An aircraft multi-source data fusion method, an electronic device, and a storage medium

Abstract

The application provides a multi-source data fusion method, an electronic device and a storage medium, and belongs to the technical field of multi-source data fusion. The method comprises the following steps: S1. CFD data and wind tunnel test data are acquired, and a mapping relationship between characteristic variables and target variables is established by using a machine learning model; S2. The CFD data and the wind tunnel test data are extracted by a repeated sampling method, a plurality of training sets and corresponding verification sets are established, a model is established based on different training sets, and a multi-model machine learning model is formed; S3. The mean square error of the prediction results of the multi-model machine learning model is solved by using the verification set; S4. The prediction results output by different training sets and the mean square errors obtained by different prediction results are recorded; and S5. The training set is input into the multi-model machine learning model, and the prediction results of the multi-model are combined based on a weighted method of the mean square error. The problem that the model precision and the fitting degree are poor due to model overfitting is solved.

Description

TECHNICAL FIELD

[0001] The application relates to a multi-source data fusion method, in particular to a multi-source data fusion method, an electronic device and a storage medium, and belongs to the technical field of multi-source data fusion. BACKGROUND

[0002] In the preliminary development stage of an aircraft, corresponding aerodynamic numerical calculation and aerodynamic characteristic evaluation need to be carried out according to the aerodynamic shape of the aircraft, so as to plan further optimization design scheme, and therefore it is necessary to accurately and efficiently analyze the aerodynamic shape of the aircraft. At present, a wind tunnel test and a CFD numerical simulation are main technical methods for aerodynamic characteristic analysis of the aircraft. The wind tunnel test can accurately analyze the aerodynamic characteristics of the aircraft, but has problems of high cost and long test cycle. The CFD method solves discrete equations in fluid mechanics by numerical calculation, and has a lower cost than the wind tunnel test, but has slightly lower accuracy than the wind tunnel test. In view of the disadvantages of the wind tunnel test and the CFD numerical simulation, a method that can meet low test cost and obtain high-precision data analysis results needs to be further developed on the basis of reasonably giving full play to the advantages. In order to meet this demand, a concept of multi-source aerodynamic data fusion is widely researched and applied, the essence of which is to use a proxy model that can map the relationship between low-precision data and high-precision data, and realize accurate prediction of high-precision data. However, the method currently involved uses a single modeling of a fixed training set and test set to obtain a single mapping relationship between high-precision data and low-precision data, which is prone to model overfitting, seriously affecting the accuracy and fitting degree of the model. SUMMARY

[0003] In the following, a brief summary of the present application is given in order to provide a basic understanding of some aspects of the present application. It should be understood that this summary is not a comprehensive summary of the present application. It is not intended to determine the key or important parts of the present application, nor to limit the scope of the present application. Its purpose is only to give some concepts in a simplified form as a prelude to the more detailed description discussed later.

[0004] In view of this, in order to solve the technical problem of poor model accuracy and fitting degree caused by model overfitting in the prior art, the present application provides a multi-source data fusion method based on an aircraft.

[0005] Scheme one, a multi-source data fusion method based on an aircraft, comprising the following steps:

[0006] S1. Obtain CFD data and wind tunnel test data, the CFD data are characteristic variables, and the wind tunnel test data are target variables; a mapping relationship between the characteristic variables and the target variables is established by using a machine learning model;

[0007] S2. The CFD data and the wind tunnel test data are extracted by a repeated sampling method to establish multiple training sets and corresponding verification sets, a model is established based on different training sets, and a multi-model machine learning model is formed;

[0008] S3. The mean square errors of the prediction results of the multi-model machine learning model are solved by the verification set respectively;

[0009] S4. The prediction results output by different training sets and the mean square errors obtained by different prediction results are recorded;

[0010] S5. The training set is input into the multi-model machine learning model, and the prediction results of the multi-model are combined based on the weighted method of the mean square error.

[0011] Preferably, the CFD data and the wind tunnel test data are obtained in the following manner:

[0012] Step A, the wind tunnel test data are summarized and arranged, including a test information table and a train table;

[0013] Step B, the standard templates of the conventional force test are arranged for different wind tunnel data storage methods;

[0014] Step C, an output interface of a data program is created to convert the original data in the wind tunnel test;

[0015] Step D, geometric shape parameters and inflow condition parameters are obtained from the converted wind tunnel test data and CFD numerical simulation data;

[0016] Step E, force and moment coefficients of the target variables are obtained, including lift coefficient, drag coefficient, side force coefficient, pitch moment coefficient, yaw moment coefficient and roll moment coefficient;

[0017] Step F, the data of the characteristic variables and the target variables are preprocessed, including data cleaning, data anomaly processing, data missing value filling and data feature engineering.

[0018] Preferably, the CFD data are lift coefficients, drag coefficients, side force coefficients, pitch moment coefficients, yaw moment coefficients and roll moment coefficients obtained by an aircraft at different Mach numbers, attack angles, sideslip angles and CFD calculations;

[0019] The wind tunnel test data are lift coefficients, drag coefficients, side force coefficients, pitch moment coefficients, yaw moment coefficients and roll moment coefficients obtained by wind tunnel tests respectively.

[0020] Preferably, the machine learning model uses a variety of linear regression models based on regularization or a variety of nonlinear regression models based on decision tree fitting.

[0021] Preferably, the method for solving the mean square error of the prediction result of the multi-model machine learning model by the verification set is: let be the target true value in the verification set, be the target prediction value in the verification set, and n be the total number of samples in the verification set. The mean square error (MSE) is abbreviated as M, and M can be expressed as: .

[0022] Preferably, the method for combining the prediction results of the multi-model based on the weighted method of the mean square error is: based on the mean square error obtained by different prediction results, the weight information of the proxy model is solved , and the expression is:

[0023]

[0024] wherein, is the mean square error of the model, p is the total number of models, q is the adjustment index, and the value of q needs to be selected according to the difference between the model and the data set; when the value of q is positive, the weight is inversely proportional to its mean square error; when the value of q is 0, the weight of each model is the same; when the value of q is negative, the weight is proportional to its mean square error; let be the model formed under the repeated sampling method, and the expression based on the result under the re-sampling weighting method is:

[0025] .

[0026] Scheme two, an electronic device, comprising a memory and a processor, the memory stores a computer program, when the processor executes the computer program, the steps of the method for fusing multi-source data of an aircraft based on the scheme one are realized.

[0027] Scheme three, a computer readable storage medium, which stores a computer program, when the computer program is executed by a processor, the method for fusing multi-source data of an aircraft based on the scheme one is realized.

[0028] The beneficial effects of the present application are as follows: the present application forms multiple model results by repeated sampling method, thereby obtaining a multi-mapping relationship between low-precision data and high-precision data, and further using a weighted method of mean square error to fuse the mapping relationship of multiple models, thereby effectively reducing the influence of simulation overfitting, and further improving the precision and fitting degree of the proxy model for predicting the aerodynamic coefficient. The present application can also effectively reduce the number of traditional wind tunnel tests and CFD numerical simulation calculations, which meets the low test cost and can obtain high-precision data analysis results, and provides a scheme for the aerodynamic characteristic analysis and design of the aircraft. BRIEF DESCRIPTION OF DRAWINGS

[0029] The drawings described herein are used to provide further understanding of the present application, and form a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application, and do not constitute an improper limitation on the present application. In the drawings:

[0030] Fig. 1 It is a flow chart of a method for fusing multiple source data of an aircraft.

[0031] Fig. 2 It is a flow chart of a method for selecting a proxy model. DETAILED DESCRIPTION

[0032] In order to make the technical solutions and advantages in the embodiments of the present application more clear and apparent, the exemplary embodiments of the present application are further described in detail below with reference to the drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, and not an exhaustive enumeration of all embodiments. It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0033] Embodiment 1, reference Figs. 1-2 The present embodiment is a method for fusing multiple source data of an aircraft, comprising the following steps:

[0034] S1. Obtain CFD data and wind tunnel test data, the CFD data being a characteristic variable, and the wind tunnel test data being a target variable; use a machine learning model to establish a mapping relationship between the characteristic variable and the target variable;

[0035] Specifically, the CFD calculation data (low-precision data) is set as a characteristic variable to assist in analyzing the wind tunnel test data (high-precision data), thereby obtaining a mapping relationship between the CFD calculation data (low-precision data) and the wind tunnel test data (high-precision data);

[0036] Specifically, the CFD data and the wind tunnel test data are obtained in the following manner:

[0037] Step A, summarize the wind tunnel test data, including a test information table and a train table;

[0038] Step B, standard template of conventional force test is arranged for different wind tunnel data storage mode;

[0039] Step C, output interface of data program is created to transform raw data in wind tunnel test;

[0040] Step D, geometric shape parameters and inflow condition parameters are obtained from transformed wind tunnel test data and CFD numerical simulation data;

[0041] Step E, force and moment coefficients of target variables are obtained, including lift coefficient, drag coefficient, side force coefficient, pitching moment coefficient, yawing moment coefficient and rolling moment coefficient;

[0042] Step F, data preprocessing is performed on data of characteristic variables and target variables, including data cleaning, data anomaly processing, data missing value filling and data feature engineering, etc.

[0043] CFD data are lift coefficient, drag coefficient, side force coefficient, pitching moment coefficient, yawing moment coefficient and rolling moment coefficient obtained by CFD calculation of the aircraft at different Mach numbers, attack angles, sideslip angles and CFD;

[0044] Wind tunnel test data are lift coefficient, drag coefficient, side force coefficient, pitching moment coefficient, yawing moment coefficient and rolling moment coefficient obtained by wind tunnel test respectively;

[0045] The machine learning model uses multiple linear regression models based on regularization; when the training set is small, the linear regression method will face the problem of model overfitting. In order to avoid this situation, necessary regularization processing is adopted for the linear regression model. Regularization has the characteristics of enhancing the generalization ability of the model. Regularization reduces the weight coefficient (regression coefficient) by adding a penalty term to the loss function to reduce the complexity of the model, to a certain extent, to avoid overfitting and reduce the variance of the model. Regularization reduces the weight coefficient (regression coefficient) by adding a penalty term to the loss function to reduce the complexity of the model, and the linear regression method based on regularization has Lasso regression, Ridge regression and ElasticNet regression, which perform regularization processing through different penalty term expressions.

[0046] The expression of regularization has two forms of L1 and L2, which are combined with the linear regression loss function of mean square error and the penalty term to obtain the latest loss function. Let be the true value of the characteristic variable, be the true value of the target variable, be the predicted value of the target variable, be the total number of samples, the total number of characteristic variables, the strength of regularization, the weight coefficient (regression coefficient);

[0047] Then:

[0048] The expression based on L1 regularization is:

[0049]

[0050] The expression based on L2 regularization is:

[0051]

[0052] The Lasso regression method introduces the above-mentioned L1 regularization, and the Ridge regression method introduces L2 regularization. The ElasticNet regression method is between Lasso and Ridge, and simultaneously applies L1 and L2 regularization in the loss function.

[0053] The machine learning model uses a variety of nonlinear regression models based on decision tree fitting; including the Random Forest method and the AdaBoost method. The Random Forest method extracts multiple training sets through random sampling with replacement to form multiple decision tree models, and for each decision tree model, effective splitting is performed according to information gain or Gini coefficient, and finally multiple decision trees are combined into a random forest. The principle of the AdaBoost method is to form a dependent relationship among multiple decision trees, and by analyzing the learning error rate of the previous decision tree, more weight is allocated to the sample with a high learning error rate, so that the sample with high weight is more valued in the subsequent decision tree. Through iterative training of decision trees and updating of sample weight values, such an iterative process is terminated until the maximum number of iterations is reached or a specified error rate is met.

[0054] S2. The machine learning model (proxy model) selected in S1 is used to extract CFD data and wind tunnel test data through repeated sampling method, to establish multiple training sets and corresponding validation sets, to establish models based on different training sets, and to form a multi-model machine learning model;

[0055] Specifically, the repeated sampling method refers to randomly sampling the sample data multiple times without replacement, and grouping the sampled samples into a training set and the unsampled samples into a validation set.

[0056] Specifically, the training set is 80% to 90% of the state points in the data set.

[0057] Specifically, the training set and the validation set are allocated in a ratio of 9:1 or 8:2, wherein the training set is again divided into a new training set and a new validation set by the resampling weighting method, and the new training set and the new validation set are allocated in a ratio of 9:1 or 8:2. The validation set is used to adjust the hyperparameters of the trained model to achieve the optimal model effect and to solve the mean square error; and the test set is used to finally judge the prediction accuracy and generalization ability of the model after the resampling weighting method is adjusted.

[0058] Specifically, the sample data is subjected to multiple random sampling without replacement, and the sampled samples form a training set and the unsampled samples form a test set, and multiple models are established based on different training sets.

[0059] S3. The agent model after using the repeated sampling method in S2, multiple training sets and corresponding validation sets are established by the repeated sampling method, and the mean square error of the prediction results of the multiple model machine learning models is solved by the validation set respectively;

[0060] Let be the target true value in the test set, be the target predicted value in the test set, n be the total number of samples in the test set, and the mean square error M be represented as

[0061]

[0062] S4. Record the prediction results output by different training sets and the mean square errors obtained by different prediction results.

[0063] Specifically, the multiple new training sets obtained by the repeated sampling method are modeled and trained, and the model information is recorded. Further, the corresponding validation sets are evaluated and verified by the mean square error, and the value of the mean square error is recorded.

[0064] S5. Input the training set into the multiple model machine learning model, and combine the prediction results of the multiple models based on the weighted method of the mean square error.

[0065] Specifically, the multiple model machine learning model is obtained after multiple modeling;

[0066] Specifically, the weighting method is based on the mean square error of the agent model under different training sets, i.e., the mean square error obtained based on different prediction results, to solve the agent model weight information , and the expression is:

[0067]

[0068] wherein, is the mean square error of the model, p is the total number of models, q is the adjustment index, the value of q needs to be selected according to the difference between the model and the data set; when the value of q is positive, the weight assigned to each model is inversely proportional to its mean square error; when the value of q is 0, the weight assigned to each model is the same; when the value of q is negative, the weight assigned to each model is proportional to its mean square error . . . .

[0069] . . .

[0070] For example, n models are formed through step S4, and the mean square errors obtained by each model are combined and recorded as MSE={ }. Secondly, the proxy model weight information expression can be solved to obtain the weight of each model W={ }. Thirdly, the prediction value of the proxy model formed by mapping the characteristic variable and the target variable is multiplied by the corresponding weight value, and finally the fused prediction result is obtained.

[0071] In embodiment 2, the computer device of the application can be a device including a processor and a memory, such as a single-chip microcomputer including a central processing unit. When the processor executes the computer program stored in the memory, the steps of the aircraft multi-source data fusion method described above are realized.

[0072] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), ready-to-use programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0073] The memory can mainly include a program storage area and a data storage area. The program storage area can store an operating system, application programs required by at least one function (such as a sound playing function, an image playing function, etc.), and the like. The data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), and the like. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state memory devices.

[0074] Embodiment 3, computer readable storage medium embodiment

[0075] The computer readable storage medium of the present application can be any form of storage medium readable by the processor of the computer device, including but not limited to non-volatile memory, volatile memory, ferroelectric memory, etc., and the computer readable storage medium stores a computer program. When the processor of the computer device reads and executes the computer program stored in the memory, the steps of the above-mentioned aircraft multi-source data fusion method can be realized.

[0076] The computer program includes computer program code, which can be in the form of source code, object code, executable files, or some intermediate forms, etc. The computer readable medium can include any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the content included in the computer readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction, for example, in some jurisdictions, according to legislation and patent practice, the computer readable medium does not include electrical carrier signals and telecommunication signals.

[0077] While the application has been described in accordance with the various embodiments shown and described, it is to be understood that the application is not limited to those precise embodiments, and that various modifications and changes can be made by those skilled in the art without departing from the scope of the present application. Furthermore, the language used in this specification has been principally selected for readability and instructional purposes and can not have been selected to delineate or circumscribe the patent rights to which it refers. Accordingly, the present application is intended to be illustrative, but not limiting, of the scope of the application, which is set forth with particularity in the claims that follow.

Claims

1. An aircraft multi-source data fusion method based on, characterized by, Comprising the following steps: S1. Obtain CFD data and wind tunnel test data, the CFD data is a characteristic variable, the wind tunnel test data is a target variable, and a mapping relationship between the characteristic variable and the target variable is established by using a machine learning model; The way of obtaining CFD data and wind tunnel test data is as follows: Step A, the wind tunnel test data is summarized, including test information table and train table; Step B, the standard template of the conventional force test is arranged for different wind tunnel data storage modes; Step C, the output interface of the data program is created, and the original data in the wind tunnel test is converted; Step D, the geometric shape parameters and inflow condition parameters are obtained from the converted wind tunnel test data and CFD numerical simulation data; Step E, the force and moment coefficients of the target variable are obtained, including lift coefficient, drag coefficient, side force coefficient, pitch moment coefficient, yaw moment coefficient and roll moment coefficient; Step F, the data of the characteristic variable and the target variable are preprocessed, including data cleaning, data anomaly processing, data missing value filling and data feature engineering; S2. The CFD data and the wind tunnel test data are extracted by repeated sampling method, a plurality of training sets and corresponding verification sets are established, a model is established based on different training sets, and a multi-model machine learning model is formed; S3. The mean square error of the prediction results of the multi-model machine learning model is solved by the verification set respectively; S4. The prediction results output by different training sets and the mean square errors obtained by different prediction results are recorded; S5. The training set is input into the multi-model machine learning model, and the prediction results of the multi-model are combined based on the weighted method of the mean square error; The training set is input into the multi-model machine learning model, and the method for combining the prediction results of the multi-model based on the weighted method of mean square error is: solving the proxy model weight information based on the mean square error obtained from different prediction results The expression is: ； wherein, is the mean squared error of the model, p is the total number of models, and q is a tuning index, the value of which needs to be chosen depending on the difference between the model and the dataset; When q is positive, the assigned weight is inversely proportional to the mean squared error of the model; when q is zero, the assigned weight is the same for all models; when q is negative, the assigned weight is proportional to the mean squared error of the model When q is positive, the assigned weight is inversely proportional to the mean squared error of the model; when q is zero, the assigned weight is the same for all models; when q is negative, the assigned weight is proportional to the mean squared error of the model When q is positive, the assigned weight is inversely proportional to the mean squared error of the model; when q is zero, the assigned weight is the same for all models; when q is negative, the assigned weight is proportional to the mean squared error of the model When q is positive, the assigned weight is inversely proportional to the mean squared error of the model; when q is zero, the assigned weight is the same for all 。 2. The method according to claim 1, wherein, The CFD data is the lift coefficient, drag coefficient, side force coefficient, pitch moment coefficient, yaw moment coefficient and roll moment coefficient obtained by the aircraft at different Mach numbers, attack angles, side slip angles and CFD calculation; The wind tunnel test data is the lift coefficient, drag coefficient, side force coefficient, pitch moment coefficient, yaw moment coefficient and roll moment coefficient obtained by the wind tunnel test respectively.

3. The method according to claim 2, wherein, The machine learning model uses a plurality of linear regression models based on regularization or a plurality of nonlinear regression models based on decision tree fitting.

4. The method according to claim 3, wherein, The method for solving the mean square error of the prediction result of the multi-model machine learning model through the verification set is: let the target real value in the test set be y, the target prediction value in the test set be y', and n be the total number of samples in the test set, and the mean square error M is expressed as: .

5. An electronic device, comprising: It comprises a memory and a processor, the memory stores a computer program, and the processor implements the steps of the aircraft multi-source data fusion method according to any one of claims 1-4 when executing the computer program.

6. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the aircraft multi-source data fusion method according to any one of claims 1-4.

Images