Method and system for evaluating equipment performance with reduced effects of data noise and imbalance

Abstract

The application provides a weapon equipment performance evaluation method and system for reducing the influence of data noise and imbalance, and belongs to the technical field of weapon equipment. The method comprises the following steps: generating a test result data matrix corresponding to each capability element in the weapon equipment; performing missing data statistics on the test result data of each performance index in each test of each capability element of the weapon equipment performance; generating a Gaussian noise matrix corresponding to each capability element in the weapon equipment; generating a performance index matrix corresponding to each capability element in the weapon equipment; performing estimation on each capability in the weapon equipment; setting a self-encoder and a regressor; performing self-encoder training to obtain the encoder output result in the trained self-encoder; and performing regressor training to obtain the weapon equipment performance evaluation result output by the trained regressor. The application can reduce the influence of data noise and data imbalance on the weapon equipment performance evaluation result, and ensures the accuracy of the evaluation result.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of weapon equipment, and particularly relates to an equipment performance evaluation method and system for reducing the influence of data noise and imbalance. BACKGROUND

[0002] Equipment performance test and evaluation is a necessary means for checking equipment performance, evaluating equipment quality, and determining equipment development and procurement strategies, and is an important approach to improving equipment combat effectiveness.

[0003] Traditional equipment performance evaluation methods need to construct an equipment performance index system, configure reasonable evaluation calculation methods, and use performance test result data to realize equipment performance evaluation calculation. Equipment performance evaluation requires clear and comprehensive understanding of the mission, development standards, development process, and equipment performance test and evaluation methods of the equipment, and puts forward high technical theory and business knowledge requirements for evaluation implementers.

[0004] In recent years, the development of machine learning, data mining and other fields provides a technical basis for automated equipment performance test and evaluation. Equipment performance evaluation is essentially a numerical prediction and calculation. By selecting a reasonable machine learning algorithm, mining the potential rules in the test result data, and using the historical evaluation results of the traditional evaluation method as training data to train a regression prediction model, the equipment performance can be automatically evaluated, and the evaluation efficiency can be improved.

[0005] The quality of model training data often determines the accuracy of the equipment performance evaluation model. Training data usually comes from equipment performance tests, including real equipment tests and simulation tests. Real equipment tests are tests conducted on real equipment in a realistic battlefield environment, while simulation tests are dynamic tests conducted by simulation systems on potential or objectively existing equipment performance. Real equipment tests are easily disturbed by environmental factors, while simulation data lacks sufficient authenticity. Therefore, test results are prone to errors, and there is certain data noise in the collected data, which will interfere with the model training process. At the same time, since the equipment entering the performance test phase has usually been tested and optimized many times, the test results of the equipment tend to be positive in most cases, and there are more positive evaluations in the historical evaluation results, lacking sufficient negative evaluation results, resulting in imbalance in the evaluation data. Data noise and imbalance will affect the model training process, making it impossible to accurately mine the real data rules, resulting in unreliable model evaluation results.

[0006] The autoencoder can learn the potential features of data by encoding and decoding to reconstruct the original data. The real data is input into the autoencoder after adding noise, and the autoencoder needs to learn more stable data representation by removing noise to reconstruct the input data during training. The learned data representation is essentially the potential features of the denoised data.

[0007] As a network structure algorithm model, the multilayer perceptron has strong data fitting ability, and usually uses mean square error as the loss function. The standard mean square error does not distinguish between positive and negative samples, and when the positive and negative samples of the training data are unbalanced, the prediction result of the trained model will be biased towards the side with more positive and negative samples. The prediction target of the regression model is a continuous value. Reasonable estimation of the distribution of the training target data before training and correction of the loss function using the estimation result can reduce the influence of data imbalance to a certain extent. SUMMARY

[0008] One of the purposes of the present application is to provide a weapon equipment performance evaluation method for reducing the influence of data noise and imbalance, which can reduce the influence of data noise and data imbalance on the evaluation result of weapon equipment performance, ensure the accuracy of the evaluation result, and improve the evaluation effect.

[0009] The second purpose of the present application is to provide a weapon equipment performance evaluation system for reducing the influence of data noise and imbalance.

[0010] In order to achieve one of the above purposes, the present application adopts the following technical solutions:

[0011] A weapon equipment performance evaluation method for reducing the influence of data noise and imbalance, the equipment performance evaluation method comprising the following steps:

[0012] Step S1, collecting the test result data of each performance index in each test of each capability element of weapon equipment performance to generate a test result data matrix corresponding to each capability element in the weapon equipment;

[0013] Step S2, performing missing data statistics on the test result data of each performance index in each test of each capability element of the weapon equipment performance to generate a missing data simulation matrix corresponding to each capability element in the weapon equipment;

[0014] Step S3, generating a Gaussian noise matrix corresponding to each capability element in the weapon equipment according to the collection error of each test result data;

[0015] Step S4, generating a performance index matrix corresponding to each capability element in the weapon equipment according to the test result data matrix, the Gaussian noise matrix and the missing data simulation matrix corresponding to each capability element in the weapon equipment;

[0016] Step S5, estimating each capability in the weapon equipment according to each row of test result data in the test result data matrix corresponding to each capability element in the weapon equipment, to generate a performance estimation result matrix corresponding to each capability element in the weapon equipment;

[0017] Step S6, setting an auto-encoder and a regressor, and connecting an encoder in the auto-encoder with the regressor;

[0018] Step S7, taking the transpose matrix of the performance index matrix corresponding to each capability element in the weapon equipment as an input of the auto-encoder, and performing auto-encoder training according to the test result data matrix corresponding to each capability element in the weapon equipment, to obtain an output result of the trained encoder;

[0019] Step S8, taking the output result of the trained encoder as an input of the regressor, and performing regressor training according to the performance estimation result matrix corresponding to each capability element in the weapon equipment, to obtain a weapon equipment performance evaluation result output by the trained regressor.

[0020] Further, in the step S2, the specific process of generating the missing data simulation matrix corresponding to each capability element in the weapon equipment includes:

[0021] Step S21, performing missing data statistics on the test result data of each performance index in each test of each capability element in the weapon equipment, to calculate the missing times of each performance index in each capability element;

[0022] Step S22, calculating the product of the number of performance indexes in each capability element and the number of tests;

[0023] Step S23, calculating the ratio of the missing times of each performance index in each capability element to the product result;

[0024] Step S24, randomly generating a missing data simulation matrix according to the ratio of the missing times of each performance index in each capability element to the product result.

[0025] Further, in the step S4, the specific process of generating the performance index matrix corresponding to each capability element in the weapon equipment includes:

[0026] Step S41, adding the test result data matrix corresponding to each capability element in the weapon equipment and the Gaussian noise matrix;

[0027] Step S42, performing Hadamard product on the addition result and the corresponding missing data simulation matrix to obtain a performance index matrix corresponding to each capability element in the weapon equipment;

[0028] The number of rows of the test result data matrix and the Gaussian noise matrix is the number of tests, and the number of columns is the number of performance indexes;

[0029] Each element value of the test result data matrix is test result data, and the element value of each column in the Gaussian noise matrix is a random variable subject to Gaussian distribution.

[0030] Further, in the step S7, the specific process of training the autoencoder includes:

[0031] Step S71, setting autoencoder parameters;

[0032] Step S72, sequentially performing weighting, biasing and normalization processing on the transposed matrix of the performance index matrix corresponding to each capability element in the weapon equipment to obtain an aggregated information matrix of each capability element in the weapon equipment;

[0033] Step S73, performing information recovery processing on the aggregated information matrix of each capability element in the weapon equipment;

[0034] Step S74, judging whether the information recovery processing result converges to the transposed matrix of the corresponding test result data matrix, if yes, stopping training and taking the aggregated information matrix of each capability element in the weapon equipment as the input of the regressor, and if no, adjusting the autoencoder parameters and returning to step S72.

[0035] Further, in the step S8, the specific process of training the regressor includes:

[0036] Step S81, setting regressor parameters;

[0037] Step S82, sequentially performing weighting, biasing and normalization processing on the received aggregated information matrix of each capability element in the weapon equipment;

[0038] Step S83, judging whether the normalization processing result converges to the corresponding performance estimation result matrix, if yes, stopping training and taking the normalization processing result as the weapon equipment performance evaluation result, and if no, adjusting the regressor parameters and returning to step S82.

[0039] In order to achieve the second purpose, the application adopts the following technical scheme:

[0040] A weapon equipment performance evaluation system for reducing the influence of data noise and imbalance, the equipment performance evaluation system comprising:

[0041] A collection module for collecting test result data of each performance index in each capability element of weapon equipment performance in each test to generate a test result data matrix corresponding to each capability element of the weapon equipment;

[0042] A first generation module for performing missing data statistics on the test result data of each performance index in each capability element of weapon equipment performance in each test to generate a missing data simulation matrix corresponding to each capability element of the weapon equipment;

[0043] A second generation module for generating a Gaussian noise matrix corresponding to each capability element of the weapon equipment according to the collection error of each test result data;

[0044] A third generation module for generating a performance index matrix corresponding to each capability element of the weapon equipment according to the test result data matrix, the Gaussian noise matrix and the missing data simulation matrix corresponding to each capability element of the weapon equipment;

[0045] An estimation module for estimating each capability of the weapon equipment according to each row of test result data in the test result data matrix corresponding to each capability element of the weapon equipment to generate a performance estimation result matrix corresponding to each capability element of the weapon equipment;

[0046] A setting module for setting an auto-encoder and a regressor and connecting an encoder in the auto-encoder with the regressor;

[0047] An auto-encoder training module for performing auto-encoder training according to the test result data matrix corresponding to each capability element of the weapon equipment, taking the transpose matrix of the performance index matrix corresponding to each capability element of the weapon equipment as the input of the auto-encoder to obtain the output result of the trained encoder;

[0048] A regressor training module for performing regressor training according to the performance estimation result matrix corresponding to each capability element of the weapon equipment, taking the output result of the trained encoder as the input of the regressor to obtain the weapon equipment performance evaluation result output by the trained regressor.

[0049] Further, the first generation module comprises:

[0050] The missing data statistical submodule is configured to statistically analyze the test result data of each performance index in each capability element of the weapon equipment performance in each test to calculate the missing times of each performance index in each capability element.

[0051] The first calculation submodule is configured to calculate the product of the number of performance indexes in each capability element and the number of tests.

[0052] The second calculation submodule is configured to calculate the ratio of the missing times of each performance index in each capability element to the product result.

[0053] The random generation submodule is configured to randomly generate a missing data simulation matrix according to the ratio of the missing times of each performance index in each capability element to the product result.

[0054] Further, the third generation module comprises:

[0055] The adding submodule is configured to add the test result data matrix corresponding to each capability element in the weapon equipment and the Gaussian noise matrix.

[0056] The Hadamard product submodule is configured to perform Hadamard product on the addition result and the corresponding missing data simulation matrix to obtain the performance index matrix corresponding to each capability element in the weapon equipment.

[0057] The test result data matrix and the Gaussian noise matrix have the number of tests as the number of rows and the number of performance indexes as the number of columns.

[0058] Each element value of the test result data matrix is test result data, and each column of the Gaussian noise matrix has random variables subject to Gaussian distribution as element values.

[0059] Further, the autoencoder training module comprises:

[0060] The first setting submodule is configured to set the autoencoder parameters.

[0061] The first normalization processing submodule is configured to sequentially perform weighting, biasing and normalization processing on the transposed matrix of the performance index matrix corresponding to each capability element in the weapon equipment to obtain the aggregation information matrix of each capability element in the weapon equipment.

[0062] The information recovery processing submodule is configured to perform information recovery processing on the aggregation information matrix of each capability element in the weapon equipment.

[0063] The first judging submodule is used for judging whether the information recovery processing result converges to the transpose matrix of the corresponding test result data matrix, if yes, stopping training, and taking the aggregated information matrix of each capability element in the weapon equipment as the input of the regressor; if no, adjusting the auto-encoder parameter, and transmitting to the first normalization processing submodule.

[0064] Further, the regressor training module comprises:

[0065] The second setting submodule is used for setting the regressor parameter.

[0066] The second normalization processing submodule is used for sequentially performing weighting, biasing and normalization processing on the received aggregated information matrix of each capability element in the weapon equipment.

[0067] The second judging submodule is used for judging whether the normalization processing result converges to the corresponding performance estimation result matrix, if yes, stopping training, and taking the normalization processing result as the weapon equipment performance evaluation result; if no, adjusting the regressor parameter, and transmitting to the second normalization processing submodule.

[0068] In summary, the technical scheme provided by the present application has the following technical effects:

[0069] The application generates a test result data matrix corresponding to each capability element of the weapon equipment through the test result data of each performance index in each test of each capability element of the weapon equipment performance; obtains a missing data simulation matrix corresponding to each capability element of the weapon equipment through the data missing condition of the test result data matrix; obtains a Gaussian noise matrix corresponding to each capability element of the weapon equipment through the collection error of the test result data of each performance index in each test of each capability element of the weapon equipment performance; and obtains a performance index matrix containing data noise and data imbalance by using the test result data matrix, the missing data simulation matrix and the Gaussian noise matrix corresponding to each capability element of the weapon equipment; under the condition that the regressor parameters remain unchanged, the test result data matrix and the performance index matrix corresponding to each capability element of the weapon equipment are used for autoencoder training to ensure the denoising effect of the autoencoder; the performance prediction result matrix corresponding to each capability element of the weapon equipment is obtained by using the traditional evaluation method through the test result data of each performance index in each test of each capability element of the weapon equipment performance; the regressor is trained by using the output result of the encoder in the trained autoencoder and the performance prediction result matrix to obtain a trained regressor; then the trained autoencoder and the trained regressor are used to realize the evaluation of the weapon equipment performance; the application can obtain the latent features of the test data through the autoencoder, reconstruct the original data, achieve the data denoising effect, correct the loss function of the regression model through the probability density of the sample, adjust the gradient contribution of the high and low frequency samples in the training, and finally realize the reduction of the influence of data noise and data imbalance on the accuracy of the evaluation model, ensure the accuracy of the evaluation result, reduce the technical threshold of completing the evaluation task, and improve the evaluation efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0070] In order to more clearly illustrate the specific embodiments of the present application or the technical solutions in the prior art, the drawings needed in the specific embodiments or the prior art description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0071] Figure 1 The flow chart of the weapon equipment performance evaluation method for reducing the influence of data noise and imbalance according to the embodiment of the present application. DETAILED DESCRIPTION

[0072] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0073] This embodiment presents a method for evaluating weapon and equipment performance to reduce the impact of data noise and imbalance, referencing... Figure 1 The equipment performance evaluation method includes the following steps:

[0074] S1. Collect the test result data of each performance index in each capability element of the weapon system in each test, so as to generate the test result data matrix corresponding to each capability element in the weapon system.

[0075] The test results data from multiple actual and simulation tests were collected, and the index data generated from multiple evaluations were stitched together row by row to generate a test result data matrix X for each performance index (such as maximum speed, turning speed, and maximum acceleration index) in each equipment capability (such as mobility, firepower, and communication capabilities). 0 :

[0076]

[0077] Among them, X 0 A matrix of test results data with m rows and n columns; x ij Let represent the test results data for the j-th performance index in the i-th test, where i = 1, 2, ..., m, j = 1, 2, ..., n, m is the number of tests, and n is the number of performance indices. For example, if mobility includes the maximum speed index, turning speed index, and maximum acceleration index, then n = 3.

[0078] The experimental result data matrix X in this embodiment 0 This contains m rows of test result data, representing the test results collected from m actual or simulation tests. Each row corresponds to the test results of all performance indicators collected after one test. Test Result Data Matrix X 0 This matrix contains n columns of data, representing the test results for n performance indicators. Each column corresponds to the test results for a specific performance indicator across all tests. For example, it represents all test results for the maximum speed after m tests. The test result data matrix X 0 element x in ij This represents the test results data for the j-th performance index in the i-th test.

[0079] Test result data matrix X 0 Each row of test result data in the test result data matrix X corresponds to one test, and the test result data after the test can be used as a characterization of the capability of one aspect of the equipment. For example, one test collects performance index results such as maximum speed, turning speed, and maximum acceleration, which together characterize the mobility capability of the weapon equipment. A total of m tests are performed, and the entire test result data matrix X 0 The mobility capability of the weapon equipment is characterized m times.

[0080] S2, missing data statistics are performed on the test result data of each performance index in each capability element of the performance of the weapon equipment in each test to generate a missing data simulation matrix corresponding to each capability element in the weapon equipment.

[0081] The missing data simulation matrix MI of this embodiment is an m-row-by-n-column matrix, and the elements in the missing data simulation matrix MI are composed of 0 and 1 values randomly generated according to a data missing rate MissRate. When the element MI ij in the missing data simulation matrix MI is 0, it indicates that data loss occurs when the jth performance index (such as maximum speed) is collected in the ith test; when the element MI ij in the missing data simulation matrix MI is 1, it indicates that no data loss occurs when the jth performance index (such as maximum speed) is collected in the ith test. The data missing rate MissRate can be manually set according to expert experience or can be obtained according to historical test result data statistics of the weapon equipment, that is, MissRate = number of missing data in historical test result data / total number of index data in historical test result data.

[0082] In summary, in this embodiment, the specific process of generating the missing data simulation matrix corresponding to each capability element in the weapon equipment includes:

[0083] Step S21, missing data statistics are performed on the test result data of each performance index in each capability element of the performance of the weapon equipment in each test to calculate the missing number of each performance index in each capability element;

[0084] Step S22, the product of the number of performance indexes in each capability element and the number of tests is calculated;

[0085] Step S23, the ratio of the missing number of each performance index in each capability element to the product result is calculated;

[0086] Step S24, the missing data simulation matrix is randomly generated according to the ratio of the missing number of each performance index in each capability element to the product result.

[0087] S3, generating a Gaussian noise matrix corresponding to each capability element in the weapon equipment according to the collection error of each test result data.

[0088] The Gaussian noise matrix GM in the embodiment is an m-row n-column matrix, each column data GM i is a group of random variables with length m, GM i Each data in GM i obeys Gaussian distribution, with mean μ and variance σ, and the values of μ and σ can be set according to the statistical values of the corresponding columns of historical test result data, such as μ and σ being 0 and 0.1 respectively.

[0089] S4, generating a performance index matrix corresponding to each capability element in the weapon equipment according to the test result data matrix, the Gaussian noise matrix and the missing data simulation matrix corresponding to each capability element in the weapon equipment.

[0090] The GM i in the Gaussian noise matrix in the embodiment is the numerical error of a certain performance index (maximum speed) that may occur during each test due to measurement or other reasons. Therefore, after adding the Gaussian noise matrix GM to the test result data matrix X 0 , the test result data of all performance indexes after the occurrence of collection error in the test can be simulated.

[0091] In order to simulate the data noise caused by data collection error and missing problem in the test result data, the following processing is performed:

[0092]

[0093] The above formula indicates that after adding the Gaussian noise matrix GM to the test result data matrix X 0 , and performing Hadamard product with the missing data simulation matrix MI, an m-row n-column performance index matrix X of the model input is obtained.

[0094] For the indexes that appear data loss, the numerical value of the test result will be set to 0, and for the indexes that do not appear data loss, the numerical value of the test result remains unchanged.

[0095] After simulating the data collection error and data missing of the test result data matrix X 0 , the generated performance index matrix X can be regarded as the result of the test result data matrix X 0 after adding data noise.

[0096] Based on the above principle, in the embodiment, the specific process of generating the performance index matrix corresponding to each capability element in the weapon equipment includes:

[0097] Step S41, adding the test result data matrix corresponding to each capability element in the weapon equipment and the Gaussian noise matrix.

[0098] Step S42, Hadamard product of the addition result and the corresponding missing data simulation matrix to obtain the performance index matrix corresponding to each capability element in the weapon equipment.

[0099] S5, according to each row of test result data in the test result data matrix corresponding to each capability element in the weapon equipment, the performance of each capability in the weapon equipment is estimated to generate the performance estimation result matrix corresponding to each capability element in the weapon equipment.

[0100] Suppose the test result data matrix X of the performance index of a certain type of tank armor equipment 0 , through the existing equipment performance evaluation method, the test result data of the performance index is calculated to obtain the performance estimation result matrix corresponding to each capability element in the weapon equipment. The performance estimation result matrix Y is:

[0101] Y=(y1,y2,...,y m )

[0102] The performance estimation result matrix Y of the embodiment is a matrix of 1 row and m columns, and each element y i in it is the performance estimation result data of each performance index of the i-th test, that is, the estimation result corresponding to the i-th row data of X 0 . For example, the test collects index data such as maximum speed, turning speed and maximum acceleration, and according to the above data, the mobility is evaluated, and the result of each test can complete the estimation of the mobility, and the combination of multiple estimation results, that is, the estimation result matrix Y of the mobility.

[0103] S6, setting an autoencoder and a regressor, and connecting the encoder in the autoencoder with the regressor.

[0104] S7, according to the test result data matrix corresponding to each capability element in the weapon equipment, the transpose matrix of the performance index matrix corresponding to each capability element in the weapon equipment is taken as the input of the autoencoder, and the autoencoder training is performed to obtain the output result of the trained encoder.

[0105] The autoencoder inputs the transpose matrix X T of the performance index matrix X with data noise in the input layer, and the encoder encodes the transpose matrix X TAfter the weighting, biasing and normalization processing are sequentially performed, the latent data features (i.e., the aggregated information matrix) are obtained, and after the latent data features are decoded (i.e., information recovery processing), the decoding results are output through the output layer until the decoding results approximate the transpose of the test result data matrix X 0 The specific process of the autoencoder training in this embodiment includes:

[0106] Step S71, setting the autoencoder parameters;

[0107] Step S72, sequentially performing weighting, biasing and normalization processing on the transpose matrix of the performance index matrix corresponding to each capability element in the weapon equipment, to obtain the aggregated information matrix of each capability element in the weapon equipment;

[0108] Step S73, performing information recovery processing on the aggregated information matrix of each capability element in the weapon equipment;

[0109] Step S74, judging whether the information recovery processing result converges to the transpose matrix of the corresponding test result data matrix, if yes, stopping the training, and taking the aggregated information matrix of each capability element in the weapon equipment as the input of the regressor (i.e., the output result of the hidden layer of the trained autoencoder); if no, adjusting the autoencoder parameters, and returning to step S72.

[0110] S8, taking the output result of the trained encoder as the input of the regressor according to the performance estimation result matrix corresponding to each capability element in the weapon equipment, performing regressor training to obtain the weapon equipment performance evaluation result output by the trained regressor.

[0111] The regressor of this embodiment takes the encoder output of the autoencoder as the input layer, inputs the latent data features after encoding, and outputs the prediction result through the regression output layer until the prediction result approximates the estimation result matrix Y. The specific process of the regressor training includes:

[0112] Step S81, setting the regressor parameters;

[0113] Step S82, sequentially performing weighting, biasing and normalization processing on the received aggregated information matrix of each capability element in the weapon equipment;

[0114] Step S83, judging whether the normalization processing result converges to the corresponding performance estimation result matrix, if yes, stopping the training, and taking the normalization processing result as the weapon equipment performance evaluation result (i.e., the regressor output result); if no, adjusting the regressor parameters, and returning to step S82.

[0115] In this embodiment, the specific process of adjusting the regressor parameters includes:

[0116] 1. Kernel density estimation is performed on the performance estimation result matrix;

[0117] 2. Probability density and loss coefficient of each kernel density estimation result are calculated:

[0118] c = a pdf(y)

[0119] wherein y is a value in the performance estimation result matrix, corresponding to the capability estimation of a row of data in the test result data matrix; pdf(y) is the corresponding sample probability density of the performance estimation result y; c is the loss coefficient corresponding to the sample with the performance estimation result y. The greater the probability density pdf(y) is, the smaller the loss coefficient c is; otherwise, the greater the loss coefficient c is. a is a constant, and 0 < a < 1;

[0120] 3. The loss function of the regressor is modified using the loss coefficient;

[0121] 4. The parameter adjustment of the regressor is performed using the gradient descent optimization method based on the modified loss function of the regressor.

[0122] The embodiment generates a test result data matrix corresponding to each capability element in the weapon equipment by the test result data of each performance indicator in each test of each capability element of the weapon equipment performance, obtains a missing data simulation matrix corresponding to each capability element in the weapon equipment by the data missing condition of the test result data matrix, obtains a Gaussian noise matrix corresponding to each capability element in the weapon equipment by the collection error of the test result data of each performance indicator in each test of each capability element of the weapon equipment performance, and obtains a performance indicator matrix containing data noise and data imbalance by the test result data matrix, the missing data simulation matrix and the Gaussian noise matrix corresponding to each capability element in the weapon equipment. The embodiment trains the autoencoder by the test result data matrix and the performance indicator matrix corresponding to each capability element in the weapon equipment under the condition that the parameters of the regressor remain unchanged, guarantees the denoising effect of the autoencoder, obtains a performance estimation result matrix corresponding to each capability element in the weapon equipment by a traditional evaluation method and the test result data of each performance indicator in each test of each capability element of the weapon equipment performance, trains the regressor by the encoder output result in the trained autoencoder and the performance estimation result matrix, obtains the trained regressor, and realizes the evaluation of the weapon equipment performance by the trained encoder and the trained regressor. The embodiment can obtain the latent features of the test data by the autoencoder, reconstruct the original data, achieve the data denoising effect, adjust the gradient contribution of high-frequency and low-frequency samples in the training by correcting the loss function of the regression model according to the probability density of the samples, and finally realize the reduction of the influence of data noise and data imbalance on the accuracy of the evaluation model, guarantee the accuracy of the evaluation result, reduce the technical threshold of completing the evaluation task, and improve the evaluation efficiency.

[0123] The above embodiment can be implemented by using the technical solutions given in the following embodiments.

[0124] Another embodiment provides a weapon equipment performance evaluation system for reducing the influence of data noise and imbalance, which comprises:

[0125] A collection module is configured to collect test result data of each performance indicator in each test of each capability element of the weapon equipment performance, and generate a test result data matrix corresponding to each capability element in the weapon equipment.

[0126] A first generation module is configured to count missing data of the test result data of each performance indicator in each test of each capability element of the weapon equipment performance, and generate a missing data simulation matrix corresponding to each capability element in the weapon equipment.

[0127] A second generation module is configured to generate a Gaussian noise matrix corresponding to each capability element in the weapon equipment according to the collection error of each test result data.

[0128] a third generation module, configured to generate a performance index matrix corresponding to each capability element in the weapon equipment according to a test result data matrix, a Gaussian noise matrix and a missing data simulation matrix corresponding to each capability element in the weapon equipment;

[0129] an estimation module, configured to estimate each capability in the weapon equipment according to each row of test result data in the test result data matrix corresponding to each capability element in the weapon equipment, to generate a performance estimation result matrix corresponding to each capability element in the weapon equipment;

[0130] a setting module, configured to set an auto-encoder and a regressor, and connect an encoder in the auto-encoder with the regressor;

[0131] an auto-encoder training module, configured to perform auto-encoder training according to the test result data matrix corresponding to each capability element in the weapon equipment, take a transposed matrix of the performance index matrix corresponding to each capability element in the weapon equipment as an input of the auto-encoder, and obtain an output result of a trained encoder;

[0132] a regressor training module, configured to perform regressor training according to the performance estimation result matrix corresponding to each capability element in the weapon equipment, take the output result of the trained encoder as an input of the regressor, and obtain a weapon equipment performance evaluation result output by a trained regressor.

[0133] Further, the first generation module comprises:

[0134] a missing data statistics submodule, configured to perform missing data statistics on test result data of each performance index in each test of each capability element in the weapon equipment, to calculate a missing number of each performance index in each capability element;

[0135] a first calculation submodule, configured to calculate a product of a number of performance indexes in each capability element and the number of tests;

[0136] a second calculation submodule, configured to calculate a ratio of the missing number of each performance index in each capability element to the product result;

[0137] a random generation submodule, configured to randomly generate a missing data simulation matrix according to the ratio of the missing number of each performance index in each capability element to the product result.

[0138] Further, the third generation module comprises:

[0139] An adding sub-module is configured to add the test result data matrix and the Gaussian noise matrix corresponding to each capability element in the weapon equipment;

[0140] A Hadamard product sub-module is configured to perform Hadamard product on the addition result and the corresponding missing data simulation matrix to obtain the performance index matrix corresponding to each capability element in the weapon equipment.

[0141] The test result data matrix and the Gaussian noise matrix have the number of rows as the number of tests and the number of columns as the number of performance indexes;

[0142] Each element value of the test result data matrix is test result data, and each column of the Gaussian noise matrix has an element value as a random variable subject to Gaussian distribution.

[0143] Further, the auto-encoder training module comprises:

[0144] A first setting sub-module is configured to set the auto-encoder parameters;

[0145] A first normalization processing sub-module is configured to sequentially perform weighting, biasing and normalization processing on the transposed matrix of the performance index matrix corresponding to each capability element in the weapon equipment to obtain the aggregated information matrix of each capability element in the weapon equipment;

[0146] An information recovery processing sub-module is configured to perform information recovery processing on the aggregated information matrix of each capability element in the weapon equipment;

[0147] A first judgment sub-module is configured to judge whether the information recovery processing result converges to the transposed matrix of the test result data matrix, if yes, stop training, and take the aggregated information matrix of each capability element in the weapon equipment as the input of the regressor; if no, adjust the auto-encoder parameters and transmit them to the first normalization processing sub-module.

[0148] Further, the regressor training module comprises:

[0149] A second setting sub-module is configured to set the regressor parameters;

[0150] A second normalization processing sub-module is configured to sequentially perform weighting, biasing and normalization processing on the received aggregated information matrix of each capability element in the weapon equipment,

[0151] A second judgment sub-module is configured to judge whether the normalization processing result converges to the corresponding performance estimation result matrix, if yes, stop training, and take the normalization processing result as the weapon equipment performance evaluation result; if no, adjust the regressor parameters and transmit them to the second normalization processing sub-module.

[0152] The principles, formulas and parameter definitions involved in the above embodiments are all applicable, and will not be described again.

[0153] Please note that the technical features of the above embodiments can be combined in any way. In order to make the description simple, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combination of the technical features does not exist contradictory, it should be considered as the scope of the description. The above embodiments only express several embodiments of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the patent. It should be noted that for ordinary skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are all within the scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. A method of reducing the effects of data noise and imbalance on the performance evaluation of weapon systems, characterized in that, The equipment performance evaluation method comprises the following steps: Step S1, collecting test result data of each performance index in each capability element of weapon equipment performance in each test to generate a test result data matrix corresponding to each capability element in the weapon equipment; Step S2, performing missing data statistics on the test result data of each performance index in each capability element of the weapon equipment performance in each test to generate a missing data simulation matrix corresponding to each capability element in the weapon equipment; Step S3, generating a Gaussian noise matrix corresponding to each capability element in the weapon equipment according to the collection error of each test result data; Step S4, generating a performance index matrix corresponding to each capability element in the weapon equipment according to the test result data matrix, the Gaussian noise matrix and the missing data simulation matrix corresponding to each capability element in the weapon equipment; Step S5, performing pre-estimation on each capability in the weapon equipment according to each row of test result data in the test result data matrix corresponding to each capability element in the weapon equipment to generate a performance pre-estimation result matrix corresponding to each capability element in the weapon equipment; Step S6, setting a self-encoder and a regressor, and connecting the encoder in the self-encoder with the regressor; Step S7, taking the transpose matrix of the performance index matrix corresponding to each capability element in the weapon equipment as the input of the self-encoder according to the test result data matrix corresponding to each capability element in the weapon equipment, performing self-encoder training to obtain the output result of the trained encoder; Step S8, taking the output result of the trained encoder as the input of the regressor according to the performance pre-estimation result matrix corresponding to each capability element in the weapon equipment, performing regressor training to obtain the weapon equipment performance evaluation result output by the trained regressor.

2. The method of evaluating the performance of a weapon system of claim 1, wherein, In the step S2, the specific process of generating the missing data simulation matrix corresponding to each capability element in the weapon equipment comprises: Step S21, performing missing data statistics on the test result data of each performance index in each capability element of the weapon equipment performance in each test to calculate the sum of missing times of all performance indexes in each capability element; Step S22, calculating the product of the number of performance indexes in each capability element and the number of tests; Step S23, calculating the ratio of the sum of missing times of all performance indexes in each capability element to the product result; Step S24, randomly generating a missing data simulation matrix according to the ratio of the sum of missing times of all performance indexes in each capability element to the product result.

3. The method of evaluating the performance of a weapon system of claim 1 wherein, In the step S4, the specific process of generating the performance index matrix corresponding to each capability element in the weapon equipment comprises: Step S41, adding the test result data matrix and the Gaussian noise matrix corresponding to each capability element in the weapon equipment; Step S42, performing Hadamard product on the addition result and the corresponding missing data simulation matrix to obtain the performance index matrix corresponding to each capability element in the weapon equipment; The row number of the test result data matrix and the Gaussian noise matrix is the number of tests, and the column number is the number of performance indexes; Each element value of the test result data matrix is test result data, and the element value of each column in the Gaussian noise matrix is a random variable subject to Gaussian distribution.

4. The method of evaluating performance of a weapon system of claim 1, wherein, In the step S7, the specific process of the autoencoder training includes: Step S71, setting autoencoder parameters; Step S72, sequentially performing weighting, biasing and normalization processing on the transposed matrix of the performance index matrix corresponding to each capability element in the weapon equipment, to obtain the aggregation information matrix of each capability element in the weapon equipment; Step S73, performing information recovery processing on the aggregation information matrix of each capability element in the weapon equipment; Step S74, judging whether the information recovery processing result converges to the transposed matrix of the corresponding test result data matrix, if yes, stopping the training, and taking the aggregation information matrix of each capability element in the weapon equipment as the input of the regressor; if no, adjusting the autoencoder parameters, and returning to step S72.

5. The method of evaluating the performance of a weapon system according to any one of claims 1 to 4, characterized in that, In the step S8, the specific process of the regressor training includes: Step S81, setting regressor parameters; Step S82, sequentially performing weighting, biasing and normalization processing on the received aggregation information matrix of each capability element in the weapon equipment; Step S83, judging whether the normalization processing result converges to the corresponding performance estimation result matrix, if yes, stopping the training, and taking the normalization processing result as the weapon equipment performance evaluation result; if no, adjusting the regressor parameters, and returning to step S82.

6. A weapon system performance evaluation system that reduces the effects of data noise and imbalance, comprising: The equipment performance evaluation system includes: A collection module, configured to collect test result data of each performance index in each capability element of weapon equipment performance in each test, to generate a test result data matrix corresponding to each capability element in the weapon equipment; A first generation module, configured to perform missing data statistics on the test result data of each performance index in each capability element of weapon equipment performance in each test, to generate a missing data simulation matrix corresponding to each capability element in the weapon equipment; A second generation module, configured to generate a Gaussian noise matrix corresponding to each capability element in the weapon equipment according to the collection error of each test result data; A third generation module, configured to generate a performance index matrix corresponding to each capability element in the weapon equipment according to the test result data matrix, the Gaussian noise matrix and the missing data simulation matrix corresponding to each capability element in the weapon equipment; An estimation module, configured to estimate each capability in the weapon equipment according to each row of test result data in the test result data matrix corresponding to each capability element in the weapon equipment, to generate a performance estimation result matrix corresponding to each capability element in the weapon equipment; A setting module, configured to set an autoencoder and a regressor, and connect an encoder in the autoencoder with the regressor; The self-encoder training module is configured to perform self-encoder training by taking the transposed matrix of the performance index matrix of each capability element in the weapon equipment as an input of the self-encoder according to the test result data matrix of each capability element in the weapon equipment, so as to obtain an output result of the trained encoder. The regressor training module is configured to perform regressor training by taking the output result of the trained encoder as an input of the regressor according to the performance estimation result matrix of each capability element in the weapon equipment, so as to obtain a performance evaluation result of the weapon equipment output by the trained regressor.

7. The weapon system performance assessment system of claim 6, wherein, The first generation module comprises: The missing data statistics submodule is configured to perform missing data statistics on the test result data of each performance index in each test of each capability element in the weapon equipment performance, so as to calculate the sum of missing times of all performance indexes in each capability element. The first calculation submodule is configured to calculate the product of the number of performance indexes and the number of tests in each capability element. The second calculation submodule is configured to calculate the ratio of the sum of missing times of all performance indexes in each capability element to the product result. The random generation submodule is configured to randomly generate a missing data simulation matrix according to the ratio of the sum of missing times of all performance indexes in each capability element to the product result.

8. The weapon system performance assessment system of claim 6, wherein, The third generation module comprises: The addition submodule is configured to add the test result data matrix and the Gaussian noise matrix of each capability element in the weapon equipment. The Hadamard product submodule is configured to perform Hadamard product on the addition result and the corresponding missing data simulation matrix, so as to obtain the performance index matrix of each capability element in the weapon equipment. The test result data matrix and the Gaussian noise matrix have the number of tests as the number of rows and the number of performance indexes as the number of columns. Each element value of the test result data matrix is test result data, and each element value of each column of the Gaussian noise matrix is a random variable subject to Gaussian distribution.

9. The weapon system performance assessment system of claim 6, wherein, The self-encoder training module comprises: The first setting submodule is configured to set self-encoder parameters. The first normalization processing submodule is configured to sequentially perform weighting, biasing and normalization processing on the transposed matrix of the performance index matrix of each capability element in the weapon equipment, so as to obtain the aggregated information matrix of each capability element in the weapon equipment. The information recovery processing submodule is configured to perform information recovery processing on the aggregated information matrix of each capability element in the weapon equipment. The first judgment submodule is configured to judge whether the information recovery processing result converges to the transposed matrix of the corresponding test result data matrix, if yes, stop training, and take the aggregated information matrix of each capability element in the weapon equipment as an input of the regressor, if not, adjust the self-encoder parameters and transmit them to the first normalization processing submodule.

10. The weapon system performance assessment system of any of claims 6-9, wherein, The regressor training module comprises: The second setting submodule is configured to set regressor parameters. The second normalization processing submodule is configured to sequentially perform weighting, biasing and normalization processing on the received aggregated information matrix of each capability element in the weapon equipment. The second judging submodule is configured to judge whether the normalized processing result converges to the corresponding performance estimation result matrix. If yes, the training is stopped, and the normalized processing result is taken as the weapon equipment performance evaluation result. If no, the regressor parameter is adjusted, and is transmitted to the second normalization processing submodule.

Images