A proxy model assisted optimization method for population pre-screening and spatial reduction

Abstract

The application discloses a kind of population pre-screening and space reduction proxy model aided optimization method, it is related to model optimization field, first, the extremely low computing cost advantage of proxy model is used to pre-screen population individual, then based on the individual characteristics after pre-screening, optimization space is reduced.Subsequently, subsequent operation is based on the individual selected in advance and reduced optimization space, to realize the goal of fast optimization convergence.On the other hand, in the optimization process, the method makes full use of the acceleration auxiliary role of proxy model in engine overall performance simulation, reduces the evaluation cost of objective function, further reduces optimization calculation amount.This not only improves optimization efficiency, but also guarantees the accuracy of original overall performance simulation model.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of model optimization, and particularly to a proxy model aided optimization method for population pre-screening and space reduction. BACKGROUND

[0002] The design of engine control laws directly affects the overall performance of the engine under different operating conditions. Optimizing a reasonable control strategy can effectively improve the engine performance and ensure the engine operates under stable and safe conditions. The main control variables of the engine usually include shaft speed and adjustable geometry variables. For engine architecture schemes characterized by variable cycles, there are numerous adjustable geometry variables, which brings significant challenges to the design of control laws.

[0003] Currently, the research on the design of engine control laws mostly adopts the method of manual debugging by researchers. This method requires designers to have high professional knowledge and to have a deep understanding of the influence of each control variable on the engine performance, and to comprehensively adjust to avoid risks such as over-temperature, over-speed and surge. However, this process is tedious and requires repeated iteration of design parameters and component characteristics, which seriously affects the design efficiency. Therefore, some research introduces optimization algorithms to realize the automatic optimization design of control laws. However, optimization algorithms need to select a small number of competitive solutions from a large number of candidate solutions when evaluating the objective function, which requires frequent calls to the engine performance simulation model for evaluation of the objective function, resulting in huge computational overhead.

[0004] The mainstream research mainly adopts the following two types of solutions: one is to improve the optimization algorithm, including adopting different optimization strategies, introducing adaptive mutation operators, and increasing optimization boundary constraints. Although certain progress has been made in computational efficiency through various enhancement techniques, it is still difficult to significantly reduce the computational cost without reducing the robustness of the algorithm; the second is to use a proxy model to approximate the engine performance simulation model to speed up the calculation of the objective function. However, this method still faces great limitations and deficiencies when applied to the optimization of aero-engine performance simulation. Specifically, although the proxy model can greatly reduce the computational load by directly replacing the engine performance simulation model, the inevitable prediction error of the proxy model may cause the optimization result to deviate from the optimal solution.

[0005] Efficient global optimization (EGO) is a further development of the application of proxy models in optimization problems. EGO updates the local accuracy of the proxy model by constructing a reasonable sampling strategy, thereby gradually approaching the local or global optimal solution. However, during the optimization process, EGO needs to repeatedly call the high-cost simulation model to supplement sample points and constantly update and train the proxy model. This repeated calculation process brings additional computational consumption, which may even offset the efficiency improvement brought by the replacement of the simulation model by the proxy model.

[0006] In summary, although the current research has made some progress in the optimization design of engine control law, there is still a lot of room for improvement in terms of optimization accuracy and efficiency. Therefore, it is urgent to develop a new engine control law optimization design method that can significantly reduce the computational cost without compromising the robustness and accuracy of the algorithm, thereby meeting the optimization design requirements of engine control law. SUMMARY

[0007] To solve the above problems, based on the defects and deficiencies of the prior art, the key technology of the present application is to propose a population pre-screening and space reduction proxy model assisted optimization algorithm. Based on this method, the rapid automatic optimization design of engine control law can be realized. The present application includes the training and application of engine overall performance proxy model and auxiliary proxy model, the pre-screening of initial population and optimization space reduction operation, and the program algorithm flow of the whole control law design method.

[0008] The population pre-screening and space reduction proxy model assisted optimization algorithm includes the following steps:

[0009] S1, according to the characteristics of the engine, the engine control law variables involved in the control law optimization design are defined as independent variables, and the preliminary optimization interval IV range of the independent variables is determined;

[0010] S2, training an engine overall performance proxy model with the independent variables and engine operating conditions as input variables and engine performance as output variable;

[0011] Training an auxiliary proxy model with the independent variables and engine operating conditions as input variables and the guess vector X obtained by iterative solution convergence of the engine simulation model as output variable;

[0012] S3, based on the preliminary optimization interval IV range of the independent variables, combined with the engine overall performance proxy model, the creation and pre-screening of the initial population of the optimization algorithm are performed to obtain the pre-screening initial population;

[0013] S4, according to the individual characteristics of the pre-screening initial population, the preliminary optimization interval is reduced to obtain the reduced optimization interval;

[0014] S5, based on the pre-screening initial population and the reduced optimization interval, a target fitness driven population optimization algorithm is selected, and the optimization search steps of the algorithm are executed to complete the auxiliary optimization of the engine overall performance proxy model combined with the auxiliary proxy model.

[0015] Preferably, the control law variables in S1 include shaft speed and adjustable geometry variables, and the preliminary optimization interval is the upper and lower bounds of the control law variables, expressed as:

[0016]

[0017] wherein m is the number of independent variables, IV1 is the first independent variable, IV2 is the second independent variable, IV m is the mth independent variable, IV m is the lower bound of IV m , is the upper bound of IV m .

[0018] Preferably, the specific content of S3 of preliminarily optimizing the interval based on the given independent variables, in combination with the engine overall performance proxy model, for creating and pre-screening the initial population of the optimization algorithm is as follows:

[0019] S301, using the Latin hypercube sampling method, generating a pre-screening population containing N*P individuals in the preliminarily optimized interval IV range of the independent variables, wherein N represents the pre-screening multiple, and P represents the number of individuals of the standard optimization population;

[0020] S302, constructing the objective fitness function evaluation formula Fit=f(P) according to the optimization objective and the constraint condition, wherein P is the engine overall performance, and Fit is the individual objective fitness.

[0021] S303, based on the engine overall performance proxy model, calculating the engine performance corresponding to each individual in the population, and then based on the objective fitness function evaluation formula, evaluating the objective fitness of all individuals in the population to obtain the objective fitness value;

[0022] S304, sorting the individuals in the pre-screening population from low to high according to the objective fitness value, and selecting the P individuals with the highest ranking of the objective fitness value.

[0023] Preferably, the specific content of S4 of reducing the preliminarily optimized interval to obtain the reduced optimization interval according to the individual characteristics of the pre-screening initial population is as follows:

[0024] According to the extreme values of the independent variables in the individuals of the pre-screening initial population, i.e. the individual characteristics, updating the upper and lower bounds of the preliminarily optimized interval, and introducing a relaxation term R j / k for adjustment, wherein the updating expression is as follows:

[0025]

[0026] IV j IV 1,j ,IV 2,j ,...,IV P,j )-R j / k, IV j ),j=1,2,...,m

[0027]

[0028] in, Let j be the new upper bound of the j-th independent variable. IV j 'Is the new lower bound for the j-th independent variable, IV 1,j IV is the j-th independent variable for the first individual. 2,j IV is the j-th independent variable for the second individual. P,j Let R be the j-th independent variable for the P-th individual. j Let k be the initial optimization interval length for the j-th independent variable, and k be the number of equal divisions of the optimization interval length, which is generally taken as around 20.

[0029] The preferred expression for reducing the optimization interval is:

[0030]

[0031] Among them, IV range 'To reduce the optimization interval, IV m 'For IV m The new lower bound after reduction, IV m The new upper bound after reduction.

[0032] Preferably, the specific content of the optimization search step in S5 is as follows:

[0033] Predict the guess vector X corresponding to the independent variables of a new individual using an auxiliary agent model;

[0034] This is used as the initial guess for the engine performance simulation model. The engine performance is obtained by solving the model, and the target fitness value is calculated based on the target fitness function Fit=f(P).

[0035] The optimization process ends when the termination condition of the optimization algorithm is met.

[0036] In summary, this application addresses the control law optimization problem of an adaptive cyclic aero-engine by improving upon traditional engine performance optimization algorithms. However, the method of this invention is applicable to population optimization algorithms driven by various objective fitness levels and aero-engines of different configurations. Compared with existing engine control law optimization design methods, the solution of this invention has the following advantages:

[0037] 1. The method of this invention enables automatic optimization design of engine control laws, avoiding the tediousness and inefficiency of manually adjusting and designing control laws. Furthermore, compared to traditional optimization methods, it significantly reduces computational costs and improves the efficiency of control law optimization design without compromising algorithm robustness and accuracy.

[0038] 2、The optimization algorithm has good universality, can be flexibly applied to various control law optimization problems of different engine types, and can be adaptively improved based on various target fitness driving population optimization algorithms.

[0039] The technical method of the application is further described in detail below by means of the drawings and examples. BRIEF DESCRIPTION OF DRAWINGS

[0040] Figure 1 The flow chart of the proxy model assisted optimization method of population pre-screening and space reduction of the application;

[0041] Figure 2 The schematic diagram of the self-adaptive cycle engine of the application;

[0042] Figure 3 The comparison chart of the optimal individual fitness of the traditional optimization method and the optimization method model of the application;

[0043] Figure 4 The comparison chart of the number of model calls of the traditional optimization method and the optimization method model of the application, wherein, Figure 4 (a) in the figure is the number of model calls of the traditional optimization method, Figure 4 (b) in the figure is the number of model calls of the optimization method model of the application.

[0044] Reference signs:

[0045] 1, front fan; 2, rear fan; 3, mode selection valve; 4, front duct ejector; 5, outer duct nozzle; 6, rear duct ejector; 7, main nozzle; 8, core engine driven fan; 9, compressor; 10, main combustion chamber; 11, high pressure turbine; 12, low pressure turbine; 13, afterburner; 14, first outer duct; 15, second outer duct; 16, third outer duct. DETAILED DESCRIPTION

[0046] The technical method of the application is further described below by means of the drawings and examples. It should be noted that: unless otherwise specifically stated, the relative arrangement, numerical expression and numerical value of the components and steps set forth in these examples do not limit the scope of the application.

[0047] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way limiting to the scope of the application or its applications or uses.

[0048] Techniques, systems, and devices known to those of ordinary skill in the relevant art can not be discussed in detail herein, but should be considered as part of the specification, where appropriate.

[0049] In all examples shown and discussed herein, any specific values should be interpreted as merely exemplary, and not as a limitation. Thus, other examples of the exemplary embodiments can have different values.

[0050] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0051] The present application is directed to the essence of the problem that the existing engine control law optimization design method is difficult to balance between optimization accuracy and optimization efficiency, learns and absorbs the potential technical features in the current mainstream research, fully develops the core advantages of the existing engine performance proxy model technology, and adaptively improves the optimization algorithm, and proposes a proxy model assisted optimization method with population pre-screening and optimization space reduction characteristics.

[0052] Embodiment one

[0053] The specific technical solutions of the present application are as follows:

[0054] S1, according to the characteristics of the engine, the engine control law variables participating in the control law optimization design are defined as independent variables, and the preliminary optimization interval IV of the independent variables is determined range ;

[0055] Preferably, the control law variables in S1 include shaft speed and adjustable geometry variables, and the preliminary optimization interval is the upper and lower bounds of the control law variables, expressed as:

[0056]

[0057] Wherein, m is the number of control law variables.

[0058] S2, taking the independent variables and engine operating conditions as input variables, taking the engine performance as output variable, generating sample set based on engine overall performance simulation model and training to generate engine overall performance proxy model;

[0059] Wherein, the proxy model construction method used for training can select one of Kriging, radial basis function interpolation, support vector regression and artificial neural network.

[0060] Taking the independent variables and engine operating conditions as input variables, taking the guess value vector X obtained by iterative solution convergence of the engine simulation model as output variable, generating sample set based on engine overall performance simulation model and training to generate auxiliary proxy model;

[0061] Wherein, the proxy model construction method used for training can select one of Kriging, radial basis function interpolation, support vector regression and artificial neural network.

[0062] The engine overall performance proxy model and the generated auxiliary proxy model are two independent models, one of which predicts performance and one of which predicts a guess value vector.

[0063] S3, preliminary optimization interval IV based on independent variables range In combination with the engine overall performance proxy model, the creation and pre-screening of the initial population of the optimization algorithm are performed to obtain a pre-screened initial population.

[0064] Preferably, the specific content of S3, in which the preliminary optimization interval IV based on the given independent variables is combined with the engine overall performance proxy model, for the creation and pre-screening of the initial population of the optimization algorithm, is as follows:

[0065] S301, using the Latin hypercube sampling method, a pre-screening population containing N*P individuals is generated in the preliminary optimization interval IV of the independent variables, wherein N represents the pre-screening multiple, and P represents the number of individuals in the standard optimization population. range

[0066] S302, according to the optimization target and the constraint condition, a target fitness function evaluation formula Fit=f(P) is constructed, wherein P is the engine overall performance, and Fit is the individual target fitness.

[0067] For different optimization requirements, the optimization target may be to maximize the thrust, or to minimize the specific fuel consumption, etc., the constraint condition may be different for different engine types, and the construction of the target fitness function is also different.

[0068] The method itself mainly lies in the framework of the optimization algorithm, and there is no limitation on these aspects;

[0069] S303, based on the engine overall performance proxy model, the engine performance corresponding to each individual in the population is calculated, and then the target fitness of all individuals in the population is evaluated based on the target fitness function evaluation formula to obtain the target fitness value.

[0070] S304, the individuals in the pre-screening population are sorted from low to high according to the target fitness value, and the P individuals with high ranking are selected as the initial population of the optimization algorithm.

[0071] S4, according to the individual characteristics of the pre-screened initial population, the preliminary optimization interval is reduced to obtain a reduced optimization interval.

[0072] Preferably, the specific content of S4, in which the preliminary optimization interval is reduced to obtain a reduced optimization interval according to the individual characteristics of the pre-screened initial population, is as follows:

[0073] According to the extreme values of the independent variables in the individuals of the pre-screened initial population, i.e., the individual characteristics, the upper and lower bounds of the preliminary optimization interval are updated, and a relaxation term R is introduced j ​ / k is adjusted, wherein the updating expression is:

[0074]

[0075] IV j max(min(IV 1,j ,IV 2,j ,…,IV P,j )-R j / k, IV j ),j=1,2,…,m

[0076]

[0077] wherein, is a new upper bound of the jth independent variable, IV j is a new lower bound of the jth independent variable, IV 1,j is the jth independent variable of the first individual, IV 2,j is the jth independent variable of the second individual, IV P,j is the jth independent variable of the Pth individual, R j is a preliminary optimization interval length of the jth independent variable, and k is an equal division fraction of the optimization interval length, generally taking about 20.

[0078] Preferably, the expression for reducing the optimization interval is:

[0079]

[0080] wherein, range is the reduced optimization interval, IV m is a new lower bound of IV m after reduction, is a new upper bound of IV m after reduction.

[0081] S5. Based on the pre-screening initial population and the reduced optimization interval, a target fitness-driven population optimization algorithm (such as genetic algorithm, differential evolution algorithm, particle swarm optimization algorithm, etc.) is selected to perform an optimization search step of the algorithm to complete the auxiliary optimization of the engine overall performance proxy model in combination with the auxiliary proxy model.

[0082] Preferably, the specific content in the optimization search step of the algorithm in S5 is:

[0083] The auxiliary proxy model is used to predict the guess value vector X corresponding to the independent variables of the new individual;

[0084] The guess value vector X is used as the initial guess value of the engine performance simulation model to obtain the engine performance and calculate the target fitness value according to the target fitness function Fit=f(P);

[0085] When the termination condition of the optimization algorithm is met, the optimization process ends.

[0086] There are two main reasons for the time-consuming optimization of the conventional engine control law: first, the iterative convergence process of the optimization algorithm is slow due to the unsatisfactory target fitness of the initial population and the large optimization interval, and it takes dozens to hundreds of iterations to converge to the optimal solution. Second, the evaluation of the target fitness depends on the time-consuming model simulation process, and if the calculation speed of the simulation model cannot be significantly improved, the optimization efficiency is also difficult to effectively improve.

[0087] The present application proposes two key improvement measures: population pre-screening and space reduction, to effectively solve the problem of slow iterative convergence of the optimization algorithm. Specifically, the fitness of the population individuals is calculated by the proxy model of the overall performance of the engine in the pre-screening stage. This method has very low calculation cost and allows to handle a number of individuals far exceeding the standard optimization population size, thereby achieving high-density coverage in the sample optimization space. On this basis, the individuals are sorted and screened according to the target fitness to obtain an initial population with ideal target fitness. In addition, individuals with superior target fitness usually have similar independent variable value characteristics and are concentrated in the vicinity of the optimal solution. Based on this characteristic information, the optimization space can be effectively reduced, significantly reducing the range of subsequent optimization search, thereby speeding up the iterative convergence process. To avoid the prediction error introduced by directly using a proxy model to replace the simulation model, the present application uses a proxy model to accelerate the engine performance simulation model. The core principle is to significantly reduce the number of iterative calculations of the simulation model by providing a reasonable initial guess, thereby reducing the calculation cost of target fitness evaluation.

[0088] The optimization algorithm proposed in the present application uses a variable credibility strategy. In the early stage of optimization, a low-credibility proxy model is used to guide the population to locate near the optimal solution; subsequently, a high-precision proxy model is used to assist the simulation model to further explore the region where the optimal solution is located. This strategy not only significantly improves the optimization efficiency, but also maintains the robustness and accuracy of the algorithm. It is worth noting that the engine performance proxy model and the auxiliary proxy model required by the optimization algorithm only need to be trained once and can be repeatedly applied to the optimization process of different working conditions and optimization targets. Only when new control law variables are added, the optimization interval is expanded, or the engine design parameters are changed, the model needs to be retrained.

[0089] Embodiment two

[0090] In order to facilitate the understanding of the above technical solutions of the present application, the above technical solutions of the present application will be described in detail through specific embodiments.

[0091] Figure 2A schematic diagram of an adaptive cycle aeroengine is given, showing its main components and ducted structure. As an engine with the ability of cycle regulation as its distinctive feature, it can achieve the change distribution of cycle energy and flow and the switching of two working modes through the control combination of numerous adjustable geometry components, so as to adapt to diversified combat environments and task requirements. However, with the increase of control law variables, the determination of the control law becomes more complex and challenging. If the optimization algorithm can achieve good results on the adaptive cycle engine, it is also expected to achieve excellent performance in other engine configurations with comparable or more simplified complexity, thereby effectively verifying the wide applicability and reliability of the technical solutions proposed in the present application.

[0092] For the adaptive cycle engine, the main control law variables include: engine low-pressure shaft relative physical speed NL, rear fan adjustable stator blade angle α RFan , core engine driven fan adjustable stator blade angle α CDFS , compressor adjustable stator blade angle α HPC , low-pressure turbine guide vane angle α LPT , main nozzle throat area change percentage A8, front duct ejector area change percentage A FVABI and rear duct ejector area change percentage A RVABI . Taking its ground take-off (0Ma0km) dual-outer duct mode as an example, the control law optimization is carried out to obtain the corresponding control law variable combination that maximizes the thrust under the condition of meeting the constraints. The preliminary selected optimization interval of each variable in the dual-outer duct mode of the engine is shown in Table 1:

[0093] Table 1 Preliminary optimization interval of key control law variables of adaptive cycle engine

[0094]

[0095]

[0096] In terms of optimization strategy selection, on the one hand, the population pre-screening and space reduction proxy model assisted genetic optimization algorithm (GA+KDSM) proposed in the present application is adopted, and on the other hand, only the traditional unimproved genetic algorithm (GA) is used for optimization effect comparison.

[0097] For the GA+KDSM method, the pre-screening factor N is set to 50 and the population size P is set to 1500. In the first training process, the total time for training the engine overall performance surrogate model and the auxiliary surrogate model is about 15 minutes, which is reasonable and efficient compared to the batch multi-point control law optimization method. Table 2 shows the reduced optimization space, and the results show that the optimization space is significantly reduced, especially for the low-pressure shaft relative physical speed and the main nozzle throat area percentage change, which are the two parameters that have a greater impact on the engine thrust performance. The length of the change interval is shortened by 19.44 times and 6.11 times, respectively, which will significantly speed up the subsequent optimization search process. For the traditional GA method, the population size P is also set to 1500 to ensure consistency.

[0098] Table 2 Reduced optimization space of key control law variables of adaptive cycle engine

[0099]

[0100] Figure 3 The optimal individual fitness variation trend comparison of two optimization strategies in completing a control law optimization process is shown. As can be seen from the figure, compared with the traditional genetic optimization algorithm, the optimization algorithm proposed in the present application starts optimization search with a lower optimal individual fitness at the beginning of iteration, and has a faster convergence speed in subsequent iterations. Specifically, the convergence stop number of the traditional genetic optimization algorithm is 32 generations, while the optimization algorithm of the present application only needs 15 generations, achieving more than twice the calculation efficiency improvement. In terms of optimization results, both algorithms can obtain global optimal solution, and the relative errors of thrust and specific fuel consumption obtained are only 0.068% and 0.092%, respectively.

[0101] Further, Figure 4 The number of model calls in the optimization search process of the two optimization algorithms is compared. For the optimization algorithm proposed in the present application, the engine performance surrogate model is used to calculate the fitness when creating the initial population, which avoids calling the time-consuming engine performance simulation model; in the subsequent optimization iteration process, the number of model calls per generation is also much lower than that of the traditional unimproved optimization algorithm. Overall, until the optimization iteration converges, the traditional optimization algorithm calls the engine simulation model about 5.1675 million times, while the optimization algorithm of the present application only calls 338.5 thousand times. Therefore, in terms of optimization time, the traditional optimization algorithm takes about 28 minutes to optimize the single-point control law, while the optimization algorithm of the present application only takes about 2 minutes, significantly improving the optimization efficiency.

[0102] The embodiment provides an agent model aided optimization method with population pre-screening and optimization space reduction characteristics. The method firstly performs pre-screening on population individuals by using the extremely low calculation cost advantage of the agent model, and then reduces the optimization space based on the characteristics of the pre-screened individuals. Subsequently, subsequent operations are based on the pre-selected individuals and the reduced optimization space, so that the goal of fast optimization convergence is achieved. On the other hand, in the optimization process, the method fully utilizes the acceleration auxiliary role of the agent model in the overall performance simulation of the engine, reduces the evaluation cost of the objective function, and further reduces the optimization calculation amount. This not only improves the optimization efficiency, but also guarantees the accuracy of the original overall performance simulation model.

[0103] Finally, it should be noted that the above embodiments are only used to illustrate the technical method of the present application, but not to limit it. Although the present application has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical method of the present application can still be modified or replaced equivalently, and these modifications or equivalent replacements cannot make the modified technical method deviate from the spirit and scope of the technical method of the present application.

Claims

1. A proxy model assisted optimization method for population pre-screening and spatial reduction, characterized in that, The method comprises the following steps: S1, according to the engine type characteristics, the engine control law variable participating in the control law optimization design is selected as the independent variable, and the preliminary optimization interval of the independent variable is determined ; S2, training an engine overall performance proxy model with independent variables and engine operating conditions as input variables and engine performance as output variables; S3, training an auxiliary proxy model with independent variables and engine operating conditions as input variables and the guess vector X obtained by iterative solution convergence of the engine simulation model as output variables; S3, preliminary optimization interval based on independent variable In combination with the engine overall performance proxy model, the initial population of the optimization algorithm is created and pre-screened to obtain a pre-screened initial population; S4, reducing the preliminary optimization interval to obtain a reduced optimization interval according to the individual characteristics of the pre-screened initial population; S5, selecting a target fitness driven population optimization algorithm based on the pre-screened initial population and the reduced optimization interval, and performing an optimization search step of the algorithm to complete auxiliary optimization of the engine overall performance proxy model in combination with the auxiliary proxy model; In S4, the specific content of reducing the preliminary optimization interval to obtain a reduced optimization interval according to the individual characteristics of the pre-screened initial population is: According to the extreme value of the independent variable in the individual of the pre-screening initial population, i.e. the individual characteristics, the upper and lower bounds of the preliminary optimization interval are updated, and a relaxation term is introduced is adjusted, wherein the update expression is: ； wherein, new upper bound for the jth independent variable, new lower bound for the jth independent variable, the jth independent variable of the first individual, the jth independent variable of the second individual, the jth independent variable of the Pth individual, preliminary optimization interval length for the jth independent variable, k is the number of equal partitions of the optimization interval length. In the optimization search step of the algorithm in S5, the specific content is: Using the auxiliary proxy model to predict the guess vector X corresponding to the independent variables of the new individual; Taking it as the initial guess value of the engine performance simulation model, solving the engine performance and calculating the target fitness value according to the target fitness function Fit=f(P); When the termination condition of the optimization algorithm is met, the optimization process ends; In S1, the control law variables include shaft speed and adjustable geometry variables, and the preliminary optimization interval is the upper and lower bounds of the control law variables, expressed as: ； wherein m is the number of arguments, is the first argument, is the second argument, is the mth argument, is the lower bound of is the upper bound of .​ 2. The method of claim 1, wherein, In S3, the specific content of creating and pre-screening the initial population of the optimization algorithm based on the given independent variable preliminary optimization interval and in combination with the engine overall performance proxy model is: S301、Adopting Latin hypercube sampling method in the preliminary optimization interval of independent variables generate a pre-screening population of individuals containing a pre-screening population of individuals, wherein denotes the pre-screening multiple, denotes the number of individuals of the standard optimization population; S302, constructing a target fitness function evaluation formula Fit=f(P) according to the optimization target and the constraint condition, wherein P is the engine overall performance and Fit is the individual target fitness; S303, calculating the engine performance corresponding to each individual in the population based on the engine overall performance proxy model, and then evaluating the target fitness of all individuals in the population based on the target fitness function evaluation formula to obtain the target fitness value; S304, sorting the individuals in the pre-screened population from low to high according to the target fitness value, and selecting the P individuals with high ranking of the target fitness value as the initial population of the optimization algorithm.

3. The method of claim 1, wherein, The expression of the reduced optimization interval is: ； wherein, is a new lower bound of the reduction of is a new upper bound of the reduction of is a new upper bound of the reduction of

Images