Aircraft multi-model fusion approximate characterization method based on interest parameter domain
By adopting a multi-model fusion approximate representation method based on the interest parameter domain in aircraft design optimization, global and local error indicators are constructed and NSGA-II optimization is carried out, the problems of insufficient exploration of optimal modeling parameters and low deviation modeling accuracy are solved. It realizes efficient and precise characterization and shortened design cycles of the aircraft system.
Patent Information
- Application Number
- CN202410816751.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-24
- Publication Date
- 2024-11-08
AI Technical Summary
The existing multi-model fusion approximate representation method has problems such as insufficient exploration of optimal modeling parameters and low modeling accuracy of high and low precision analysis modeling, resulting in insufficient approximate representation accuracy and low training efficiency.
Using a multi-model fusion approximate representation method based on the interest parameter domain, global error indicators and local error indicators are constructed through a mixed scale method, and NSGA-II is used to solve the Pareto non-inferior solution set containing scaling factors and shape functions to determine the interest parameter domain The parameter combination in this paper realizes the precise characterization of the aircraft system through weighted sum of the characterization values of each basic agent model.
This method can reduce the training time of approximate characterization models, improve the efficiency of aircraft design, improve the accuracy of approximate characterization models, and shorten the aircraft design cycle. It is suitable for the field of aircraft design optimization that includes both high-precision and low-precision analysis models.
Smart Images

Figure CN118917036A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to an aircraft multi-model fusion approximate characterization method based on an interest parameter domain, and belongs to the technical field of aircraft design optimization. Background Art
[0002] With the development of computer software and hardware technology, high-fidelity modeling methods such as Finite Element Analysis (FEA) and Computational Fluid Dynamics (CFD) have been widely used in modern aircraft design. However, while high-fidelity simulation models improve the analysis accuracy, they also cause a sharp increase in the computational cost of aircraft system design optimization. Secondly, modern aircraft design optimization problems often have both high- and low-precision analysis models. Taking aircraft aerodynamic optimization as an example, CFD methods are often used as high-precision analysis models for calculating aerodynamic heat, and panel methods and engineering estimates are used as low-precision analysis models. In order to reduce the computational time of traditional design methods, proxy model methods are widely used in aircraft design. However, existing single-precision proxy models such as radial basis function (RBF) and Kriging still face the problems of insufficient approximate representation accuracy and low training efficiency. The multi-model fusion approximate representation method can balance the solution accuracy and efficiency of different precision analysis models, thereby improving the approximate accuracy of the approximate representation model and reducing the training time. Using multi-model fusion approximate characterization models instead of high-precision analysis models for aircraft design optimization can effectively improve optimization efficiency and shorten the design cycle.
[0003] Existing multi-model fusion approximate representation methods are mainly based on two architectures, namely, the model correction architecture and the hierarchical proxy model architecture. Multi-model fusion approximate representation methods based on the model correction architecture can be mainly divided into multiplicative scaling methods, additive scaling methods and comprehensive scaling methods. Haftka (1993) proposed a multiplicative scaling method. The basic idea is to first establish a proxy model from low-precision samples, and then correct the proxy model by the difference between high-precision and low-precision responses. However, when the predicted value of the low-precision proxy model for the high-precision sample point approaches 0, the multiplicative scaling method tends to fail. Based on the above defects of the multiplicative scaling method, Eldred (2004) proposed an additive scaling method, and proved that it has better approximation accuracy than the multiplicative scaling method through a large number of numerical cases. Zhou (2015) proposed an additive scaling method based on support vector regression, using two support vector regression models to approximate the low-precision analysis model and the high-low precision model deviation term. Forrester (2007) proposed a hybrid scaling method based on multiplicative and additive scaling methods. Wang (2021) proposed a multi-model fusion approximate representation model based on a hybrid scaling method and a moving least squares proxy model. In this method, the unknown parameters in the scaling factor and the deviation function term are simultaneously calculated based on the moving least squares proxy model. Zhang (2022) proposed a multi-model fusion approximate representation method based on a credible interval based on the framework of a hybrid scaling method. The credible interval length is determined by solving two sub-optimization problems, and the proxy model weights corresponding to each parameter in the credible interval are calculated by a cross-validation method. Through a large number of numerical cases and engineering cases, the approximate ability of this method is compared with three traditional scaling methods and the LR-MFS method (Zhang, 2018). The comparison results show that this method has significant advantages in approximation accuracy and robustness.
[0004] The hierarchical surrogate model architecture was proposed by Han (2012). Its basic idea is to establish low- and high-credibility surrogate models in sequence, and directly introduce the estimated value of the low-credibility surrogate model as a global trend function into the modeling of the high-credibility surrogate model. Hu (2016) proposed an improved multi-model fusion approximate characterization method. By rewriting the constant factors originally reflecting the high- and low-precision analysis models into a polynomial response surface form, the approximate accuracy of the approximate characterization model is significantly improved. In addition, Han (2019) generalized the hierarchical surrogate model architecture from two layers to any multiple layers, and successfully applied the improved method to aircraft design optimization problems such as aerodynamic shape and rotor structure. However, the above-mentioned multi-model fusion approximate characterization method based on model correction and hierarchical surrogate model architecture still faces the problems of insufficient exploration of optimal modeling parameters and low modeling accuracy of high- and low-precision analysis model deviations.
[0005] In order to better illustrate the technical solution of the present invention, the relevant mathematical foundations involved are introduced below.
[0006] (1) RBF Proxy Model
[0007] The RBF proxy model can be expressed as a linear weighted sum of radially symmetric basis functions, as shown in formula (1):
[0008]
[0009] Where n s is the number of training sample points, w i is the i-th radial function, x i is the i-th training sample point, φ(·) is the RBF kernel function, and r(·,·) is the Euclidean distance. The weight coefficient vector w can be solved by formula (2):
[0010] w=A -1 y
[0011]
[0012] Where y is the training sample point response value vector. The radial function of the present invention adopts the inverse multi-quadratic function (IMQ), which is expressed as
[0013] φ(r,c)=(r 2 +c 2 ) -0.5 (3)
[0014] Where c is the shape parameter.
[0015] (2) Hybrid Scaling Method
[0016] The hybrid scaling method can be expressed as formula (4).
[0017]
[0018] In the formula, is the approximate characterization value, ρ is the scaling factor, A proxy model for low-precision analytical models, The high and low precision analysis model deviation. The high and low precision sample sets required to construct the multi-model fusion approximate representation model by the mixed scaling method are shown in formula (5) and formula (6)
[0019]
[0020] Where, X L is a low-precision sample point, y L is the low-precision analysis model response value, X H is a high-precision sample point, y H is the high-precision analysis model response value, n L is the number of low-precision sample points, nH is the number of high-precision sample points. The deviation of high-precision and low-precision analysis models can be calculated by formula (7):
[0021]
[0022] In the formula, Represents the response value of the RBF surrogate model of the low-precision analysis model at the i-th high-precision sample point. Summary of the invention
[0023] In order to solve the problem of high-precision approximate characterization of aircraft performance models, the purpose of the present invention is to provide an aircraft multi-model fusion approximate characterization method based on the parameter domain of interest. The method constructs global error indicators and local error indicators based on the mixed scaling method, and solves a set of Pareto non-inferior solutions including scaling factors and shape functions by NSGA-Ⅱ, and then determines the parameter combination in the parameter domain of interest, and realizes the accurate characterization of complex aircraft systems by weighted summation of the characterization values of each base proxy model. The present invention has the following advantages: first, it reduces the time consumption of approximate characterization model training and improves the efficiency of aircraft design; second, it fully explores parameter combinations through multi-objective optimization and improves the accuracy of approximate characterization models.
[0024] The purpose of the present invention is achieved through the following technical solutions.
[0025] The invention discloses an aircraft multi-model fusion approximate characterization method based on an interest parameter domain, which adopts a multi-model fusion framework of a comprehensive scaling method to fuse high-precision and low-precision aircraft model information. Global and local error indicators are established respectively, and a set of Pareto non-inferior solutions including scaling factors and shape functions are solved based on NSGA-Ⅱ, and the interest parameter domain is determined by the parameter boundary in the Pareto non-inferior solution set. Based on the scaling factor step size and the shape parameter step size, the parameter combination of each base proxy model is determined in the interest parameter domain. A weight calculation method based on cross-validation and correlation coefficient is introduced to efficiently calculate the weight of each base proxy model. The weighted sum of the characterization values of each base proxy model is used as the approximate characterization value of the multi-model fusion approximate characterization method, that is, the accurate characterization of the aircraft system is realized. The invention can overcome the problems of insufficient exploration of optimal modeling parameters and low modeling accuracy of high-low precision analysis model deviation in traditional multi-model fusion approximate characterization methods, and can efficiently realize accurate approximate characterization of complex aircraft systems, thereby shortening the aircraft design cycle.
[0026] The present invention discloses an aircraft multi-model fusion approximate characterization method based on an interest parameter domain, also known as an IPD-MFS method, comprising the following steps:
[0027] Step 1: Define the parameters of the IPD-MFS method; the parameters include the total computational cost c total , high-precision sample calculation cost cH , high-to-low precision sample computation cost ratio r, shape parameter number of steps n c and the scaling factor step number n ρ ;
[0028] Step 2: Generate a set of high-precision and low-precision sample points in the design space using the Latin hypersquare experimental design method as the training sample points of IPD-MFS;
[0029] The number of high-precision sample points:
[0030] n H =c H (8)
[0031] The number of low-precision sample points:
[0032] n L =(c total -c H )·r (9)
[0033] The high-precision analysis model and low-precision analysis model of the aircraft are called respectively to obtain the model response values at the high-precision sample points and low-precision sample points; based on the low-precision sample response values, the RBF proxy model of the low-precision analysis model is established Based on the high-precision sample response value, the deviation function response value is calculated, as shown in formula (10):
[0034]
[0035] In the formula, f H,i is the i-th high-precision response value, ρ is the scaling factor, is the predicted value of the RBF surrogate model of the low-precision analysis model for the i-th high-precision response value.
[0036] Step 3: Based on the low-precision deviation function response value obtained in step 2, establish the global error index as shown in formula (11):
[0037]
[0038] Based on the low-precision deviation function response value obtained in step 2, and the deviation function RBF proxy model established by other high-precision samples other than the i-th high-precision sample Establish the local error index as shown in formula (12):
[0039]
[0040] Step 4: Based on the global and local error indicators obtained in step 3, a multi-objective optimization problem as shown in formula (13) is established, and NSGA-II is used to solve the multi-objective optimization problem to obtain a set of Pareto non-inferior solutions including the scaling factor ρ and the shape parameter c;
[0041]
[0042] Step 5: Based on the scaling factor and shape coefficient of the Pareto non-inferior solution set in step 4, determine the parameter domain of interest, as shown in formula (14):
[0043] IPD=[ρ min ,ρ max ]×[c min ,c max ] (14)
[0044] In the formula, ρ min , max is the minimum and maximum value of the scaling factor in the Pareto non-inferior solution set; c min 、c max is the minimum and maximum value of the shape coefficient in the Pareto non-inferior solution set; based on the domain of interest parameters and the number of shape parameter steps n c and the number of scaling factors n ρ , determine the base agent model as shown in formula (15), and the parameter combination in the base agent model is shown in formula (16)
[0045]
[0046] Where I and J are determined by formula (17)
[0047]
[0048] In the formula, res(k / n ρ ) means k divided by n ρ The remainder; [] indicates the rounding down operation;
[0049] Step 6: Determine the weights of each base proxy model in equation (15) by equation (18)
[0050]
[0051] The index in formula (18) k Determined by the following formula
[0052]
[0053] In the formula, It is the average value of the high-precision analysis model response; is the average response of the kth base proxy model; the first term in formula (19) represents the correlation between the base proxy model and the high-precision analysis model, and the second term represents the inverse of the local error of the base proxy model;
[0054] Step 7: Response value of the base agent model based on step 6 and the weight ω in step 6 k , the approximate characterization value of IPD-MFS is calculated by formula (20)
[0055]
[0056] In the formula, ω represents the weight vector, Represents the base surrogate model response value vector.
[0057] The approximate characterization value determined according to formula (13) is used to realize the approximate characterization of multi-model fusion of aircraft.
[0058] Step 8: Apply the IPD-MFS method described in steps 1 to 7 to the field of aircraft design optimization involving high- and low-precision analysis models to improve the accuracy and efficiency of approximate characterization and optimization of aircraft systems.
[0059] The aircraft design optimization described in step 8 includes high-precision analysis models such as structural analysis models based on large-scale finite elements and aerodynamic analysis models based on high-precision computational fluid dynamics, as well as low-precision analysis models such as panel method and engineering estimation. The IPD-MFS method can effectively improve the efficiency of aircraft design optimization and shorten the design cycle.
[0060] The aircraft system includes fields such as single-stage horizontal take-off and landing suborbital aircraft, two-stage horizontal take-off and landing suborbital aircraft, and vertical take-off and landing suborbital aircraft.
[0061] Beneficial effects:
[0062] 1. The present invention discloses an aircraft multi-model fusion approximate characterization method based on the parameter domain of interest. The global error index and the local error index are constructed based on the mixed scaling method, and a set of Pareto non-inferior solutions including scaling factors and shape functions are solved by NSGA-Ⅱ, and then the parameter combination in the parameter domain of interest is determined. The accurate characterization of the aircraft performance is achieved by weighted summation of the characterization values of each base proxy model.
[0063] 2. The present invention discloses an aircraft multi-model fusion approximate characterization method based on the parameter domain of interest, introduces a weight calculation method based on cross-validation and correlation coefficient, solves the local error index of each base proxy model by a one-by-one verification method, and combines the correlation coefficient of the base proxy model with the high-precision analysis model to efficiently solve the weight of each base proxy model, thereby improving the approximate characterization accuracy of the aircraft system.
[0064] 3. The present invention discloses an aircraft multi-model fusion approximate characterization method based on the parameter domain of interest, which is suitable for the field of aircraft design optimization that includes both high-precision analysis models and low-precision analysis models, and can improve the efficiency of aircraft design optimization and shorten the design cycle. BRIEF DESCRIPTION OF THE DRAWINGS
[0065] Figure 1 A flowchart of an aircraft multi-model fusion approximate characterization method (IPD-MFS method) based on an interest parameter domain disclosed by the present invention;
[0066] Figure 2 is a schematic diagram of the parameter domain of interest;
[0067] Figure 3 is a suborbital vehicle configuration; wherein, Figure (a) is a front view of the suborbital vehicle, Figure (b) is a side view of the suborbital vehicle, Figure (c) is a top view of the suborbital vehicle, Figure (d) is an isometric view of the suborbital vehicle, Figure (e) is a vertical configuration parameter of the head of the suborbital vehicle, Figure (f) is a horizontal configuration parameter of the head of the suborbital vehicle, and Figure (g) is a wing configuration parameter of the suborbital vehicle;
[0068] Figure 4 The following are bar graphs comparing the IPD-MFS method with similar methods for the prediction of heat flux at stagnation points and wing leading edges; Figure (a) is a comparison chart of IPD-MFS with ECV-MFS, MLS-MFS-1, and MLS-MFS-2; Figure (b) is a comparison chart of IPD-MFS with CS-MFS, MS-MFS, OA-MFS, and HK; Figure (c) is a comparison chart of IPD-MFS with RBF and KRG;
[0069] Figure 5 Figure 1 is a box plot comparing the IPD-MFS method with similar methods for the prediction of stagnation point and wing leading edge heat flux. Figure (a) is a box plot comparing the head stagnation point heat flux prediction between IPD-MFS and ECV-MFS, MLS-MFS-1, and MLS-MFS-2. Figure (b) is a box plot comparing the head stagnation point heat flux prediction between IPD-MFS and CS-MFS, MS-MFS, OA-MFS, and HK. Figure (c) is a box plot comparing the head stagnation point heat flux prediction between IPD-MFS and RBF and KR G is a box plot for the comparison of head stagnation point heat flux prediction, Figure (d) is a box plot for the comparison of wing leading edge heat flux prediction between IPD-MFS and ECV-MFS, MLS-MFS-1, and MLS-MFS-2, Figure (e) is a box plot for the comparison of wing leading edge heat flux prediction between IPD-MFS and CS-MFS, MS-MFS, OA-MFS, and HK, and Figure (f) is a box plot for the comparison of wing leading edge heat flux prediction between IPD-MFS and RBF and KRG. DETAILED DESCRIPTION
[0070] In order to further illustrate the purpose and advantages of the present invention, the present invention is further described below in conjunction with the accompanying drawings and specific implementation processes.
[0071] Example 1: Suborbital spacecraft heat flux prediction project case
[0072] like Figure 1 As shown, the present embodiment discloses an aircraft multi-model fusion approximate characterization method based on an interest parameter domain, also known as the IPD-MFS method, and the specific implementation steps are as follows:
[0073] Step 1: Define the parameters of the IPD-MFS method; the parameters include the total computational cost c total , high-precision sample calculation cost c H , high-to-low precision sample computation cost ratio r, shape parameter number of steps n c and the scaling factor step number n ρ ;
[0074] Step 2: Generate a set of high-precision and low-precision sample points in the design space using the Latin hypersquare experimental design method as the training sample points of IPD-MFS;
[0075] The number of high-precision sample points:
[0076] n H =c H (twenty one)
[0077] The number of low-precision sample points:
[0078] n L =(c total -c H )·r (22)
[0079] The high-precision analysis model and low-precision analysis model of the aircraft are called respectively to obtain the model response values at the high-precision sample points and low-precision sample points; based on the low-precision sample response values, the RBF proxy model of the low-precision analysis model is established Based on the high-precision sample response value, the deviation function response value is calculated, as shown in formula (10):
[0080]
[0081] In the formula, f H,i is the i-th high-precision response value, ρ is the scaling factor, is the predicted value of the RBF surrogate model of the low-precision analysis model for the i-th high-precision response value.
[0082] Step 3: Based on the low-precision deviation function response value obtained in step 2, establish the global error index as shown in formula (11):
[0083]
[0084] Based on the low-precision deviation function response value obtained in step 2, and the deviation function RBF proxy model established by other high-precision samples other than the i-th high-precision sample Establish the local error index as shown in formula (12):
[0085]
[0086] Step 4: Based on the global and local error indicators obtained in step 3, a multi-objective optimization problem as shown in formula (13) is established, and NSGA-II is used to solve the multi-objective optimization problem to obtain a set of Pareto non-inferior solutions including the scaling factor ρ and the shape parameter c;
[0087]
[0088] Step 5: Based on the scaling factor and shape coefficient of the Pareto non-inferior solution set in step 4, determine the parameter domain of interest, as shown in formula (14):
[0089] IPD=[ρ min ,ρ max ]×[c min ,c max ] (27)
[0090] In the formula, ρ min , max is the minimum and maximum value of the scaling factor in the Pareto non-inferior solution set; c min 、c max is the minimum and maximum value of the shape coefficient in the Pareto non-inferior solution set; based on the domain of interest parameters and the number of shape parameter steps n c and the number of scaling factors n ρ , determine the base agent model as shown in formula (15), and the parameter combination in the base agent model is shown in formula (16)
[0091]
[0092]
[0093] Where I and J are determined by formula (17)
[0094]
[0095] In the formula, res(k / n ρ ) means k divided by n ρ The remainder; [] indicates a rounding down operation;
[0096] Step 6: Determine the weights of each base proxy model in equation (15) by equation (18)
[0097]
[0098] The index in formula (18) k Determined by the following formula
[0099]
[0100] In the formula, It is the average value of the high-precision analysis model response; is the average response of the kth base proxy model; the first term in formula (19) represents the correlation between the base proxy model and the high-precision analysis model, and the second term represents the inverse of the local error of the base proxy model;
[0101] Step 7: Response value of the base agent model based on step 6 and the weight ω in step 6 k , the approximate characterization value of IPD-MFS is calculated by formula (20)
[0102]
[0103] In the formula, ω represents the weight vector, Represents the base surrogate model response value vector.
[0104] The approximate characterization value determined according to formula (13) is used to realize the approximate characterization of multi-model fusion of aircraft.
[0105] Step 8: Apply the IPD-MFS method described in steps 1 to 7 to the field of aircraft design optimization involving high- and low-precision analysis models, to improve the approximate characterization accuracy and design optimization efficiency of performance indicators such as thermal flux density and structural strength of the aircraft system.
[0106] The aircraft design optimization described in step 8 includes high-precision analysis models such as structural analysis models based on large-scale finite elements and aerodynamic analysis models based on high-precision computational fluid dynamics, as well as low-precision analysis models such as panel method and engineering estimation. The IPD-MFS method can effectively improve the efficiency of aircraft design optimization and shorten the design cycle.
[0107] The suborbital aircraft system includes fields such as single-stage horizontal take-off and landing suborbital aircraft, two-stage horizontal take-off and landing suborbital aircraft, and vertical take-off and landing suborbital aircraft.
[0108] Taking a suborbital spacecraft heat flux prediction engineering case as an example, the application of IPD-MFS method in suborbital spacecraft system design optimization is introduced. In step 1, 40 high-precision and 200 low-precision sample points are generated by the Latin hypersquare experimental design method. The parameters of the IPD-MFS method are set as follows: The total calculation cost c total =50, high-precision sample calculation cost c H =40, high-to-low precision sample calculation cost ratio r=20, shape parameter step number n c =40 and scaling factor step number n ρ =40.
[0109] Select Figure 3 The suborbital vehicle configuration shown in the figure is used as the reference configuration, and 10 design variables including the head control point and wingspan, chord length, and sweep angle are selected. The suborbital vehicle design variables are shown in Table 1:
[0110] Table 1. Suborbital vehicle design variables
[0111]
[0112] A high-precision analysis model of the suborbital vehicle is established by CFD method. Local refinement and boundary layer mesh are added at the head and wings of the suborbital vehicle to increase the calculation accuracy and convergence. The k-ω turbulence model and dual temperature model are used to calculate the heat flux density of the whole aircraft. Monitoring points are set at the head and wings of the aircraft to obtain the heat flux density values of the stagnation point and the leading edge of the wing.
[0113] For the low-precision analysis model of the suborbital vehicle, the wing leading edge heat flux density model considering the angle of attack and the wing sweep angle is established by equation (21):
[0114]
[0115] Where, χ is the wing sweep angle; α is the angle of attack; n is the correction coefficient; q sp is the heat flow of a sphere with constant radius. n and q sp It can be determined by formula (22) and formula (23)
[0116]
[0117] In the formula, r l is the leading edge curvature radius; ρ ∞ 、v ∞ h is the incoming flow density and incoming flow velocity; s h is the sensible enthalpy of the gas at the stagnation point; w is the air wall enthalpy. s and h w It can be calculated by formula (24) and formula (25)
[0118]
[0119] In the formula, c p =1.4 is the specific heat ratio of air; T ∞ is the incoming flow temperature; R = 8.31446 J / (K·mol) is the ideal gas constant; W = 0.02897 (kg / mol) is the air gas constant; T w is the wall temperature.
[0120] The head stagnation point heat flux density is calculated by formula (26)
[0121]
[0122] In the formula, r min and r max Indicates the minimum and maximum curvature radius of the head stagnation point.
[0123] The operating conditions in this embodiment are 8 Mach, 15 / angle of attack, and an altitude of 50km. In this embodiment, the model approximation accuracy is verified by a one-by-one verification method. For each multi-model fusion approximate characterization method, 40 approximate characterization model trainings are performed; for the single-precision approximate characterization method, 50 approximate characterization model trainings are performed. For this flight condition, the average prediction error of different methods is calculated to evaluate the approximate accuracy.
[0124] Depend on Figure 4 , Figure 5 It is known that the IPD-MFS method has a smaller average prediction error for the leading edge and stagnation point heat flux density of suborbital vehicles than other methods, so it has better approximation accuracy. The comparison results of engineering cases show that the IPD-MFS method helps to improve the accuracy of multi-model fusion approximate representation, and can make the optimization results more reliable and accurate in the field of aircraft optimization.
[0125] From the above comparison, it can be seen that the IPD-MFS method can effectively improve the global approximate accuracy of the approximate model in the multi-model fusion approximate characterization method. Applying the IPD-MFS method to the optimization design method based on the proxy model can improve the optimization ability and optimization efficiency of the optimization design method, and at the same time enhance the robustness of the optimization design method. IPD-MFS is suitable for the field of aircraft design optimization that contains both high- and low-precision models (high-precision models include structural analysis models based on large-scale finite elements, aerodynamic analysis models based on high-precision computational fluid dynamics, etc., and low-precision models include panel methods, engineering estimates, etc.). The engineering case of heat flux prediction of suborbital vehicles demonstrates the rationality, effectiveness and engineering practicability of the IPD-MFS method.
[0126] The specific description above further illustrates the purpose, technical solutions and beneficial effects of the invention in detail. It should be understood that the above is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.
Claims
1. An aircraft multi-model fusion approximate characterization method based on the parameter domain of interest, called the IPD-MFS method, is characterized by: The following steps are included: Step 1: Define the parameters of the IPD-MFS method; the parameters include the total computational cost c total , high-precision sample calculation cost c H , high-to-low precision sample computation cost ratio r, shape parameter number of steps n c and the scaling factor step number n ρ ; Step 2: Generate a set of high-precision and low-precision sample points in the design space using the Latin hypersquare experimental design method as the training sample points of IPD-MFS; The number of high-precision sample points: n H =c H (1) The number of low-precision sample points: n L =(c total -c H )·r (2) The high-precision analysis model and the low-precision analysis model of the aircraft are respectively called to obtain the model response values at the high-precision sample points and the low-precision sample points accordingly; Based on the low-precision sample response value, the RBF proxy model of the low-precision analysis model is established Based on the high-precision sample response value, the deviation function response value is calculated, as shown in formula (3): In the formula, f H,i is the i-th high-precision response value, ρ is the scaling factor, is the predicted value of the RBF surrogate model of the low-precision analysis model for the i-th high-precision response value; Step 3: Based on the low-precision deviation function response value obtained in step 2, establish the global error index as shown in formula (4): Based on the low-precision deviation function response value obtained in step 2, and the deviation function RBF proxy model established by other high-precision samples other than the i-th high-precision sample Establish the local error index as shown in formula (5): Step 4: Based on the global error index and local error index obtained in step 3, a multi-objective optimization problem as shown in formula (6) is established, and NSGA-II is used to solve the multi-objective optimization problem to obtain a set of Pareto non-inferior solutions including the scaling factor ρ and the shape parameter c; Step 5: Based on the scaling factor and shape coefficient of the Pareto non-inferior solution set in step 4, determine the parameter domain of interest, as shown in formula (7): IPD=[ρ min ,r max ]×[c min ,c max ] (7) In the formula, ρ min , max is the minimum and maximum value of the scaling factor in the Pareto non-inferior solution set; c min 、c max is the minimum and maximum value of the shape coefficient in the Pareto non-inferior solution set; Based on the domain of interest parameters and the number of shape parameter steps n c and the number of scaling factors n ρ , determine the base agent model as shown in formula (8), and the parameter combination in the base agent model is shown in formula (9) Where I and J are determined by formula (10) In the formula, res(k / n ρ ) means k divided by n ρ The remainder; [] indicates the rounding down operation; Step 6: Determine the weights of each base proxy model in equation (8) by equation (11) The index in formula (11) k Determined by the following formula In the formula, It is the average value of the high-precision analysis model response; is the average response of the kth base proxy model; the first term in formula (12) represents the correlation between the base proxy model and the high-precision analysis model, and the second term represents the inverse of the local error of the base proxy model; Step 7: Response value of the base agent model based on step 6 and the weight ω in step 6 k , the approximate characterization value of IPD-MFS is calculated by formula (13) In the formula, ω represents the weight vector, represents the base agent model response value vector; The approximate characterization value determined according to formula (13) is used to realize the approximate characterization of multi-model fusion of aircraft.
2. The method according to claim 1, characterized in that: The method also includes step eight: applying the IPD-MFS method described in steps one to seven to the field of aircraft design optimization including high-precision analysis models and low-precision analysis models, so as to improve the accuracy and efficiency of approximate characterization and optimization of performance indicators such as thermal flux density and structural strength of the aircraft system. The aircraft design optimization described in step 8 includes high-precision analysis models such as structural analysis models based on large-scale finite elements and aerodynamic analysis models based on high-precision computational fluid dynamics, as well as low-precision analysis models such as panel method and engineering estimation. The IPD-MFS method can effectively improve the efficiency of aircraft design optimization and shorten the design cycle. The aircraft system includes fields such as single-stage horizontal take-off and landing suborbital aircraft, two-stage horizontal take-off and landing suborbital aircraft, and vertical take-off and landing suborbital aircraft.
Citation Information
Cited By
Lithologic oil reservoir variable credibility proxy optimization method based on multi-source multi-precision data fusion
CN119783520A
Lithologic reservoir variable credibility proxy optimization method based on multi-source multi-precision data fusion
CN119783520B
Efficient prediction method for milling deformation of thin-walled workpiece
CN119939815A
Method for determining morphological parameters of high lift device and computing equipment
CN122020865A
