Hybrid power dynamics modeling method and system based on long short-term memory neural network

Abstract

The present invention discloses a hybrid power dynamics modeling method and system based on a long short-term memory neural network, which relates to the field of modeling technology, constructs a hybrid power dynamics model of a flexible spacecraft; adjusts the physical parameters of the dynamics model within a set threshold range to obtain a plurality of adjusted dynamics models; performs programming calculations on the hybrid power dynamics model and the plurality of adjusted dynamics models to obtain a reference parameter matrix and a plurality of difference parameter matrices; subtracts each difference parameter matrix from the reference parameter matrix to obtain a difference value set; trains a long short-term memory neural network based on each difference parameter matrix and each difference set to obtain a trained long short-term memory neural network; predicts the actual modeling error based on the trained long short-term memory neural network, and then offsets the uncertainty error, improves the modeling accuracy, and provides a good foundation for subsequent vibration control.

Description

Technical Field

[0001] The present invention relates to the field of modeling technology, and in particular to a hybrid power dynamics modeling method and system based on a long short-term memory neural network. Background Art

[0002] With the rapid development of the aerospace industry, the vibration control of flexible spacecraft has become a major research hotspot in the aerospace field. Establishing a mathematical model of the spacecraft is the first and most critical step in vibration control. Therefore, the accuracy of the model directly affects the effectiveness of vibration control. Flexible structures are increasingly used in aerospace vehicles and have become one of the main structural types used in aerospace engineering. Various aerospace structures are also increasingly developing in the direction of larger sizes, lower stiffness, and lighter weight. However, this also brings a series of new challenges to the design, manufacture, and use of spacecraft structures, especially the accuracy of flexible spacecraft modeling. High-precision spacecraft models play a vital role in the effectiveness of spacecraft vibration control. Summary of the Invention

[0003] The purpose of the present invention is to provide a hybrid dynamics modeling method and system based on long short-term memory neural network, which improves the accuracy of modeling by fitting the uncertainty error in modeling.

[0004] To achieve the above object, the present invention provides the following solutions:

[0005] A hybrid dynamics modeling method based on a long short-term memory neural network, comprising:

[0006] Constructing a hybrid dynamics model of the flexible spacecraft; adjusting physical parameters of the dynamics model within a set threshold range to obtain a plurality of adjusted dynamics models;

[0007] Performing programming calculation on the hybrid dynamics model to obtain a reference parameter matrix; performing programming calculation on a plurality of the adjustment dynamics models to obtain a plurality of difference parameter matrices;

[0008] Subtracting each of the difference parameter matrices from the reference parameter matrix to obtain a difference value set;

[0009] Constructing a long short-term memory neural network based on the difference parameter matrix; training the long short-term memory neural network based on each of the difference parameter matrices and each of the difference value sets to obtain the trained long short-term memory neural network;

[0010] constructing an initial dynamic model of the required flexible spacecraft, performing programming calculations on the initial dynamic model, and obtaining a required parameter matrix;

[0011] Obtaining a prediction error value based on the demand parameter matrix and the trained long short-term memory neural network;

[0012] The initial dynamic model is adjusted based on the prediction error value to obtain a dynamic model of the required flexible spacecraft.

[0013] Preferably, the constructing of a hybrid dynamics model of the flexible spacecraft; adjusting the physical parameters of the dynamics model within a set threshold range to obtain a plurality of adjusted dynamics models, includes:

[0014] constructing an initial finite element model of the flexible spacecraft, and designing the initial finite element model according to a set set of physical parameters to obtain a design dynamic model;

[0015] Adjusting the set physical parameter set within the set threshold range to obtain a plurality of adjusted physical parameter sets;

[0016] Designing the initial finite element model according to each of the adjustment physical parameter sets to obtain a plurality of initial adjustment dynamic models;

[0017] Setting fixed surfaces and remote points on the design dynamics model to obtain the hybrid dynamics model;

[0018] Fixed surfaces and remote points are set for each of the initial adjustment dynamic models to obtain each of the adjustment dynamic models.

[0019] Preferably, the reference parameter matrix includes a mass matrix, a damping matrix and a stiffness matrix.

[0020] Preferably, the physical parameters include stiffness, density and Young's modulus.

[0021] Preferably, the long short-term memory neural network includes a forget gate, an input gate and an output gate.

[0022] The present invention also provides a hybrid dynamics modeling system based on a long short-term memory neural network, comprising:

[0023] A model building module is used to build a hybrid dynamics model of the flexible spacecraft and adjust the physical parameters of the dynamics model within a set threshold range to obtain a plurality of adjusted dynamics models;

[0024] a programming calculation module, configured to perform programming calculations on the hybrid dynamics model to obtain a reference parameter matrix, and to perform programming calculations on a plurality of the adjustment dynamics models to obtain a plurality of difference parameter matrices;

[0025] a difference module, configured to respectively perform a difference between each of the difference parameter matrices and the reference parameter matrix to obtain a difference value set;

[0026] A neural network module is used to construct a long short-term memory neural network based on the difference parameter matrix, and train the long short-term memory neural network based on each of the difference parameter matrices and each of the difference value sets to obtain the trained long short-term memory neural network;

[0027] A demand model module is used to construct an initial dynamic model of the demand flexible spacecraft and perform programming calculations on the initial dynamic model to obtain a demand parameter matrix;

[0028] A prediction module, configured to obtain a prediction error value based on the demand parameter matrix and the trained long short-term memory neural network;

[0029] A model adjustment module is used to adjust the initial dynamic model based on the prediction error value to obtain the dynamic model of the required flexible spacecraft.

[0030] Preferably, the model building module is specifically:

[0031] constructing an initial finite element model of the flexible spacecraft, and designing the initial finite element model according to a set set of physical parameters to obtain a design dynamic model;

[0032] Adjusting the set physical parameter set within the set threshold range to obtain a plurality of adjusted physical parameter sets;

[0033] Designing the initial finite element model according to each of the adjustment physical parameter sets to obtain a plurality of initial adjustment dynamic models;

[0034] Setting fixed surfaces and remote points on the design dynamics model to obtain the hybrid dynamics model;

[0035] Fixed surfaces and remote points are set for each of the initial adjustment dynamic models to obtain each of the adjustment dynamic models.

[0036] Preferably, the reference parameter matrix includes a mass matrix, a damping matrix and a stiffness matrix.

[0037] Preferably, the physical parameters include stiffness, density and Young's modulus.

[0038] Preferably, the long short-term memory neural network includes a forget gate, an input gate and an output gate.

[0039] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0040] The present invention discloses a hybrid power dynamics modeling method and system based on a long short-term memory neural network, which constructs a hybrid power dynamics model of a flexible spacecraft; adjusts the physical parameters of the dynamics model within a set threshold range to obtain a plurality of adjusted dynamic models; performs programming calculations on the hybrid power dynamics model and the plurality of adjusted dynamic models to obtain a reference parameter matrix and a plurality of difference parameter matrices; subtracts each difference parameter matrix from the reference parameter matrix to obtain a difference value set; trains the long short-term memory neural network based on each difference parameter matrix and each difference set to obtain a trained long short-term memory neural network; predicts the actual modeling error based on the trained long short-term memory neural network, and then offsets the uncertainty error, improves the modeling accuracy, and provides a good foundation for subsequent vibration control. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0042] Figure 1 This is a flow chart of the hybrid dynamics modeling method based on long short-term memory neural network of the present invention;

[0043] Figure 2 Schematic diagram of a time series with arbitrary difference values;

[0044] Figure 3 Schematic diagram of the training output and test output of the long short-term memory neural network;

[0045] Figure 4 is the error in the LSTM neural network training process;

[0046] Figure 5 is the error in the LSTM neural network testing process;

[0047] Figure 6 This is a structural diagram of the hybrid dynamics modeling system based on the long short-term memory neural network of the present invention.

[0048] Explanation of symbols: 1. Model building module; 2. Programming calculation module; 3. Difference module; 4. Neural network module; 5. Demand model module; 6. Forecasting module; 7. Model adjustment module. DETAILED DESCRIPTION

[0049] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0050] The purpose of the present invention is to provide a hybrid dynamics modeling method and system based on long short-term memory neural network, which improves the accuracy of modeling by fitting the uncertainty error in modeling.

[0051] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0052] Figure 1 This is a flow chart of the hybrid dynamics modeling method based on long short-term memory neural network of the present invention. Figure 1 As shown, the present invention provides a hybrid dynamics modeling method based on a long short-term memory neural network, comprising:

[0053] Step S1: construct a hybrid dynamics model of the flexible spacecraft; adjust the physical parameters of the dynamics model within a set threshold range to obtain a plurality of adjusted dynamics models. In this embodiment, the physical parameters include stiffness, density, and Young's modulus.

[0054] Specifically, step S1 includes:

[0055] Step S11: construct an initial finite element model of the flexible spacecraft. The initial finite element model is designed according to a set of physical parameters to obtain a designed dynamic model. Preferably, the initial finite element model is designed based on the Hamiltonian principle. In this embodiment, the initial finite element model is a nonlinear structural model.

[0056] Step S12: Adjust the set physical parameter set within the set threshold range to obtain a plurality of adjusted physical parameter sets. In this embodiment, the set threshold range is ±1%. The adjusted differences represent the uncertainty in actual modeling, such as parameter error, modeling error, and measurement error.

[0057] Step S13 , designing the initial finite element model according to each of the adjustment physical parameter sets to obtain a plurality of initial adjustment dynamic models.

[0058] Step S14: Setting fixed surfaces and remote points on the designed dynamic model to obtain the hybrid dynamic model. For example, if the flexible spacecraft is a satellite solar panel, the fixed surface is the fixed side of the panel, and the remote points are set based on the condition that the closer to the fixed side, the smaller the offset.

[0059] Step S15 , setting fixed surfaces and remote points for each of the initial adjustment dynamic models to obtain each of the adjustment dynamic models.

[0060] Step S2, performing programming calculation on the hybrid dynamics model to obtain a reference parameter matrix; performing programming calculation on a plurality of the adjustment dynamics models to obtain a plurality of difference parameter matrices.

[0061] In this embodiment, the reference parameter matrix includes a mass matrix, a damping matrix, and a stiffness matrix. The stiffness and Young's modulus determine the stiffness matrix of the hybrid dynamics model, and the density determines the mass matrix of the hybrid dynamics model. Programming calculations are performed based on the Hamilton principle, specifically the number of elements, the number of nodes, and physical parameters of the hybrid dynamics model.

[0062] Step S3, respectively, the difference parameter matrix and the reference parameter matrix are subtracted to obtain a difference set. Any difference in the difference set is a time series data. Figure 2 As shown, Figure 2 In the table, DifferenceData is the difference data, date is the data, Difference is the difference, and Time is the time.

[0063] Step S4: constructing a long short-term memory neural network based on the difference parameter matrix; training the long short-term memory neural network based on each of the difference parameter matrices and each of the difference value sets to obtain the trained long short-term memory neural network. In this embodiment, the long short-term memory neural network includes a forget gate, an input gate, and an output gate to ensure that information transmission is "timely." The overall architecture is designed as a three-layer recurrent neural network structure. The forget gate controls the weight of the time series neurons before the current moment; the input gate controls the weight of the input at the previous moment as the calculation item at the current moment; and the output gate controls the weight of the current moment and the predicted output.

[0064] In this embodiment, the stochastic gradient method is used for training. The number of data (samples) passed to the program for training is 3, and the total number of iterations is set to 100. That is, the training of the long short-term memory neural network is completed after 100 rounds of iterations. The training process is as follows: Figure 3 As shown, Figure 3In the example, PredictionOfDifference is the predicted difference, train is training, predict is prediction, Time is time, Difference is difference, Figure 3 It can be seen that the training process converges quickly and the training effect is good.

[0065] The uncertainty error data is predicted by the long short-term memory neural network, and the predicted difference data is compared with the original difference data. The root mean square error is used as an indicator to evaluate the prediction results of the long short-term memory neural network. The testing process of the long short-term memory neural network is as follows: Figure 4 As shown, from Figure 4 It can be seen that the convergence is fast and the prediction error is small. The overall training and testing process is as follows Figure 5 As shown, it can be concluded that the long short-term memory neural network has the ability to approximate modeling uncertainty errors, and the error between the output of the long short-term memory neural network and the real data is very small. Figure 4 and Figure 5 In , Epoch is the number of iterations, TrainLoss is the root mean square error of the training process, TestLoss is the root mean square error of the testing process, test is the test, and loss is the root mean square error.

[0066] Step S5: constructing an initial dynamic model of the required flexible spacecraft, performing programming calculations on the initial dynamic model, and obtaining a required parameter matrix.

[0067] Step S6: obtaining a prediction error value based on the demand parameter matrix and the trained long short-term memory neural network.

[0068] Step S7: adjusting the initial dynamic model based on the prediction error value to obtain a dynamic model of the required flexible spacecraft.

[0069] Figure 6 This is a structural diagram of the hybrid dynamics modeling system based on the long short-term memory neural network of the present invention. Figure 6 As shown, the present invention provides a hybrid power dynamics modeling system based on long short-term memory neural network, including: a model building module 1, a programming calculation module 2, a difference module 3, a neural network module 4, a demand model module 5, a prediction module 6 and a model adjustment module 7.

[0070] The model building module 1 is used to build a hybrid dynamics model of the flexible spacecraft and adjust the physical parameters of the dynamics model within a set threshold range to obtain a plurality of adjusted dynamics models.

[0071] The programming calculation module 2 is used to perform programming calculations on the hybrid dynamics model to obtain a reference parameter matrix, and to perform programming calculations on a plurality of the adjustment dynamics models to obtain a plurality of difference parameter matrices.

[0072] The subtraction module 3 is used to perform subtraction between each of the difference parameter matrices and the reference parameter matrix to obtain a difference value set.

[0073] The neural network module 4 is used to construct a long short-term memory neural network based on the difference parameter matrix, and train the long short-term memory neural network based on each of the difference parameter matrices and each of the difference value sets to obtain the trained long short-term memory neural network.

[0074] The demand model module 5 is used to construct an initial dynamic model of the demand flexible spacecraft and perform programming calculations on the initial dynamic model to obtain a demand parameter matrix.

[0075] The prediction module 6 is used to obtain a prediction error value based on the demand parameter matrix and the trained long short-term memory neural network.

[0076] The model adjustment module 7 is used to adjust the initial dynamic model based on the prediction error value to obtain the dynamic model of the required flexible spacecraft.

[0077] Optionally, the model building module 1 is specifically:

[0078] An initial finite element model of the flexible spacecraft is constructed, and the initial finite element model is designed according to a set physical parameter set to obtain a design dynamic model.

[0079] The set physical parameter set is adjusted within the set threshold range to obtain a plurality of adjusted physical parameter sets.

[0080] The initial finite element model is designed according to each of the adjustment physical parameter sets to obtain a plurality of initial adjustment dynamic models.

[0081] Fixed surfaces and remote points are set on the design dynamics model to obtain the hybrid dynamics model.

[0082] Fixed surfaces and remote points are set for each of the initial adjustment dynamic models to obtain each of the adjustment dynamic models.

[0083] Optionally, the reference parameter matrix includes a mass matrix, a damping matrix and a stiffness matrix.

[0084] Optionally, the physical parameters include stiffness, density and Young's modulus.

[0085] Optionally, the long short-term memory neural network includes a forget gate, an input gate and an output gate.

[0086] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0087] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims

1. A hybrid dynamics modeling method based on long short-term memory neural network, characterized in that: include: Constructing a hybrid dynamics model of the flexible spacecraft; adjusting physical parameters of the dynamics model within a set threshold range to obtain a plurality of adjusted dynamics models; The hybrid dynamics model of the flexible spacecraft is constructed; the physical parameters of the dynamics model are adjusted within a set threshold range to obtain a plurality of adjusted dynamics models, including: constructing an initial finite element model of the flexible spacecraft, and designing the initial finite element model according to a set set of physical parameters to obtain a design dynamic model; Adjusting the set physical parameter set within the set threshold range to obtain a plurality of adjusted physical parameter sets; Designing the initial finite element model according to each of the adjustment physical parameter sets to obtain a plurality of initial adjustment dynamic models; Setting fixed surfaces and remote points on the design dynamics model to obtain the hybrid dynamics model; Setting fixed surfaces and remote points for each of the initial adjustment dynamic models to obtain each of the adjustment dynamic models; Performing programming calculation on the hybrid dynamics model to obtain a reference parameter matrix; performing programming calculation on a plurality of the adjustment dynamics models to obtain a plurality of difference parameter matrices; Subtracting each of the difference parameter matrices from the reference parameter matrix to obtain a difference value set; Constructing a long short-term memory neural network based on the difference parameter matrix; training the long short-term memory neural network based on each of the difference parameter matrices and each of the difference value sets to obtain the trained long short-term memory neural network; constructing an initial dynamic model of the required flexible spacecraft, performing programming calculations on the initial dynamic model, and obtaining a required parameter matrix; Obtaining a prediction error value based on the demand parameter matrix and the trained long short-term memory neural network; The initial dynamic model is adjusted based on the prediction error value to obtain a dynamic model of the required flexible spacecraft.

2. The hybrid dynamics modeling method according to claim 1, characterized in that: The reference parameter matrix includes a mass matrix, a damping matrix and a stiffness matrix.

3. The hybrid dynamics modeling method according to claim 1, characterized in that: The physical parameters include stiffness, density and Young's modulus.

4. The hybrid dynamics modeling method according to claim 1, characterized in that: The long short-term memory neural network includes a forget gate, an input gate and an output gate.

5. A hybrid dynamics modeling system based on long short-term memory neural network, characterized in that: include: A model building module is used to build a hybrid dynamics model of the flexible spacecraft and adjust the physical parameters of the dynamics model within a set threshold range to obtain a plurality of adjusted dynamics models; The model building module is specifically: constructing an initial finite element model of the flexible spacecraft, and designing the initial finite element model according to a set set of physical parameters to obtain a design dynamic model; Adjusting the set physical parameter set within the set threshold range to obtain a plurality of adjusted physical parameter sets; Designing the initial finite element model according to each of the adjustment physical parameter sets to obtain a plurality of initial adjustment dynamic models; Setting fixed surfaces and remote points on the design dynamics model to obtain the hybrid dynamics model; Setting fixed surfaces and remote points for each of the initial adjustment dynamic models to obtain each of the adjustment dynamic models; a programming calculation module, configured to perform programming calculations on the hybrid dynamics model to obtain a reference parameter matrix, and to perform programming calculations on a plurality of the adjustment dynamics models to obtain a plurality of difference parameter matrices; a difference module, configured to respectively perform a difference between each of the difference parameter matrices and the reference parameter matrix to obtain a difference value set; A neural network module is used to construct a long short-term memory neural network based on the difference parameter matrix, and train the long short-term memory neural network based on each of the difference parameter matrices and each of the difference value sets to obtain the trained long short-term memory neural network; A demand model module is used to construct an initial dynamic model of the demand flexible spacecraft and perform programming calculations on the initial dynamic model to obtain a demand parameter matrix; A prediction module, configured to obtain a prediction error value based on the demand parameter matrix and the trained long short-term memory neural network; The model adjustment module is used to adjust the initial dynamic model based on the prediction error value to obtain the dynamic model of the required flexible spacecraft.

6. The hybrid dynamics modeling system according to claim 5, characterized in that: The reference parameter matrix includes a mass matrix, a damping matrix and a stiffness matrix.

7. The hybrid dynamics modeling system according to claim 5, characterized in that: The physical parameters include stiffness, density and Young's modulus.

8. The hybrid dynamics modeling system according to claim 5, characterized in that: The long short-term memory neural network includes a forget gate, an input gate and an output gate.

Images