Error prediction method and device for high fault-tolerant pure optical CT

Abstract

The application relates to an error prediction method and device of a high-fault-tolerance pure light CT. The method comprises the following steps: obtaining a target environment temperature and a vibration amplitude of the high-fault-tolerance pure light CT, performing layered processing on the target environment temperature and the vibration amplitude, obtaining sampling point data, inputting the sampling point data into a preset finite element simulation model to obtain response data, inputting the generated prediction model response data into a mathematical approximation model to obtain an output quantity of the mathematical approximation model, performing data fitting processing on the generated prediction model response data and the output quantity of the mathematical approximation model to obtain a prediction model, inputting the to-be-detected response data into the prediction model to obtain an error prediction result. The application provides an error prediction method of a high-fault-tolerance pure light CT, which can efficiently predict pure light CT errors under extreme environments.

Description

Technical Field

[0001] This invention relates to the field of error prediction technology for pure fiber optic current transformers, specifically to an error prediction method and device for a high-fault-tolerant pure optical CT. Background Technology

[0002] The State Grid Corporation of China will further promote the construction of ultra-high-voltage direct current (UHVDC) transmission lines, primarily for transmitting clean energy. This places higher demands on the insulation levels of primary and secondary equipment in the power system and on traditional power signal measurement methods. Current transformers are key power signal measurement devices in the power grid; therefore, it is essential to ensure the accuracy of their measurement results and the reliability of their operation.

[0003] In pure optical CT, fiber optic sensing loops employ either quarter-wave plate fusion splicing for polarization state transition or an integrated waveplate without fusion points. The error characteristics of graded-pitch polarization-maintaining fibers using integrated waveplate technology differ from those using traditional quarter-wave plates, necessitating analysis. Furthermore, pure optical CT operates in extreme and variable environments, easily leading to measurement errors. Therefore, predicting the measurement error of pure optical fiber current transformers not only helps clarify the differences in error characteristics between pure optical CTs employing different waveplate technologies but also improves the reliability of pure optical CT operation, possessing significant practical application value. Traditional industrial prediction methods are mostly based on finite element or boundary element models, resulting in redundant and complex simulation calculations, high computational costs, and an inability to achieve real-time prediction. Moreover, no researchers have yet conducted research on predicting measurement errors of pure optical CTs under extreme environmental conditions.

[0004] In the field of pure optical CT error characteristics, no scholars have conducted research on pure optical CT measurement errors under extreme environments. In the existing technology, most industrial prediction methods are based on finite element or boundary element models. The simulation calculation process is redundant and complicated, with high calculation costs and no real-time prediction can be achieved. Furthermore, in the field of pure optical CT error characteristics, no scholars have conducted research on the prediction of pure optical CT measurement errors under extreme environmental conditions. Summary of the Invention

[0005] In view of this, it is necessary to provide a method and device for error prediction of pure light CT with high fault tolerance, which can efficiently predict pure light CI errors under extreme environments.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides an error prediction method for high-fault-tolerant pure-light CT, comprising:

[0008] The target ambient temperature and vibration amplitude of a high-fault-tolerant pure light CT are obtained, and the target ambient temperature and vibration amplitude are processed in layers to obtain sampling point data;

[0009] The sampling point data is input into a preset finite element simulation model to obtain response data; the response data includes the generated prediction model response data and the response data to be detected.

[0010] The response data of the generated prediction model is input into the mathematical approximation model to obtain the output of the mathematical approximation model. Data fitting processing is performed on the response data of the generated prediction model and the output of the mathematical approximation model to obtain the prediction model.

[0011] The response data to be detected is input into the prediction model to obtain the error prediction result.

[0012] Furthermore, the acquisition of the target ambient temperature and vibration amplitude of the high-fault-tolerant pure-light CT, and the layering of the target ambient temperature and vibration amplitude to obtain sampling point data, includes:

[0013] The target ambient temperature and vibration amplitude of a high-fault-tolerant pure-light CT are obtained. Based on the target ambient temperature and vibration amplitude, a sampling point matrix is ​​obtained. The sampling point matrix is ​​then subjected to layered uniform sampling processing based on the Latin hyperisotope method to obtain sampling point data.

[0014] Further, the mathematical approximation model includes: a response surface model, a Kriging model, and a radial basis function neural network model; the process of fitting the correspondence between the generated prediction model response data and the output of the mathematical approximation model to obtain the prediction model includes:

[0015] The generated prediction model response data is fitted with the output of the response surface model, the output of the Kriging model, and the output of the radial basis function neural network model to obtain the goodness of fit of the response surface model, the Kriging model, and the radial basis function neural network model. The model with the largest goodness of fit among the three models is then used as the prediction model.

[0016] Furthermore,

[0017] The mathematical expression for the response surface model is as follows:

[0018]

[0019] In the formula: y is the high-fault-tolerant pure light CT error of the response surface model, x is the target ambient temperature and the vibration amplitude, xi For the i-th set of target environmental temperature and vibration amplitude data, x j For the j-th group of target environmental temperature and vibration amplitude data, β0, β i β ij β ii β ijk and β ii...i denoted as the polynomial coefficients corresponding to the high-fault-tolerant pure-light CT error of the response surface model, and dm is the highest order of the response surface model.

[0020] Furthermore,

[0021] The mathematical expression for the Kriging model is as follows:

[0022] y(x)=f T (x)β kr +z(x);

[0023] In the formula: y is the high-tolerance pure light CT error of the Kriging model, x is the target ambient temperature and the vibration amplitude, f is the global regression function vector of the Kriging model, βkr is the regression coefficient vector of the Kriging model, and z is the stochastic correlation function of the Kriging model, and the value of z follows a normal distribution with a mean of zero.

[0024] Furthermore,

[0025] The mathematical expression for the radial basis function neural network model is as follows:

[0026]

[0027] In the formula: y is the high-fault-tolerant pure-light CT error of the radial basis function neural network model, x is the target ambient temperature and the vibration amplitude, and n R Let x be the number of sample points. s ‖ represents the target ambient temperature, the vibration amplitude, and the s-th sample point x. s The Euclidean norm, φ is the basis function of the radial basis neural network model, and λ s p represents the weight coefficients of the basis function for the s-th sample point. j (x) represents the radial basis function neural network model polynomial, K represents the number of radial basis function neural network model polynomials, and C represents the number of polynomials. j The coefficients are the polynomial coefficients of the radial basis neural network model.

[0028] Further, the step of using the model with the largest goodness of fit among the goodness of fit of the response surface model, the Kriging model, and the radial basis function neural network model as the prediction model includes:

[0029] The model with the largest goodness of fit among the coefficients of determination for the goodness of fit of the response surface model, the Kriging model, and the radial basis function neural network model is used as the prediction model.

[0030] Furthermore,

[0031] The mathematical expression for the coefficient of determination is as follows:

[0032]

[0033] In the formula: R 2 The coefficient of determination for the goodness of fit is given by , SSR is the regression sum of squares, SSR is the total sum of squares, n is the total number of sample points, and y is the total number of sample points. i , as well as The sampling point data, the average value of the response data, and the high-fault-tolerant pure light CT error of the prediction model are respectively.

[0034] Furthermore, the error prediction results are evaluated using cross-validation to obtain the accuracy of the prediction model.

[0035] Secondly, the present invention also provides a high-fault-tolerant pure-light CT error prediction device comprising:

[0036] The acquisition module is used to acquire the target ambient temperature and vibration amplitude of the high-fault-tolerant pure light CT, and to perform layered processing on the target ambient temperature and vibration amplitude to obtain sampling point data;

[0037] The processing module is used to input the sampling point data into a preset finite element simulation model to obtain response data; the response data includes the generated prediction model response data and the response data to be detected.

[0038] The output module is used to input the response data of the generated prediction model into the mathematical approximation model to obtain the output of the mathematical approximation model, and to perform data fitting processing on the response data of the generated prediction model and the output of the mathematical approximation model to obtain the prediction model.

[0039] The error prediction module is used to input the response data to be detected into the prediction model to obtain the error prediction result.

[0040] This invention provides an error prediction method and apparatus for high-fault-tolerant pure-light CT. The method acquires the target ambient temperature and vibration amplitude of the high-fault-tolerant pure-light CT, and performs stratified processing on the target ambient temperature and vibration amplitude to obtain sampling point data. This sampling point data is input into a preset finite element simulation model to obtain response data, which includes generated prediction model response data and response data to be detected. The generated prediction model response data is input into a mathematical approximation model to obtain the output of the mathematical approximation model. Data fitting processing is performed on the generated prediction model response data and the output of the mathematical approximation model to obtain a prediction model. The response data to be detected is then input into the prediction model to obtain the error prediction result. The error prediction method for high-fault-tolerant pure-light CT proposed in this application establishes a prediction model by combining a mathematical approximation model to replace the original finite element multiphysics coupling simulation model. It uses the range of the target ambient temperature and vibration amplitude of the high-fault-tolerant pure-light CT as evaluation indicators for extreme environments, and performs stratified processing on the target ambient temperature and vibration amplitude to obtain representative sampling point data. The sampling point data is then input into the prediction model to obtain the error prediction result. Compared with existing technologies, this application proposes a high-fault-tolerant pure-light CT error prediction method, which can solve the technical problem of efficiently predicting pure-light CI errors under extreme environments. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 A flowchart illustrating an embodiment of an error prediction method for high-fault-tolerant pure-light CT provided by the present invention;

[0043] Figure 2 A flowchart illustrating the sampling point generation process based on the Latin hypercube method, representing an embodiment of an error prediction method for high-fault-tolerant pure-light CT provided by this invention.

[0044] Figure 3 A flowchart illustrating the prediction model establishment of an embodiment of the error prediction method for high-fault-tolerant pure-light CT provided by the present invention.

[0045] Figure 4 A flowchart illustrating the accuracy evaluation of a prediction model in an embodiment of an error prediction method for high-fault-tolerant pure-light CT provided by the present invention.

[0046] Figure 5This is a schematic diagram of the apparatus of an embodiment of a high-fault-tolerant pure-light CT error prediction method provided by the present invention. Detailed Implementation

[0047] 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 a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0048] In the description of the embodiments of this application, unless otherwise stated, "a plurality of" means two or more.

[0049] In this embodiment of the invention, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, apparatus, product or device that includes a series of steps or modules is not necessarily limited to those steps or modules that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to such process, method, product or device.

[0050] The naming or numbering of steps in the embodiments of the present invention does not mean that the steps in the method flow must be executed in the time / logical order indicated by the naming or numbering. The execution order of the named or numbered process steps can be changed according to the technical purpose to be achieved, as long as the same or similar technical effect can be achieved.

[0051] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0052] This invention provides an error prediction method for high-fault-tolerant pure-light CT, which will be described in detail below.

[0053] Figure 1 A flowchart of an embodiment of the error prediction method for high-fault-tolerant pure-light CT provided by the present invention includes:

[0054] S110. Obtain the target ambient temperature and vibration amplitude of the high-fault-tolerant pure light CT, and perform layered processing on the target ambient temperature and vibration amplitude to obtain sampling point data;

[0055] S120. Input the sampling point data into the preset finite element simulation model to obtain the response data; the response data includes the generated prediction model response data and the response data to be detected.

[0056] S130. Input the generated prediction model response data into the mathematical approximation model to obtain the output of the mathematical approximation model. Perform data fitting processing on the generated prediction model response data and the output of the mathematical approximation model to obtain the prediction model.

[0057] S140. Input the response data to be detected into the prediction model to obtain the error prediction result.

[0058] Understandably, in the pure-light CT error prediction model, the main influencing factors affecting the error characteristics of pure-light CT, namely the ambient temperature and vibration amplitude of high-fault-tolerant pure-light CT, are used as decision variables, and the measurement error result is used as the objective function. Representative sampling points are generated based on the Latin hypercube method. After inputting the sampling point data into the finite element simulation model of the optical path, the response values ​​represented by each sampling point are obtained through simulation calculation. 90% of the data is selected as the response data for generating the prediction model. An approximate modeling technique is used, combined with three mathematical approximation models, to establish a pure-light CT error characteristic prediction model. Finally, the model is... The coefficient of determination and root mean square error (RMSE) were used as evaluation indicators. The remaining 10% of the test response data were input into the generative prediction model for cross-validation to evaluate the accuracy of the model's prediction results. Simultaneously, the prediction results were compared with the actual data from error characteristic experiments to verify the model's high prediction accuracy. The experimental measurement results of pure-light CT temperature cycling and vibration tests were used as the basic data, and the prediction model was used as the validation data. Two points were randomly selected in the variable space as validation points. The error between the actual measured value and the predicted value of the objective function was within 10%, thus verifying the accuracy of the prediction model. High-fault-tolerance pure-light CT is a high-fault-tolerance glass-slide integrated pure-light CT. Compared to traditional 1 / 4 waveplate technology, pure-light CT improves upon the waveplate technology by using integrated waveplate technology. This new type of pure-light CT is called high-fault-tolerance pure-light CT.

[0059] Figure 2 A flowchart illustrating the sampling point generation process based on the Latin hypercube method, an embodiment of the error prediction method for high-fault-tolerant pure-light CT provided by this invention, includes:

[0060] The target ambient temperature and vibration amplitude of a high-fault-tolerant pure-light CT are obtained. Based on the target ambient temperature and vibration amplitude, a sampling point matrix is ​​obtained. The sampling point matrix is ​​then subjected to layered uniform sampling processing based on the Latin hyperisotope method to obtain sampling point data.

[0061] Understandably, the range of decision variables is first determined by the range of ambient temperature T and vibration amplitude P in practical engineering. The selected decision variable x1 represents ambient temperature T, and decision variable x2 represents vibration amplitude P. The decision variable space is (x1, x2), meaning the sampling point dimension P = 2. Based on the range of each decision variable, a large number of sampling points are selected in the decision variable space (x1, x2) for analysis. To ensure that the sampling points cover the entire range of decision variables, the number of sampling points N should be controlled to be above 100. However, considering the cumbersome calculation process of the sampling point response data due to an excessive number of sampling points, the number of sampling points N should not be too large. After obtaining the size of the sampling point matrix (N×P), sampling is performed in the decision variable space according to the hierarchical uniform sampling principle of the Latin hypersolution method, resulting in the final sampling point distribution. The Latin hypersolution method, as a hierarchical uniform sampling method, divides the originally large variable sampling space into several smaller variable sampling spaces, and then performs random sampling in each smaller variable sampling space. Specifically, suppose we have p variables x1, x2, ..., xp, and we need to collect N samples within each variable's range of values. We divide the sampling space of each variable into N equal intervals, and then randomly select a value within each interval. Ultimately, each variable's sampling space will have N uniformly distributed values. By randomly combining the sampling points of each variable, we can obtain N uniformly and randomly distributed samples across the entire variable space. Unlike random sampling, the Latin hypercube method maximizes the stratification of each marginal distribution, ensuring that the sampling points of each variable fully cover its range of values. This guarantees that there is exactly one sampling point projection in any interval of any dimension of the entire variable space, ensuring sampling uniformity both at the boundaries and within the variable space.

[0062] Figure 3 A flowchart illustrating the prediction model establishment process of an embodiment of the error prediction method for high-fault-tolerant pure-light CT provided by the present invention includes:

[0063] Mathematical approximation models include: response surface model, Kriging model, and radial basis function neural network model;

[0064] The generated prediction model response data is fitted with the outputs of the response surface model, the Kriging model, and the radial basis function neural network model, respectively, to obtain the goodness of fit of the response surface model, the Kriging model, and the radial basis function neural network model. The model with the largest goodness of fit among the three models is then used as the prediction model.

[0065] It is understandable that representative sampling points are obtained through the Latin hyperisotropic method. Based on the finite element multiphysics coupling simulation model established for the pure optical CT optical path, the sampling points are substituted into the simulation calculation to obtain the response value data represented by each sampling point. Since the accuracy of the prediction models obtained by different mathematical approximation methods varies greatly during the prediction model establishment process, resulting in overfitting and underfitting issues, the goodness of fit of the prediction models obtained by different mathematical approximation methods is compared to determine the most suitable mathematical approximation model. Specifically, among the goodness of fit of the response surface model, the Kriging model, and the radial basis function neural network model, the best goodness of fit is found to be that of the response surface model; therefore, the response surface model is used as the prediction model.

[0066] Furthermore,

[0067] The mathematical expression for the response surface model is as follows:

[0068]

[0069] In the formula: y represents the high-tolerance pure-light CT error of the response surface model, x represents the target ambient temperature and the vibration amplitude, and there are multiple sets of data. i For the i-th set of target environmental temperature and vibration amplitude data, x j The target environmental temperature and vibration amplitude data are given for the j-th group, dm is the highest order of the response surface model, and β0, β i β ij β ii β ijk and β ii…i Let β = [XTX]⁻¹XTY be the polynomial coefficients corresponding to the high-tolerance pure light CT error of the response surface model, X be the polynomial matrix, and Y be the sample point response value vector. The minimum number of sample points required for response surface models of different orders are: Md+1, (Md+1)(Md+2) / 2, (Md+1)(Md+2) / 2+Md, (Md+1)(Md+2) / 2+2Md, and Md is the number of design variables.

[0070] in:

[0071]

[0072] In the formula: RSS is the sum of squared residuals, y i For actual sample data, and This is the data used for the calculation of the prediction model.

[0073] Understandably, the principle of response surface methodology (RSM) is to fit the mathematical relationship between each decision variable and the objective function using polynomials of different orders. This approach offers advantages such as simple mathematical models and fast computation. However, when the order of the RSM is high, its expression often contains numerous polynomial components. Some of these components have a relatively small impact on the model, and redundant polynomial components can lead to low computational efficiency. Therefore, for high-order polynomial RSM models, redundant polynomial components are often discarded by calculating the residual sum of squares (RSS), thus achieving the goal of selecting the final RSM model expression. In the mathematical approximation model selected in this application, the highest order of the response surface model polynomial is chosen to be third order.

[0074] Furthermore,

[0075] The mathematical expression for the Kriging model is as follows:

[0076] y(x)=f T (x)β kr +z(x);

[0077] In the formula: y is the high-tolerance pure light CT error of the Kriging model, x is the target ambient temperature and the vibration amplitude, f is the global regression function vector of the Kriging model, βkr is the regression coefficient vector of the Kriging model, and z is the stochastic correlation function of the Kriging model, and the value of z follows a normal distribution with a mean of zero.

[0078] in,

[0079] Cov[z(x i ),z(x j )]=σ 2 R(x i ,x j );

[0080] In the formula: Cov[z(x i ),z(x j )] is the covariance of z, σ 2 Let R(x) be the variance of the random function, and R(x) be the variance of the random function. i ,x j ) represents the correlation function;

[0081] Furthermore, common types of R functions include: linear correlation function, exponential correlation function, Gaussian correlation function, and spherical correlation function;

[0082] in,

[0083] The expression for the linear correlation function is:

[0084] The expression for the exponential correlation function is:

[0085] The expression for the Gaussian correlation function is:

[0086] The expression for the spherical correlation function is:

[0087]

[0088] Understandably, the Kriging model, based on the theory of variograms, can provide unbiased estimates of decision variables. In the mathematical approximation models selected in this application, the Kriging model employs the Gaussian correlation function.

[0089] Furthermore,

[0090] The mathematical expression for the radial basis function neural network model is as follows:

[0091]

[0092] Common forms of activation functions include cubic, Gaussian, multivariate quadratic, and inverse multivariate quadratic forms.

[0093] The cubic form is expressed as: φ(r) = r 3 ;

[0094] The Gaussian expression is:

[0095] The expression for a multivariate quadratic form is:

[0096] The expression for the inverse quadratic form is:

[0097] In the formula: y is the high-fault-tolerant pure-light CT error of the radial basis function neural network model, x is the target ambient temperature and the vibration amplitude, and n R Let x be the number of sample points. s ‖ represents the target ambient temperature, the vibration amplitude, and the s-th sample point x. s The Euclidean norm, φ is the basis function of the radial basis neural network model, and λ s p represents the weight coefficients of the basis function for the s-th sample point. j (x) represents the radial basis function neural network model polynomial, K represents the number of radial basis function neural network model polynomials, and C represents the number of polynomials. j The coefficients are the polynomial coefficients of the radial basis neural network model.

[0098] Understandably, a radial basis function (RBF) neural network is a three-layer neural network consisting of an input layer, hidden layers, and an output layer. It possesses strong nonlinear function approximation capabilities and generalization ability. The transition from the input layer to the hidden layer in a RBF neural network is nonlinear, while the transition from the hidden layer to the output layer is linear. In the mathematical approximation model selected in this application, the RBF activation function is chosen.

[0099] Figure 4 A flowchart illustrating the accuracy evaluation of a prediction model in an embodiment of a high-fault-tolerant pure-light CT error prediction method provided by the present invention includes:

[0100] The model with the largest goodness of fit among the coefficients of determination for the goodness of fit of the response surface model, the Kriging model, and the radial basis function neural network model is used as the prediction model.

[0101] The expression for the coefficient of determination is as follows:

[0102]

[0103] In the formula: R 2 The coefficient of determination for the goodness of fit is given by , SSR is the regression sum of squares, SSR is the total sum of squares, n is the total number of sample points, and y is the total number of sample points. i , as well as The average value of the sampling point data, the average value of the response data, and the high-fault-tolerant pure light CT error of the prediction model, respectively;

[0104] Furthermore, the evaluation metrics for goodness of fit also include: sum of squared errors, root mean square error, root mean square error, regression sum of squares, and total sum of squared deviations.

[0105] in,

[0106] The expression for the sum of squared errors is:

[0107] The expression for mean squared error is:

[0108] The standard root mean square error expression is:

[0109] The expression for the regression sum of squares is:

[0110] The expression for the total sum of squared deviations is:

[0111] Furthermore, the error prediction results are evaluated using cross-validation to obtain the accuracy of the prediction model.

[0112] It is understandable that during the establishment of predictive models, the sampling points may not be sufficiently representative in the decision variable space due to improper selection of sampling point acquisition methods, or the model may not fit the abstract correspondence of the problem well due to improper selection of mathematical approximation models. Therefore, the final predictive models often suffer from overfitting or underfitting. To obtain a predictive model that meets the fitting accuracy requirements, it is often necessary to use certain evaluation methods to evaluate the accuracy of the predictive model. The evaluation of predictive models is generally carried out from two aspects: model discrimination and goodness of fit. Generally, the predictive model should first meet the requirement of discrimination, and then meet the requirement of goodness of fit. Considering that the role of the predictive model in this application is mainly to replace the computationally expensive finite element multiphysics coupled simulation model, which involves model regression problems, it is even more necessary to examine the fitting accuracy of the predictive model. Cross-validation is often used to verify whether the model parameters meet the requirements. Its principle is to split a large amount of sampling point data into two parts: the large part is used for modeling analysis, and the small part is used to verify whether the constructed model meets the requirements. The error with the model is used as the evaluation index. This invention focuses on predicting the accuracy of pure optical CT measurements. Therefore, the coefficient of determination (R²) and root mean square error (RMSE) are used as evaluation indicators for the goodness of fit of the prediction model. Generally, R² > 0.9 and RMSE < 0.2 are used as the acceptance limits for the evaluation indicators. When the indicators meet the acceptance limits, the accuracy of the prediction model can be considered to meet the requirements. This invention mainly selects the prediction model that meets the acceptance limits and has the smallest error accuracy by comparing three mathematical models. The modeling process is based on limited sampling point data. If the prediction results are directly adopted, it may lead to certain errors. Therefore, it is necessary to compare with real experimental data for verification and improvement, thereby reducing the impact of errors. The measurement results of vibration test and temperature cycling test are used as basic data, and the prediction model is used as verification data. The prediction results are compared with the simulation results. Two points are randomly selected in the variable space as verification points, and the error between the actual measured value and the predicted value of the objective function is compared. The prediction accuracy of the prediction model is determined by the verification results.

[0113] To better implement the error prediction method for high-fault-tolerant pure-light CT in this embodiment of the invention, please refer to the corresponding documentation. Figure 5 , Figure 5 A schematic diagram of an embodiment of the device 500 provided by the present invention includes:

[0114] The acquisition module 501 is used to acquire the target ambient temperature and vibration amplitude of the high-fault-tolerant pure light CT, and to perform layered processing on the target ambient temperature and vibration amplitude to obtain sampling point data;

[0115] The processing module 502 is used to input the sampling point data into a preset finite element simulation model to obtain response data; the response data includes the generated prediction model response data and the response data to be detected.

[0116] Output module 503 is used to input the generated prediction model response data into the mathematical approximation model to obtain the output of the mathematical approximation model, and to perform data fitting processing on the generated prediction model response data and the output of the mathematical approximation model to obtain the prediction model;

[0117] The matching module 504 is used to input the response data to be detected into the prediction model to obtain the error prediction result.

[0118] The high-fault-tolerant pure-light CT error prediction device provided in the above embodiments can realize the technical solution described in the above embodiments of the high-fault-tolerant pure-light CT error prediction method. The specific implementation principle of each module or unit can be found in the corresponding content in the above embodiments of the high-fault-tolerant pure-light CT error prediction method, which will not be repeated here.

[0119] The above embodiments are only used to illustrate the design concept and features of the present invention, and their purpose is to enable those skilled in the art to understand the content of the present invention and base their decisions on facts. The scope of protection of the present invention is not limited to the above embodiments. Therefore, all equivalent changes or modifications made based on the principles and design ideas disclosed in the present invention are within the scope of protection of the present invention.

Claims

1. An error prediction method for high-fault-tolerant pure-light CT, characterized in that, include: The target ambient temperature and vibration amplitude of a high-fault-tolerant pure light CT are obtained. Based on the target ambient temperature and vibration amplitude, a sampling point matrix is ​​obtained. The sampling point matrix is ​​then subjected to layered uniform sampling processing based on the Latin hyperisotope method to obtain sampling point data. The sampling point data is input into a preset finite element simulation model to obtain response data; the response data includes the generated prediction model response data and the response data to be detected. The response data of the generated prediction model is input into the mathematical approximation model to obtain the output of the mathematical approximation model. Data fitting processing is performed on the response data of the generated prediction model and the output of the mathematical approximation model to obtain the prediction model. The response data to be detected is input into the prediction model to obtain the error prediction result; The mathematical approximation models include: response surface model, Kriging model, and radial basis function neural network model; the step of performing data fitting processing on the response data of the generated prediction model and the output of the mathematical approximation model to obtain the prediction model includes: The generated prediction model response data is fitted with the output of the response surface model, the output of the Kriging model, and the output of the radial basis function neural network model to obtain the goodness of fit of the response surface model, the Kriging model, and the radial basis function neural network model. The model with the largest goodness of fit among the three models is then used as the prediction model.

2. The error prediction method for high-fault-tolerant pure-light CT according to claim 1, characterized in that, The mathematical expression for the response surface model is as follows: ； In the formula: y represents the high-tolerance pure-light CT error of the response surface model, and x represents the target ambient temperature and the vibration amplitude. For the i-th group of target environment temperature and vibration amplitude data, For the j-th group of target environmental temperature and vibration amplitude data, , , , , and denoted as the polynomial coefficients corresponding to the high-fault-tolerant pure-light CT error of the response surface model, and dm is the highest order of the response surface model.

3. The error prediction method for high-fault-tolerant pure-light CT according to claim 1, characterized in that, The mathematical expression for the Kriging model is as follows: ； In the formula: y represents the high-tolerance pure-light CT error of the Kriging model, and x represents the target ambient temperature and the vibration amplitude. Let be the global regression function vector of the Kriging model. kr is the regression coefficient vector of the Kriging model and z is the stochastic correlation function of the Kriging model, and the value of z follows a normal distribution with a mean of zero.

4. The error prediction method for high-fault-tolerant pure-light CT according to claim 1, characterized in that, The mathematical expression for the radial basis function neural network model is as follows: ； In the formula: y represents the high-fault-tolerant pure-light CT error of the radial basis function neural network model, and x represents the target ambient temperature and the vibration amplitude. The number of sample points, The target ambient temperature, the vibration amplitude, and the s-th sample point are given. The Euclidean norm, These are the basis functions of the radial basis neural network model. The weight coefficients of the basis function for the s-th sample point are... Let K be the number of polynomials in the radial basis function neural network model, and Cj be the coefficients of the radial basis function neural network model polynomials.

5. The error prediction method for high-fault-tolerant pure-light CT according to claim 1, characterized in that, The step of using the model with the largest goodness of fit among the goodness of fit of the response surface model, the Kriging model, and the radial basis function neural network model as the prediction model includes: The model with the largest goodness of fit among the coefficients of determination for the goodness of fit of the response surface model, the Kriging model, and the radial basis function neural network model is used as the prediction model.

6. The error prediction method for high-fault-tolerant pure-light CT according to claim 5, characterized in that, The mathematical expression for the coefficient of determination is as follows: ； In the formula: The coefficient of determination for the goodness of fit is given. For the regression sum of squares, Let n be the total sum of squared deviations, and n be the total number of sample points. , as well as The sampling point data, the average value of the response data, and the high-fault-tolerant pure light CT error of the prediction model are respectively.

7. The error prediction method for high-fault-tolerant pure-light CT according to claim 1, characterized in that, The error prediction results are evaluated using cross-validation to obtain the accuracy of the prediction model.

8. An error prediction device for high-fault-tolerant pure-light CT, characterized in that, include: The acquisition module is used to acquire the target ambient temperature and vibration amplitude of the high-fault-tolerant pure light CT, obtain a sampling point matrix based on the target ambient temperature and vibration amplitude, and perform layered uniform sampling processing on the sampling point matrix based on the Latin hyperisotope method to obtain sampling point data. The processing module is used to input the sampling point data into a preset finite element simulation model to obtain response data; the response data includes the generated prediction model response data and the response data to be detected. The output module is used to input the response data of the generated prediction model into the mathematical approximation model to obtain the output of the mathematical approximation model, and to perform data fitting processing on the response data of the generated prediction model and the output of the mathematical approximation model to obtain the prediction model. The error prediction module is used to input the response data to be detected into the prediction model to obtain the error prediction result; The mathematical approximation models include: response surface model, Kriging model, and radial basis function neural network model; the step of performing data fitting processing on the response data of the generated prediction model and the output of the mathematical approximation model to obtain the prediction model includes: The generated prediction model response data is fitted with the output of the response surface model, the output of the Kriging model, and the output of the radial basis function neural network model to obtain the goodness of fit of the response surface model, the Kriging model, and the radial basis function neural network model. The model with the largest goodness of fit among the three models is then used as the prediction model.

Images