Solid rocket engine overall performance rapid calculation method based on isomorphism scheme knowledge transfer

Abstract

The application discloses a kind of heterogeneous scheme knowledge transfer's solid engine overall performance rapid calculation method, comprising the following steps: based on similarity measurement criterion, select multiple solid rocket engine design source tasks in knowledge base;Based on the selected source task, construct the low-precision model of target domain, and optimize the low-precision model;Based on the low-precision model after optimization, the overall performance of solid engine is rapidly calculated.The application is applied to the technical field of solid engine, by drawing knowledge from solid rocket engine design library source domain, construct the approximate model based on knowledge transfer instead of conventional approximate model, effectively utilize prior knowledge, avoid a large number of sampling and simulation calculation, improve the modeling efficiency, thereby save initial calculation resource.

Description

Technical Field

[0001] This invention relates to the field of solid rocket motor technology, specifically a method for rapid calculation of the overall performance of a solid rocket motor based on knowledge transfer from heterogeneous schemes. Background Technology

[0002] Solid rocket engines are widely used in the propulsion systems of missiles, rockets, and other spacecraft. The quality of engine modeling directly affects the results of subsequent optimization design, thus impacting rocket performance. Engine modeling is a crucial component of engine design and one of the most challenging techniques. Its main task is to obtain output values ​​through extensive sampling and high-precision simulation within the design domain, and then use surrogate modeling techniques to construct a predictive surrogate model of the real model to meet the requirements of subsequent optimization design.

[0003] Currently, the commonly used design and modeling methods for solid rocket motors mainly employ the uniform experimental design approach. This involves extensive sampling and simulation within the design domain to obtain a large number of high-fidelity sample points. Then, based on these high-fidelity sample points, a surrogate modeling technique is used to construct a predictive surrogate model for the actual simulation model, which is then used for subsequent search and optimization. However, while a specific problem requires extensive sampling and simulation, a different design requirement necessitates re-sampling and simulation, consuming significant computational resources. Furthermore, insufficient local accuracy of the model leads to unreliable results in the final optimization. Summary of the Invention

[0004] To address the problems of low efficiency, large sample size, and wasted computational resources in the design and modeling of solid rocket engines in the existing technologies, which require resampling and calculation for different internal ballistic parameters, this invention provides a method for rapid calculation of the overall performance of solid rocket engines based on heterogeneous scheme knowledge transfer. This method is based on surrogate model-driven knowledge transfer modeling and utilizes past design knowledge to achieve rapid calculation of the overall performance of solid rocket engines.

[0005] To achieve the above objectives, this invention provides a method for rapid calculation of the overall performance of a solid rocket motor based on knowledge transfer from heterogeneous schemes, comprising the following steps:

[0006] Step 1: Based on the similarity measurement criteria, select multiple source missions for solid rocket engine designs from the knowledge base;

[0007] Step 2: Construct a low-precision model of the target domain based on the selected source task, and optimize the low-precision model;

[0008] Step 3: Based on the optimized low-precision model, quickly calculate the overall performance of the solid rocket motor.

[0009] In one embodiment, in step 2, the low-precision model specifically refers to:

[0010]

[0011] Among them, f L (x) represents the low-precision model, x represents the sample points, n represents the number of source tasks, and p j f represents the weight of the j-th source task. j (x) represents the global model of the j-th source task, and λ represents the hyperparameter.

[0012] In one embodiment, during the construction of the low-precision model, a global model of the source task is constructed using the sample radial basis function method, as follows:

[0013]

[0014] Where, N j ω represents the number of sample points contained in the j-th source task. i Represents the basis function coefficients. This represents the Gaussian function.

[0015] In one embodiment, during the construction of the global model of the source task, σI is superimposed on the diagonal of the coefficient matrix of the basis function coefficients, i.e.:

[0016] Φ′=Φ+σI

[0017] Where Φ represents the original coefficient matrix, Φ′ represents the coefficient matrix after superposition of σI, σ represents the smoothing factor, and I represents the identity matrix.

[0018] In one embodiment, the smoothing factor σ = 0.001.

[0019] In one embodiment, the weight of the j-th source task is specifically:

[0020]

[0021] Where, q j This represents the optimization coefficient.

[0022] In one embodiment, step 2, the process of optimizing the low-precision model, is as follows:

[0023] Using minimizing the root mean square error of the low-precision model's predictions on source domain samples as the objective function, the optimal hyperparameter λ and the optimization coefficient q are searched. j ,Right now:

[0024]

[0025] Where R(q) j ,λ) represents the prediction root mean square error as a function of the hyperparameter λ and the optimization coefficient q.j Related functions, N i x represents the number of sample points contained in all source tasks. i f represents the i-th sample point among all sample points contained in all source tasks. L (x i ) indicates that the low-precision model is accurate for sample point x. i The output predicted value, y(x) i ) represents the sample point x i The corresponding actual output.

[0026] In one embodiment, during the optimization of the low-precision model, a particle swarm optimization algorithm is used to calculate the hyperparameter λ and the optimization coefficient q. j Wherein, the hyperparameter λ takes values ​​ranging from 1 to λ to 4, and the optimization coefficient q j The range of values ​​is 0. j <1.

[0027] In one embodiment, step 1, the process of selecting the source task, specifically involves:

[0028] The similarity between the target mission of the current solid rocket motor design and each mission in the knowledge base is calculated as follows:

[0029] S iT =exp(-d p,iT ), i = 1, 2, ..., k

[0030] d p,iT =||m i ,m T || p

[0031] Among them, S iT Let m represent the similarity between the target task and the i-th task in the knowledge base. i Let m represent the meta-feature of the i-th task in the knowledge base. T d represents the meta-features of the target task. p,iT Let ||m| represent the p-norm of the target task and the i-th task in the knowledge base. i ,m T || p This represents calculating the p-norm of the target task and the i-th task in the knowledge base, where k represents the number of tasks in the knowledge base;

[0032] The n tasks with the highest similarity to the target task are selected from the knowledge base as the source tasks.

[0033] Compared with the prior art, the present invention has the following beneficial technical effects:

[0034] ​1. This invention selects similar solid rocket motor design source tasks through similarity measurement and transfers knowledge from them, aiming to build a low-precision prediction model for the target domain. It effectively utilizes prior knowledge, avoids a large amount of sampling and simulation calculations, improves modeling efficiency, and thus saves initial computing resources.

[0035] 2. The low-precision prediction model constructed in this invention can improve local accuracy through continuous sampling in subsequent optimization, which can effectively meet the optimization requirements. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0037] Figure 1 This is a flowchart of a method for rapid calculation of the overall performance of a solid rocket motor based on knowledge transfer from heterogeneous schemes, as described in this invention.

[0038] Figure 2 This is a schematic diagram of the propellant loading geometry of the engine wing-pillar type propellant grain in an embodiment of the present invention, wherein: (a) is an axial sectional view and (b) is a side view.

[0039] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0040] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0041] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0042] Furthermore, in this invention, descriptions involving "first," "second," etc., are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0043] In this invention, unless otherwise explicitly specified and limited, the terms "connection," "fixed," etc., should be interpreted broadly. For example, "fixed" can mean a fixed connection, a detachable connection, or an integral part; it can mean a mechanical connection, an electrical connection, a physical connection, or a wireless communication connection; it can mean a direct connection or an indirect connection through an intermediate medium; it can mean the internal communication of two elements or the interaction between two elements, unless otherwise explicitly limited. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0044] Furthermore, the technical solutions of the various embodiments of the present invention can be combined with each other, but only if they are based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0045] This embodiment discloses a rapid calculation method for the overall performance of solid rocket motors based on knowledge transfer from heterogeneous schemes. The method first proposes a mission similarity evaluation criterion. By calculating the similarity between the source and target missions of each solid rocket motor design, multiple missions are selected as source missions for knowledge transfer while avoiding negative transfer. A weighted low-precision model of the target domain is constructed using the selected solid rocket motor designs. A radial basis function is employed, and a particle swarm optimization algorithm is proposed to obtain the error model between the low-precision approximation model and its source domain samples, thereby constructing an approximate model of the target mission and improving the model's generalization performance. Finally, the overall performance of the solid rocket motor is rapidly calculated based on the constructed low-precision approximation model.

[0046] refer to Figure 1 The method for rapid calculation of the overall performance of a solid rocket motor based on knowledge transfer of heterogeneous schemes in this embodiment specifically includes the following steps:

[0047] Step 1: Based on the similarity measurement criteria, select multiple source missions for solid rocket engine designs from the knowledge base;

[0048] Step 2: Construct a low-precision model of the target domain based on the selected source task, and optimize the low-precision model;

[0049] Step 3: Based on the optimized low-precision model, the overall performance of the solid rocket motor is calculated quickly.

[0050] In the design of solid rocket motors, performance parameters such as maximum thrust, thrust curve, impulse ratio, burn time, total mass, and combustion efficiency all characterize the overall performance of the solid rocket motor. Therefore, in this embodiment, one or more of the following parameters are used as objective functions in the solid rocket motor design process: maximum thrust, thrust curve, impulse ratio, burn time, total mass, and combustion efficiency. That is, after substituting sample points into a low-precision model, the output of the low-precision model is the predicted value of the set performance parameters, while the actual output of the sample points is the performance parameter obtained from actual simulation of those sample points.

[0051] In this embodiment, the design parameters of solid rocket motors are divided into two categories: design variables and meta-features. Design variables are generally configuration parameters, mainly composed of the propellant configuration parameters and nozzle size parameters of the solid rocket motor. Meta-features are parameters specified by the design task, affecting the engine simulation model; the same design variable will output different results under different meta-features. A sample point includes one or more design variables, such as the aft wing length, aft wing width, depth, and aft wing tilt angle of a styloidal propellant charge.

[0052] In this embodiment, the source domain refers to the completed design task, and the target domain is the design task that needs to be performed now. The relationship between the target domain and the source domain is expressed through meta-features measured by a similarity criterion. The solid rocket engine design source task requiring knowledge transfer is selected, and its specific implementation process is as follows:

[0053] First, for k tasks T1, T2, ..., T in the knowledge base k Extract these tasks and target task T T The normalized features are {m1, m2, ..., m} k} and m T Among them, the normalized element feature can be selected from one or more of the design task-specified parameters such as the outer diameter of the propellant, the length of the propellant, and the number of rear wings of the solid rocket engine;

[0054] Secondly, after extracting the normalized features, the similarity between the target mission of the current solid rocket motor design and each mission in the knowledge base is calculated as follows:

[0055]

[0056] Among them, S iT Let m represent the similarity between the target task and the i-th task in the knowledge base. i Let d represent the meta-feature of the i-th task in the knowledge base. p,iTLet p be the norm of the target task and the i-th task in the knowledge base, where p is typically 2, and ||m i ,m T || p Let p be the p-norm of the target task and the i-th task in the knowledge base, and k be the number of tasks in the knowledge base.

[0057] Finally, n tasks with the highest similarity to the target task are selected from the knowledge base as source tasks. In this embodiment, the number of source tasks selected is set to 3 to 5.

[0058] In this embodiment, the low-precision model of the target domain constructed based on the selected source task is as follows:

[0059]

[0060] Where x represents the sample point, f L (x) represents the predicted output of the objective function value of the low-precision model for sample point x, where n represents the number of source tasks, and p j f represents the weight of the j-th source task. j (x) represents the global model of the j-th source task, and λ represents the hyperparameter.

[0061] In constructing the low-precision model, the sample radial basis function method is used to construct the global model of the source task. The radial basis functions are weighted and superimposed using simple odd functions to predict new sample points, as shown below:

[0062]

[0063] Where, N j ω represents the number of sample points contained in the j-th source task. i Represents the basis function coefficients. The Gaussian function, which takes the Euclidean distance from an unknown sample point to a known sample point as the independent variable, takes the following form:

[0064]

[0065] Where, λ i This refers to the shape parameter. The shape parameter value is λ. i Once determined, in order to calculate the corresponding basis function coefficients ω i Using interpolation conditions or least squares fitting, N j Substituting each sample point into the basic form (3) of the approximate model, and introducing interpolation conditions... The system of linear equations relating to the basis function coefficients is as follows:

[0066]

[0067] By solving the above system of linear equations, the basis function coefficient vector ω can be obtained as follows:

[0068] ω=Φ -1 y (6)

[0069] Where ω represents the basis function coefficient vector, Φ represents the coefficient matrix calculated by substituting all sample inputs in the radial basis method, each element in the matrix is ​​calculated using equation (5), and y represents the objective function value vector of all sample points.

[0070] Typically, overly dense sample points can lead to Runge's phenomenon during radial basis function model training, resulting in decreased model accuracy. Therefore, in this embodiment, σI is superimposed on the diagonal of the coefficient matrix Φ, i.e.:

[0071] Φ′=Φ+σI (7)

[0072] Where Φ represents the original coefficient matrix, Φ′ represents the coefficient matrix after superposition of σI, σ represents the smoothing factor, and I represents the identity matrix. In this embodiment, the smoothing factor is preferably σ = 0.001.

[0073] In this embodiment, the weight of the j-th source task is specifically as follows:

[0074]

[0075] Where, q j This represents the optimization coefficient.

[0076] In this embodiment, optimizing the low-precision model specifically refers to optimizing the hyperparameter λ and the optimization coefficient q in the low-precision model. j , specifically,

[0077] The objective function is to minimize the root mean square error of the low-precision model's predictions on the source domain samples, and to search for the optimal hyperparameters λ and optimization coefficients q. j ,Right now:

[0078]

[0079] Where R(q) j ,λ) represents the prediction root mean square error as a function of the hyperparameter λ and the optimization coefficient q. j Related functions, N i x represents the number of sample points contained in all source tasks. i f represents the i-th sample point among all sample points contained in all source tasks. L (x i ) indicates a low-precision model for sample point x i The output predicted value, y(x) i ) represents the sample point x i The corresponding actual output.

[0080] In the specific implementation process, the particle swarm optimization algorithm is used to calculate the hyperparameter λ and the optimization coefficient q. j The hyperparameter λ takes values ​​ranging from 1 to λ to 4, and the optimization coefficient q j The range of values ​​is 0. j <1.

[0081] After optimizing the low-precision model, the overall performance of the solid rocket motor can be calculated quickly. For example, for sample points x in the target domain... * The performance prediction requirement is input into the optimized low-precision model to obtain the corresponding performance parameter prediction result f. L (x * ).

[0082] The following section provides a further explanation of the rapid calculation method for the overall performance of a solid rocket motor using knowledge transfer from heterogeneous schemes in this embodiment, with specific examples.

[0083] To address the problem of rapidly calculating the overall performance of solid rocket motors, a similarity metric is first used to assess the similarity between the source and target tasks. Then, 3-5 similar tasks are selected for knowledge transfer modeling to construct a prior low-precision model of the target task. Finally, the overall performance of the solid rocket motor is rapidly calculated based on this prior low-precision model. The specific steps are as follows:

[0084] 1) Select the solid rocket engine design source tasks that require knowledge transfer using similarity measurement criteria. Usually, three tasks are selected for knowledge transfer.

[0085] 2) Utilize the selected solid rocket motors to design a weighted low-precision model of the target domain;

[0086] 3) Construct a solid rocket motor design model using the radial basis function method;

[0087] 4) The particle swarm optimization algorithm is used to find the optimal weight coefficients and hyperparameters;

[0088] 5) A low-precision model for solid rocket motor design was derived;

[0089] 6) Perform rapid calculations on the overall performance of solid rocket motors.

[0090] Taking the design and modeling of a certain configuration of engine wing-stave type propellant grain as an example, its propellant geometry configuration is as follows: Figure 2 As shown. The variable parameters for establishing the model are given. The design variables are the geometric parameters of the charge, including the length L12 of the rear wing styloft, the width bw of the rear wing, the depth R7, and the rear wing tilt angle α; the element-value is the outer diameter D of the charge. p ​The parameters are: charge length L1, number of rear wings b; other parameters include the charge inner diameter d. p Set it as a constant. The range of variables is shown in Table 1.

[0091] Table 1. Design variables and ranges for front and rear wing-shaped charge configurations.

[0092]

[0093] Considering the complexity of the problem and the computational cost of sample evaluation, eight tasks with 50 uniformly distributed samples were generated to populate the knowledge base. The meta-features of these tasks were randomly generated in the parameter space, and their specific values ​​are shown in Table 2. The meta-features of the target task and the similarity values ​​(calculated by normalizing the meta-features) are also shown in the table.

[0094] Table 2 Knowledge Base and Target Task Settings

[0095]

[0096] Three sources with the highest similarity values ​​were selected (①, ⑦, and ⑧, according to Table 2) as source tasks. A low-precision model of the engine wing-stave type propellant grain was constructed. Accuracy verification was performed by randomly sampling eight sample points within the design domain, and the errors are shown in Table 3.

[0097] Table 3. Relative Errors Between Inaccurate and True Model Outputs

[0098]

[0099]

[0100] Based on the comparison between the low-precision surrogate model constructed using this method and the output of the target task, it can be concluded that the solid rocket engine model constructed using the knowledge transfer modeling method can transfer knowledge from the source task, avoid a large amount of simulation calculation, improve modeling efficiency, and provide guidance for subsequent optimization.

[0101] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural transformations made using the contents of the present invention's specification and drawings under the inventive concept of the present invention, or direct / indirect applications in other related technical fields, are included within the patent protection scope of the present invention.

Claims

1. A method for rapid calculation of the overall performance of a solid rocket motor based on knowledge transfer from heterogeneous schemes, characterized in that, Includes the following steps: Step 1: Based on the similarity measurement criteria, select multiple source missions for solid rocket motor designs from the knowledge base; Step 2: Construct a low-precision model of the target domain based on the selected source task, and optimize the low-precision model. Specifically, the low-precision model is as follows: in, f L ( x ) indicates a low-precision model. x Represents sample points, n Indicates the number of source tasks. p j Indicates the first j The weight of each source task, f j ( x ) indicates the first j A global model for each source task. λ Indicates hyperparameters; In constructing the low-precision model, the global model of the source task is constructed using the sample radial basis function method, as follows: in, N j Indicates the first j The number of sample points contained in each source task ω i Represents the basis function coefficients. φ i ( r ) represents the Gaussian function; No. j The specific weights of each source task are as follows: in, q j Represents the optimization coefficient; The process of optimizing the low-precision model is as follows: The optimal hyperparameters are searched with the objective function of minimizing the root mean square error of the low-precision model on the source domain samples. λ With the optimization coefficient q j ,Right now: in, R ( q j , λ ) indicates that the root mean square error of the prediction is equal to the hyperparameter. λ With the optimization coefficient q j Related functions, N i This represents the number of sample points contained in all source tasks. x i Represents the first of the sample points contained in all source tasks. i One sample point, f L ( x i ) indicates that the low-precision model is accurate for sample points. x i The output predicted value, y ( x i ) represents sample points x i The corresponding actual output; Step 3: Based on the optimized low-precision model, quickly calculate the overall performance of the solid rocket motor.

2. The method for rapid calculation of overall performance of solid rocket motors based on knowledge transfer of heterogeneous schemes according to claim 1, characterized in that, In the process of constructing the global model of the source task, the diagonal superposition of the coefficient matrix of the basis function coefficients is performed. ,Right now: in, This represents the original coefficient matrix. Indicates superposition The coefficient matrix after that, Let I represent the smoothing factor, and let I represent the identity matrix.

3. The method for rapid calculation of overall performance of solid rocket motors based on knowledge transfer of heterogeneous schemes according to claim 2, characterized in that, The smoothing factor =0.

001.

4. The method for rapid calculation of overall performance of solid rocket motors based on knowledge transfer of heterogeneous schemes according to claim 1, 2, or 3, characterized in that, In the process of optimizing the low-precision model, the particle swarm optimization algorithm is used to calculate the hyperparameters. λ With the optimization coefficient q j The hyperparameters λ The range of values ​​for is 1≤ λ ≤4, the optimization coefficient q j The range of values ​​is 0 < q j <1.

5. The method for rapid calculation of overall performance of solid rocket motors based on knowledge transfer of heterogeneous schemes according to claim 1, 2, or 3, characterized in that, In step 1, the process of selecting the source task is as follows: The similarity between the target mission of the current solid rocket motor design and each mission in the knowledge base is calculated as follows: in, S iT This indicates that the target task and the knowledge base are related. i The similarity of the tasks, m i Indicates the knowledge base number i Meta-features of each task, m T Meta-features representing the target task d p,iT This indicates that the target task and the knowledge base are related. i One task p Norm, This indicates that the computation target task is related to the knowledge base number 1. i One task p Norm, k This indicates the number of knowledge base tasks; Select from the knowledge base n The task with the highest similarity to the target task is selected as the source task.

Images