests on each sub-link separately, and dynamically adjusts the demand factor training iterative model as needed, which can greatly improve the grasp of the overall machine test evaluation and improve the accurate evaluation of the test results. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is a system structure diagram of a digital-real fusion test evaluation method based on result fusion of the present invention. DETAILED DESCRIPTION

[0020] The present invention is described in further detail below in conjunction with the accompanying drawings.

[0021] The present invention relates to a digital-real fusion test evaluation method based on result fusion, comprising the design of a digital-real test result data fusion module, the design of a digital-real fusion test demand dynamic adaptation module, and the design of a digital-real fusion test result evaluation module. The method can better absorb the advantages of rich digital test result samples and high credibility of actual installation test results, weaken the inherent disadvantages of the two, and realize reliable and effective fusion evaluation of digital-real test results.

[0022] Figure 1 FIG. 1 is a system structure diagram of a digital-real fusion test evaluation method based on result fusion according to the present invention. Figure 1 As shown, the system for executing the method includes a digital-real test result data fusion module 1, a digital-real fusion test requirement dynamic adaptation module 2, and a digital-real fusion test result evaluation module 3. Each module works in coordination to execute the method according to the present invention. Figure 1 The method is described. The method comprises the following steps:

[0023] Step 1: The digital test results and the actual test results are integrated through the digital test result fusion module, which can make up for the missing test data and eliminate redundant and inefficient data. The specific implementation is as follows:

[0024] a. In view of the fact that it is difficult to conduct destructive, extreme, and dangerous tests on equipment in practice, resulting in incomplete test result data, and that it is convenient to conduct tests in digital simulation but it is difficult to conduct tests that are completely realistic and highly reliable, resulting in redundant test result data but low reliability, a data redundancy and gap filling mechanism is designed based on the result data of digital tests and actual installation tests;

[0025] b. After multiple links (n) and multiple rounds (m) of installation and digital testing of the equipment, the following are obtained through the data redundancy removal and gap filling mechanism:

[0026] ,

[0027] ,

[0028] in, This is the actual test result. It is the result of the mth round of the nth stage in the actual installation test result. It is the digital test result.

[0029] Remove or reduce the weight of some redundant data with low credibility. Redundant data with low credibility refers to the part of the actual installation test that already has the result data of this link, and its digital test is far different from the actual installation test results. For some missing link result data in the actual installation test, use the digital test result data after multiple weighted averages to make up for it, which is:

[0030] ,

[0031] ,

[0032] ,

[0033] in, Represents the weight coefficient, which determines the degree of reduction of redundant data. Not incomplete, indicates that, therefore Right now There is redundancy, is a very small constant, Express Output value after redundancy removal. Incomplete, in Express Finally, the output value of the integrated digital test results and the actual test results is constructed by removing redundancy and filling in gaps. ,in Represents the matrix The i-th row of Represents the matrix The jth row of .

[0034] Step 2: Through the dynamic adaptation module of digital-real fusion test requirements, a multi-objective test requirement model for the equipment is constructed. Then, based on the key primary test requirements under ideal working conditions of the equipment and the applicable compliance requirements under widely used working conditions, the main station function indicators and auxiliary station operation indicators are analyzed, and the model is iteratively optimized and verified. The specific implementation is as follows:

[0035] a. Construct a multi-objective test demand model for equipment based on multiple indicators such as main combat function indicators and auxiliary combat operation indicators. The main combat function indicators include the equipment's load, cruising range, and maximum power, and the auxiliary combat operation indicators include the equipment's rapid replacement and repair capability of parts, the tolerance level for adverse working conditions, and the disturbance level of faults to the system. Finally, a multi-objective test demand model covering the equipment's main combat function indicators and auxiliary combat operation indicators is constructed; the model input includes load, cruising range, maximum power, rapid replacement and repair capability, the tolerance level for adverse working conditions and the disturbance level of faults, as well as the demand index of each indicator, which is defined as: , where TRI is the total input, ~ There are six indicators: ~ is the demand index of each indicator. Output , ~ They represent the index test indexes of maximum power, fuel efficiency, transmission system efficiency, maximum fuel storage capacity, equipment skeleton material, consistency level of key components, load resistance level of components, suspension reliability, robustness of electronic control system, and dustproof / waterproof / corrosion-resistant / high temperature / low temperature / sealing performance of the structure. A training layer is set up inside the model. Since there is a correlation between the indicators, the relationship between the indicators and the output ten indexes is multivariate. Therefore, the correlation between the six indicators is determined by the multivariate regression method, and then the test index of each indicator in TRO is generated. The formula is as follows:

[0036] ,

[0037] in, is the regression coefficient, which indicates the correlation between the output indicator and the input indicator. Represents the corresponding equipment performance test demand indicators, specifically The determination of the value requires model training. After collecting enough samples, the multivariate regression method is used to fit the sample data to solve the regression coefficient. , the least squares method is used during the training process to minimize the error between the predicted value and the actual value.

[0038] b. Design, train, iterate and verify the multi-objective test requirement model. First, use the analytic hierarchy process (AHP) to perform a hierarchical analysis of each indicator, then combine the Delphi method and the expert experience evaluation method to establish an initial indicator factor model, then use the game theory method, reverse position repeated evaluation, iteratively adjust the indicator factor model, and finally use the TOPSIS method to analyze the distance between the current model parameters and the negative ideal model to verify the model solution. Among them, the model parameters include all input indicators and Demand Index ,The negative ideal model refers to the solution corresponding to the worst value in each indicator.

[0039] In the process of finding the negative ideal model, the negative ideal solution of each indicator is first determined, that is, the worst value (most unfavorable value) of the indicator. For example, for the cruising range indicator, its worst value is the required minimum value. Then the negative ideal solutions of all indicators are listed to form a negative ideal solution vector: Similarly, we can get .in, is a negative ideal solution. Then the distances of various solutions from the ideal solution and the negative ideal solution are calculated using the Euclidean distance formula.

[0040] ,

[0041] ,

[0042] in, is the value of the p-th solution on the q-th index, and They are the ideal solution and the negative ideal solution respectively. Finally, the relative proximity is calculated To weigh the pros and cons of each option:

[0043] ,

[0044] The larger the value, the closer the solution is to the ideal and the better the solution is. The values ​​are sorted from largest to smallest.

[0045] Step 3: Perform the following operations through the digital-real fusion test result evaluation module:

[0046] a. Construct an adaptively improved Bayesian neural network model to achieve in-depth evaluation of the logarithmic real fusion test results. The Bayesian neural network model includes an input layer, a hidden layer, and an output layer. Receive output that combines digital test results with actual test results :

[0047] ;

[0048] The hidden layer uses variational inference method to optimize network parameters (weights and biases). is the weight, is the bias, is the activation function, Is the hidden layer output:

[0049] ,

[0050] The output layer is responsible for generating the final evaluation results. is the weight, is the output bias, is the output activation function:

[0051] .

[0052] b. Use the back-propagation mechanism and iterative training of the model to optimize the performance of the Bayesian neural network model, and ensure that the Bayesian neural network model can dynamically adapt to the multi-objective test requirement model constructed in step 2. Dynamic adaptation refers to the need to adjust the weights and bias parameters. After obtaining the optimized network parameters in step a, the suboptimal network parameters are stored, and when iterative training is performed to optimize the performance of the Bayesian neural network model, a small amount of computing power is invested in training the suboptimal network parameters. If the best training effect of the suboptimal parameters is better than the training effect of the optimal network parameters, the suboptimal network parameters are trained together. Similarly, the range of suboptimal network parameters is expanded in turn. The loss function L of the back-propagation mechanism is expressed as:

[0053] ,

[0054] in, is the true value, is the predicted value, and M is the number of samples. After calculating the gradient of the loss function with respect to the parameters of the Bayesian neural network model through the back propagation algorithm, the network weights are updated using the gradient, where is the learning rate:

[0055] ,

[0056] ,

[0057] Finally, combine the multi-objective test requirement model output solution constructed in step 2 and , regularly evaluate and update the factor weights of the multi-objective test requirement model:

[0058] ,

[0059] Where F is the current factor weight, k is the adjustment coefficient, is the degree of improvement in the evaluation results.

[0060] The multi-objective test requirement model constructed in step 2 not only provides the initial conditions for the iterative training in step 3, but also serves as an evaluation criterion during the training process to continuously promote the optimization of the model. This closed-loop system ensures the flexibility and adaptability of the digital-real fusion test evaluation method, enabling it to respond to different test requirements and environmental changes in real time.

[0061] In summary, the method includes: designing a digital-physical test result data fusion module, which integrates digital testing and actual installation testing into one, and can check and fill in the gaps in data sample missing problems of both digital testing and actual installation testing, and deal with the redundancy of data samples of both; designing a digital-physical fusion test demand dynamic adaptation module, which can dynamically adjust the target orientation of test evaluation based on the accuracy of actual installation testing and the comprehensiveness of digital test samples according to dynamic needs; a digital-physical fusion test result evaluation module, which integrates the aforementioned sample fusion data with test requirements, operates on input data through adaptively improved Bayesian neural networks, and forms a digital-physical fusion test result evaluation model, and then iteratively trains the model based on the back-propagation mechanism to achieve reliable and effective fusion evaluation of digital-physical test results.

[0062] The contents not described in detail in the specification of the present invention belong to the prior art known to the professional and technical personnel in this field.

[0063] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A digital-real fusion test evaluation method based on result fusion, characterized in that: include: Step 1: The digital test results and the actual test results are integrated through the digital test result data fusion module, and the missing test data are supplemented and the redundant and inefficient data are eliminated; Step 2: Build a multi-objective test demand model for equipment through the dynamic adaptation module of digital-real fusion test demand. Then, analyze the main station function indicators and auxiliary station operation indicators based on the key primary test requirements under ideal working conditions of the equipment and the applicable compliance requirements under general working conditions, and iterate and optimize and verify the model. Step 3: Perform the following operations through the digital-real fusion test result evaluation module: a. Construct an adaptively improved Bayesian neural network model to achieve in-depth evaluation of the logarithmic real fusion test results. The Bayesian neural network model includes an input layer, a hidden layer, and an output layer. Receive the output of the fusion of digital test results and actual test results in step 1: ； The hidden layer uses variational inference method to optimize network parameters. is the weight, is the bias, is the activation function: ； The output layer is responsible for generating the final evaluation results. is the weight, is the output bias, is the output activation function: ； b. Use the back-propagation mechanism and iterative training of the model to optimize the performance of the Bayesian neural network model, ensuring that the Bayesian neural network model can dynamically adapt to the multi-objective test requirement model constructed in step 2.

2. The digital-real fusion test evaluation method based on result fusion according to claim 1 is characterized in that: Step one includes: When the equipment is subjected to multi-stage and multi-round installation and digital testing, the number of stages is n and the number of rounds is m, and the installation test results are obtained. And digital test results ; Remove or reduce the weight of some redundant data with low credibility. The redundant data with low credibility refers to the part of the actual installation test that has the result data of this link, and the part of the digital test that is far different from the actual installation test result. The missing link result data in some actual installation tests are supplemented by the digital test result data after multiple weighted averages. The process is expressed as: ， ， ， in, Represents the weight coefficient, which determines the degree of reduction of redundant data. Suppose the actual test results in the experiment Not incomplete, Indicates that the digital test results Right now There is redundancy, is a very small constant, Indicates the test results of the numbers The output value after redundancy removal, Indicates the test results of the actual installation Fill in the missing output value; The final output of the fusion digital test results and the actual test results is: .

3. The digital-real fusion test evaluation method based on result fusion according to claim 2 is characterized in that: Step 2 includes: a. Construct a multi-objective test requirement model for equipment based on multiple indicators such as main combat function indicators and auxiliary combat operation indicators. The main combat function indicators include the equipment's load, cruising range, and maximum power. The auxiliary combat operation indicators include the equipment's ability to quickly replace and repair parts, the equipment's tolerance to harsh operating conditions, and the level of disturbance caused by faults to the system. Finally, a multi-objective test requirement model covering the equipment's main combat function indicators and auxiliary combat operation indicators is constructed; b. Design training, iteration and verification methods for the multi-objective test demand model, including: first, use the hierarchical analysis method to conduct a hierarchical analysis of each indicator, then combine the Delphi method and expert experience evaluation method to establish an initial indicator factor model, then use the game theory method to reverse the position and repeatedly evaluate, iteratively adjust the indicator factor model, and finally use the TOPSIS method to analyze the distance between the current parameters of the model and the negative ideal model to verify the advantages and disadvantages of the model solution.

4. The digital-real fusion test evaluation method based on result fusion according to claim 3 is characterized in that: Step three includes: The loss function L of the back-propagation mechanism is expressed as: ， in, is the true value, is the predicted value, M is the number of samples, and after calculating the gradient of the loss function with respect to the network parameters through the back-propagation mechanism, the gradient is used to update the network weights, where For the learning rate: ， ， Finally, combined with the multi-objective test requirement model constructed in step 2, the factor weights of the multi-objective test requirement model are regularly evaluated and updated: ， Where F is the current factor weight, k is the adjustment coefficient, is the degree of improvement in the evaluation results.

Images