A multi-configuration launch vehicle simulation method, system, device and medium based on an LSTM neural network

Abstract

The present application belongs to the field of aerospace technology, and particularly relates to a multi-form launch vehicle simulation method, system, device and medium based on an LSTM neural network; the simulation method first sets initialization variables and initialization environment parameters of the launch vehicle, then optimizes the learning rate and gradient clipping rate of the built LSTM neural network by using a loss function optimization algorithm, then trains the optimized LSTM neural network by using an Adam optimizer, then inputs the dynamic variables and environment variables into the trained LSTM neural network to simulate the launch vehicle attitude adjustment force and thrust torque, and finally obtains the launch vehicle flight simulation result; the simulation system comprises an initial parameter acquisition module, an LSTM neural network training module, an instantaneous parameter acquisition module and a rocket flight simulation module; the present application can greatly improve the accuracy and reliability of rocket flight simulation, and enable the rocket to maintain a stable flight state when facing complex and variable external interference environments.

Description

Technical Field

[0001] This invention belongs to the field of aerospace technology, and specifically relates to a multi-morphological launch vehicle simulation method, system, equipment and medium based on LSTM neural network. Background Technology

[0002] Rocket launches involve complex multiphysics systems, encompassing dynamics, aerodynamics, propulsion, fuel consumption, and other factors. The rocket propulsion system needs to adjust thrust according to the demands of each stage of flight to overcome the effects of gravity and air resistance, achieving the expected flight trajectory. In the field of rocket flight simulation, relatively traditional simulation methods such as PID control algorithms still exist. These methods lack real-time performance and predictive capabilities. Due to issues with preset parameters, their predicted trajectories are not ideal and cannot accurately simulate real rocket launches. Specifically, in existing rocket launch systems, control and trajectory optimization mainly rely on rule-guided control systems and simple feedback mechanisms, but these methods have the following major drawbacks:

[0003] 1. Lack of adaptability:

[0004] Traditional rocket launch control systems often adjust thrust and combustion rate based on predetermined parameters and fixed control logic. These preset rules cannot adapt to rapid changes in the external environment (such as drastic changes in air pressure, wind speed, etc.), resulting in insufficient system flexibility and responsiveness.

[0005] 2. Limitations of real-time optimization:

[0006] Existing control methods are difficult to perform real-time global optimization based on constantly changing parameters during flight. For thrust distribution, attitude control, and other aspects, existing control methods (such as PID control) cannot quickly and effectively cope with complex disturbances.

[0007] 3. Poor attitude control:

[0008] Under complex aerodynamic conditions, rocket attitude control needs to cope with a variety of factors such as lateral wind speed and turbulence. Traditional control systems usually rely on simple feedback mechanisms, such as offsetting lateral wind speed with lateral thrust. This approach cannot guarantee the stable attitude of the rocket under multiple disturbances.

[0009] In the prior art, patent publication number "CN118518102A" entitled "An Online Flight Trajectory Planning Method Based on Parameter Identification" discloses an online flight trajectory planning method based on parameter identification. This method establishes a deviation identification model for each trajectory-sensitive parameter, adopts an online deviation identification method for trajectory-sensitive parameters, and identifies the online deviation of each parameter based on sensor measurement data during flight. Based on the deviation of the trajectory-sensitive parameters, the corresponding flight trajectory is selected from the trajectory library to complete the online flight trajectory planning. However, this method is more suitable for aircraft with non-adjustable thrust or no thrust, and does not have a significant advantage for aircraft with adjustable thrust. Furthermore, if the dynamic characteristics of the aircraft deviate significantly from the model, the planning effect may be affected. Also, for scenarios with rapidly changing flight environments (such as strong winds and turbulence), the trajectory library of this method may be insufficient to cope with dynamic and complex environmental conditions, affecting the robustness of the system.

[0010] In the prior art, patent publication number "CN 109343341 A" entitled "An Intelligent Control Method for Vertical Recovery of Launch Vehicles Based on Deep Reinforcement Learning" discloses an intelligent control method for vertical recovery of launch vehicles based on deep reinforcement learning. By establishing a simulation model of vertical recovery of launch vehicles, establishing a Markov decision process, and using a neural network model, it realizes an autonomous intelligent control method for launch vehicles, which has the effect of vertical recovery attitude control and trajectory planning. However, it cannot intelligently control thrust, and the neural network model does not consider the learning rate reward problem.

[0011] In the prior art, patent publication number "CN 116697829 A" entitled "A Rocket Landing Guidance Method and System Based on Deep Reinforcement Learning" discloses a rocket landing guidance method and system based on deep reinforcement learning. By building a simulation environment of a six-degree-of-freedom dynamic model of a rocket, a Markov decision process is established, a neural network is built according to the deep reinforcement learning algorithm, and the trained neural network model is used to guide the rocket landing flight, realizing the function of landing flight of a reusable launch vehicle. It has the effects of high algorithm efficiency, low fuel consumption, and good landing accuracy, but it has the problem of insufficient environmental impact factors. Summary of the Invention

[0012] To overcome the shortcomings of the prior art, the present invention aims to propose a multi-morphological launch vehicle simulation method, system, device, and medium based on LSTM neural network. This method builds a neural network through deep learning algorithm, trains the built neural network, and then calls the trained neural network for simulation verification. It realizes the trajectory optimization of the launch vehicle flight process and the autonomous control capability after being disturbed. It can perform high-precision simulation of rocket flight trajectory and thrust control under the presence of multiple environmental disturbance factors.

[0013] Firstly, a multi-morphological launch vehicle simulation method based on LSTM neural networks includes the following steps:

[0014] S1: Set the initialization variables for the launch vehicle, including the initial mass of the launch vehicle. Thrust per engine Initial fuel combustion rate ,altitude Launch angle Configuration Load Total mass of the rocket Current speed Current time Initial fuel mass Total mass of the rocket ;

[0015] S2: Set the initialization environment parameters for the launch vehicle, including turbulence parameters. Initial air pressure Wind speed Initial atmospheric temperature air resistance turbulent disturbance Initial atmospheric density ;

[0016] S3: Construct an LSTM (Long Short-Term Memory) neural network for rocket thrust compensation, wherein the LSTM neural network is configured with an optimizer, learning rate, gradient clipping rate, LSTM output layer dimension, and activation function.

[0017] S4: The learning rate and gradient clipping rate of the LSTM neural network described in step S3 are optimized using a loss function optimization algorithm to obtain an optimized LSTM neural network. The Adam (adaptive moment estimation) optimizer is then used to train the optimized LSTM neural network to obtain a trained LSTM neural network.

[0018] S5: Based on the initialization variables described in step S1, solve for the dynamic variables of the launch vehicle using the dynamic equations. The dynamic variables of the launch vehicle include the total thrust F and acceleration. Quality changing over time The derivative of height h ,gravity Attitude adjustment force and launch vehicle thrust vector control torque;

[0019] S6: Based on the environmental parameters initialized in step S2, solve for the environmental variables during rocket flight using the environmental parameter expressions. These environmental variables include atmospheric density. Atmospheric pressure ,temperature Water vapor pressure Wind speed ;

[0020] S7: Input the dynamic variables and environmental variables calculated in steps S5 and S6 respectively into the LSTM neural network trained in step S4 to simulate the launch vehicle's attitude adjustment force and thrust torque, and obtain the launch vehicle flight simulation results. The LSTM neural network construction process in step S3 is as follows:

[0021] S3.1: Obtain time series data, including the current time, from the initialization variables described in step S1 and the initialization environment parameters described in step S2. Current speed Initial atmospheric density Thrust per engine Initial fuel mass ;

[0022] S3.2: Normalize the time series data described in step S3.1. As shown in formula (1):

[0023] (1)

[0024] In formula (1), x represents unnormalized data. , It is the mean. It is the standard deviation;

[0025] S3.3: Construct the LSTM neural network described in step S3 using the normalized data described in step S3.2, including the input gate, forget gate, and output gate.

[0026] Furthermore, the parameters in the loss function optimization algorithm described in step S4 are as follows:

[0027] Mean square error As shown in formula (2):

[0028] (2)

[0029] In formula (2), N is the total number of data points. It is the actual value at time t. It is the predicted value at time t;

[0030] Angle change penalty As shown in formula (3):

[0031] (3)

[0032] In formula (3), It represents the change in the predicted value between time t and time t+1; It represents the change in the true value between the t-th and t+1-th time points;

[0033] Angular velocity constraints As shown in formula (4):

[0034] (4)

[0035] If the rate of angle change exceeds a set maximum value, an angle rate change penalty is applied. ,in

[0036] Total loss function As shown in formula (5):

[0037] (5)

[0038] In formula (5), and These are the weighting coefficients for change_penalty and rate_penalty, respectively, which control the degree of influence of the penalty term.

[0039] Furthermore, the momentum update, second-order moment estimation, and parameter update in the Adam optimizer described in step S4 are shown in equations (6), (7), and (8), respectively:

[0040] (6)

[0041] In formula (6), It is a momentum estimate, representing the cumulative information of the direction and magnitude of the gradient. It is the first moment estimate of the gradient (momentum). It is the decay rate of the first moment. It is the gradient at the current time;

[0042] (7)

[0043] In formula (7), It is a second-order moment estimate, representing the weighted sum of squares of the gradient magnitude, used to adjust the update step size. It is a second-moment estimate of the gradient. It is the decay rate of the second moment;

[0044] (8)

[0045] In formula (8), These are the model parameters at the current time step. These are the model parameters for the next time step. It is the global learning rate, which controls the step size of parameter updates. It is a smoothing term used to avoid division by zero errors.

[0046] Furthermore, the expression for the dynamic equation described in step S5 is shown in formula (9):

[0047] (9)

[0048] In formula (9), The total mass of the rocket. yes The total mass of the rocket and the mass of its fuel at any given moment, and the thrust. air resistance ,gravity Total thrust formed by random disturbances Determines acceleration ; It is the instantaneous combustion rate of the fuel; It is the effective exhaust speed. It is the fuel consumption efficiency coefficient; F is the rocket's acceleration, and F is the total thrust. It's speed. It's the attitude angle. It refers to altitude.

[0049] Furthermore, the expressions for the attitude adjustment force and the launch vehicle thrust vector control torque mentioned in step S5 are shown in formula (10):

[0050] (10)

[0051] In formula (10), For attitude adjustment force, It is the proportional gain coefficient, which controls the thrust compensation amount for lateral wind speed. For wind speed, For the thrust vector control torque of the launch vehicle, Adjust the attitude angle.

[0052] Furthermore, the environmental parameter expression in step S6 is shown in formula (11):

[0053] (11)

[0054] In formula (11), , These represent atmospheric density and atmospheric pressure at the Earth's surface, respectively. It is the von Kármán constant. It is the friction speed. It is height The average wind speed at that location; This represents the vertical temperature lapse rate. Sea level temperature; tropopause (approximately 11...) Afterwards, the temperature becomes constant; It is height The water vapor pressure at that location; It is the water vapor pressure at sea level; It is the molar mass of water vapor; It is the gas constant of water vapor; It is atmospheric temperature. The initial atmospheric temperature. It is the product of the vertical temperature lapse rate and the variable. It is the molar mass of water vapor. It is the gas constant of water vapor.

[0055] Secondly, a multi-morphological launch vehicle simulation system based on LSTM neural networks is provided. The system utilizes the aforementioned multi-morphological launch vehicle simulation method based on LSTM neural networks. The multi-morphological launch vehicle simulation system based on LSTM neural networks includes an initial parameter acquisition module, an LSTM neural network training module, an instantaneous parameter acquisition module, and a rocket flight simulation module.

[0056] Initial parameter acquisition module: sets the initialization variables of the launch vehicle, including the initial mass of the launch vehicle. Thrust per engine Initial fuel combustion rate ,altitude Launch angle Configuration Load Total mass of the rocket ,speed ,time Initial fuel mass Set the initialization environment parameters for the launch vehicle, including turbulence parameters. Initial air pressure Wind speed initial temperature air resistance turbulent disturbance Initial atmospheric density Total mass of the rocket ;

[0057] LSTM Neural Network Training Module: An LSTM neural network for rocket thrust compensation is constructed. The LSTM neural network includes an optimizer, learning rate, gradient clipping rate, LSTM output layer dimension, and activation function. The learning rate and gradient clipping rate of the LSTM neural network are optimized using a loss function optimization algorithm to obtain an optimized LSTM neural network. Then, the Adam optimizer is used to train the optimized LSTM neural network to obtain the trained LSTM neural network.

[0058] Instantaneous parameter acquisition module: Based on the initialization variables, solves for the dynamic variables of the launch vehicle through dynamic equations. The dynamic variables of the launch vehicle include total thrust F, acceleration, etc. Quality changing over time The derivative of height h ,gravity Attitude adjustment force and launch vehicle thrust vector control torque; based on the initialized environmental parameters, solve for the environmental variables during rocket flight using environmental parameter expressions, the environmental variables including atmospheric density. Atmospheric pressure ,temperature Water vapor pressure Wind speed ;

[0059] Rocket flight simulation module: The dynamic variables and environmental variables are input into the trained LSTM neural network to simulate the launch vehicle's attitude adjustment force and thrust torque, and the launch vehicle flight simulation results are obtained.

[0060] Thirdly, an electronic device including a memory and a processor:

[0061] Memory: Used to store the computer program that implements the LSTM neural network-based multi-morphological launch vehicle simulation method described above;

[0062] Processor: Used to implement the LSTM neural network-based multimorphic launch vehicle simulation method when executing the computer program.

[0063] Fourthly, a computer-readable storage medium stores a computer program that, when executed by a processor, implements the aforementioned multi-morphological launch vehicle simulation method based on an LSTM neural network.

[0064] Compared with the prior art, the beneficial effects of the technical solution of the present invention are as follows:

[0065] 1. In this invention, an LSTM neural network is built through step 3. Environmental parameters (wind speed, atmospheric pressure, atmospheric density, temperature, water vapor pressure, etc.) are input into the LSTM neural network to adjust the rocket's azimuth angle, thrust, and speed in real time. The network is then used for secondary learning through negative feedback and stored in long-term memory.

[0066] 2. This invention optimizes the LSTM neural network through the loss function optimization algorithm in step 4, and trains the LSTM neural network using the Adam optimizer to improve the flight stability of the launch vehicle. Specifically, through the gradient pruning technique and adaptive learning rate adjustment of the LSTM neural network in step 4, the optimizer can still effectively optimize and adjust the model parameters even in extreme cases. It can continuously improve and optimize the thrust allocation strategy during flight, effectively avoiding instability problems in neural network training. Especially when the thrust changes rapidly, the adaptive adjustment of the LSTM neural network makes the training process more stable, reduces the risk of model non-convergence, and enables the rocket to learn and adapt to different flight environments and states (including highly dynamic and complex flight scenarios), avoiding gradient explosion or vanishing phenomena, and further improving the accuracy of thrust compensation.

[0067] 3. This invention acquires environmental data in real time through steps 1 to 4 and trains the rocket using an LSTM neural network. Based on the real-time data collected and calculated in steps 5 to 7, it performs rocket flight simulation. This allows for the simulation of environmental parameters such as air pressure, wind speed, temperature, humidity, and turbulence, and makes corresponding simulation adjustments to the rocket's attitude and thrust torque. Specifically, step 7 uses the linkage between environmental data acquisition and the thrust distribution model, employing stochastic simulation methods (such as stochastic wind speed and turbulence intensity) to control the rocket's attitude and adjust the thrust distribution in real time. This thrust control method based on multidimensional data improves the rocket's flight stability, enabling it to maintain a stable flight state even when facing complex and variable external interference environments.

[0068] In summary, this invention utilizes a trained LSTM neural network and integrates multiple physical field effects such as aerodynamics, gravity, combustion chemistry, and environmental disturbances during rocket flight. It uses multi-equation coupling to simulate the dynamic characteristics of the rocket in various flight stages. By training the LSTM neural network to obtain the optimal parameters under different environments during rocket flight, the accuracy and reliability of the simulation are greatly improved. It can accurately demonstrate the trajectory optimization and prediction during rocket flight and adjust the thrust to ensure that the rocket still flies along the predetermined trajectory when subjected to disturbances. Attached Figure Description

[0069] Figure 1 This is a flowchart of a multi-morphological launch vehicle simulation method based on LSTM neural network provided by the present invention.

[0070] Figure 2 This is a flowchart of the LSTM neural network construction process provided by the present invention.

[0071] Figure 3 This is a schematic diagram of the rocket configuration provided in an embodiment of the present invention.

[0072] Figure 4 This is a comparison chart of the validation loss rate and training loss rate of the LSTM neural network provided in this embodiment of the invention with other models.

[0073] Figure 5 This is the height change curve simulated using an LSTM neural network, provided in an embodiment of the present invention.

[0074] Figure 6 This is a height variation curve simulated without using an LSTM neural network, provided in an embodiment of the present invention.

[0075] Figure 7 This is a comparison chart of fuel consumption before and after using the LSTM neural network provided in this embodiment of the invention.

[0076] Figure 8 This is a graph showing the speed comparison of the LSTM neural network before and after its use, as provided in this embodiment of the invention.

[0077] Figure 9 This is a diagram showing the change in attitude control angle without using an LSTM neural network, provided in an embodiment of the present invention.

[0078] Figure 10 This is a diagram showing the change in attitude control angle after using an LSTM neural network, as provided in an embodiment of the present invention. Detailed Implementation

[0079] The following is combined Figures 1 to 10 The present invention will be further described in detail with reference to examples and embodiments. The launch vehicle described in the present invention refers only to second-stage and third-stage rockets. Figure 1 This is a flowchart of a multi-morphological launch vehicle simulation method based on LSTM neural networks. The rocket simulation method includes the following steps:

[0080] S1: Set the initialization variables for the launch vehicle, including the initial mass of the launch vehicle. Thrust per engine Initial fuel combustion rate ,altitude Launch angle Configuration Load Total mass of the rocket Current speed Current time Initial fuel mass Total mass of the rocket ;

[0081] This embodiment utilizes a Python compiler to implement neural network learning and rocket flight simulation functions, wherein the initial mass of the launch vehicle is... The load is 10,000 kg. 500KG, initial launch angle ,altitude ,speed ,time Initial fuel mass Initial fuel combustion rate Thrust per engine configuration ; Figure 3 This diagram illustrates the rocket configuration of this embodiment. That is, four rocket engines revolve around one rocket engine to provide propulsion power;

[0082] S2: Set the initialization environment parameters for the launch vehicle, including turbulence parameters. Initial air pressure Wind speed Initial temperature air resistance turbulent disturbance Initial atmospheric density ;

[0083] In this embodiment, the turbulence parameters Initial air pressure Wind speed initial temperature air resistance turbulent disturbance Initial atmospheric density ;

[0084] S3: Construct an LSTM neural network for rocket thrust compensation, wherein the LSTM neural network is configured with an optimizer, learning rate, gradient clipping rate, LSTM output layer dimension, and activation function;

[0085] The LSTM neural network construction process in step S3 is as follows:

[0086] S3.1: Obtain time series data, including the current time, from the initialization variables described in step S1 and the initialization environment parameters described in step S2. Current speed Initial atmospheric density Thrust per engine Initial fuel mass ;

[0087] S3.2: Normalize the time series data described in step S3.1. As shown in formula (1):

[0088] (1)

[0089] In formula (1), x represents unnormalized data. , It is the mean. It is the standard deviation;

[0090] S3.3: Construct the LSTM neural network described in step S3 using the normalized data described in step S3.2, including the input gate, forget gate, and output gate;

[0091] The input gate, forget gate, and output gate process data in sequence. The input gate controls whether information enters the memory unit, the forget gate determines whether information in the memory unit is forgotten, and the output gate controls whether information in the memory unit is output. The memory unit is updated by combining the input gate and the forget gate.

[0092] Input gate vector As shown in formula (12):

[0093] (12)

[0094] In formula (12), It is the input gate vector, which controls the degree of influence of the time series data described in step S3.1 on the current memory cell, and the element value is between [0, 1]. It is the Sigmoid activation function, which maps time series data to the [0,1] interval; It is the hidden state vector of the previous time step, which contains the output information of the previous time step; It is the input vector at the current moment, containing the observation data at the current moment; It is the weight matrix of the input gate; It is the hidden state vector from the previous time step. With the input vector at the current time step The vector formed by concatenation (splicing) is used to multiply with the weight matrices of each gate; It is the bias vector of the input gate;

[0095] Forget gate vector As shown in formula (13):

[0096] (13)

[0097] In formula (13), It is the forgetting gate vector, which determines the degree of forgetting of information in the memory unit, and the element value is between [0, 1]. It is the weight matrix of the forget gate; It is the bias vector of the forget gate;

[0098] Output gate vector As shown in formula (14):

[0099] (14)

[0100] In formula (14), It is the output gate vector, which controls how much information in the memory cell is output to the hidden state. The element value is between [0, 1]. It is the weight matrix of the output gate; It is the bias vector of the output gate;

[0101] Current memory cell state vector As shown in formula (15):

[0102] (15)

[0103] In formula (15), It is the state vector of the memory unit at the current moment, which accumulates important information up to the current moment; It is the state vector of the memory cell at the previous time step; It is the weight matrix used to generate candidate memories; It is the bias vector used to generate candidate memories;

[0104] The hidden state of LSTM involves the hidden state vector at the current time step. As shown in formula (16):

[0105] (16)

[0106] In formula (16), It is the hidden state vector at the current time step, which serves as the output at the current time step and is also used for the calculation of the next time step. It is the output gate vector; It is the state vector of the memory cell at the current moment.

[0107] Of the parameters mentioned above, , , , It is the weight matrix of each gate and candidate memory, which contains the weights of the hidden state. and input The linear transformation parameters; , , , It is the bias vector of each gate and candidate memory, adjusted by the linear transformation.

[0108] Figure 2 The LSTM neural network construction process in this invention is illustrated in Table 1. The specific parameters of the LSTM neural network are shown in Table 1.

[0109] Table 1 Specific parameters of LSTM neural network

[0110] Parameter name Parameter value illustrate lstm_units 100 Output dimension of LSTM layer lstm_activation relu Activation function of LSTM layer lstm_input_shape (1, 6) Input data shape of LSTM layer lstm_kernel_regularizer l2(0.001) Kernel regularizer of LSTM layer batch_normalization none The BatchNormalization layer has no explicitly set parameters. dense1_units 50 Output dimension of the first fully connected layer dense1_activation relu Activation function of the first fully connected layer dense1_kernel_regularizer l2(0.001) The kernel regularizer of the first fully connected layer dense2_units 1 The output dimension of the second fully connected layer optimizer tf.keras.optimizers.Adam(learning_rate=0.001, clipnorm=1.0) The optimizer is Adam, the learning rate is 0.001, and the gradient clipping rate is 1.0. loss_function mse The loss function is the mean squared error.

[0111] S4: Define the learning rate adjustment strategy for the LSTM neural network described in step S3 to prevent the LSTM neural network from becoming unstable. Specifically, the learning rate and gradient clipping rate of the LSTM neural network described in step S3 are optimized using a loss function optimization algorithm to obtain an optimized LSTM neural network. Then, the Adam (adaptive moment estimation) optimizer is used to train the optimized LSTM neural network to obtain a trained LSTM neural network.

[0112] By combining LSTM neural networks with real-time data adjustment strategies, the thrust distribution of multi-engine rockets is optimized in different flight phases, making the rocket's flight more stable and precise. The LSTM neural network can perform thrust compensation adjustments based on real-time acquired environmental data (such as pressure, wind speed, temperature, humidity, etc.), ensuring that even if some rocket engines fail, the system can still reasonably allocate the thrust of the remaining engines, guaranteeing flight safety and stability.

[0113] The parameters in the loss function optimization algorithm in step S4 are as follows:

[0114] The mean squared error (MSE) is shown in formula (2):

[0115] (2)

[0116] In formula (2), N is the total number of data points. It is the actual value at time t. It is the predicted value at time t;

[0117] Angle change penalty As shown in formula (3):

[0118] (3)

[0119] In formula (3), It represents the change in the predicted value between time t and time t+1; It represents the change in the true value between the t-th and t+1-th time points;

[0120] Angular velocity constraints As shown in formula (4):

[0121] (4)

[0122] If the rate of angle change exceeds a set maximum value, an angle rate change penalty is applied. ,in In this embodiment, the maximum angle change is 5°.

[0123] Total loss function As shown in formula (5):

[0124] (5)

[0125] In formula (5), and These are the weighting coefficients for change_penalty and rate_penalty, respectively, which control the degree of influence of the penalty term.

[0126] In this embodiment, N=1000, t=1000s, λ1=0.4, λ2=0.6;

[0127] `change_penalty` is a penalty for angle changes, encouraging smooth attitude changes, while `rate_penalty` is a penalty for angle rate changes, ensuring that the rate of angle change does not exceed a certain threshold.

[0128] Adam itself has a built-in mechanism for dynamically adjusting the learning rate, so there is no need to manually set the learning rate decay. However, in some cases, combining learning rate decay can further optimize the training process and improve training results.

[0129] The Adam optimization algorithm is a stochastic optimization method suitable for large-scale data and non-stationary objectives. Its core idea is as follows:

[0130] Furthermore, the momentum update, second-order moment estimation, and parameter update in the Adam optimizer are shown in equations (6), (7), and (8), respectively:

[0131] (6)

[0132] In formula (6), It is a momentum estimate, representing the cumulative information of the direction and magnitude of the gradient. It is the first moment estimate of the gradient (momentum). It is the decay rate of the first moment. It is the gradient at the current time, in this embodiment It is 0.9;

[0133] (7)

[0134] In formula (7), It is a second-order moment estimate, representing the weighted sum of squares of the gradient magnitude, used to adjust the update step size. It is a second-moment estimate of the gradient. It is the decay rate of the second moment, in this embodiment It is 0.999;

[0135] (8)

[0136] In formula (8), These are the model parameters at the current time step. These are the model parameters for the next time step. It is the global learning rate, which controls the step size of parameter updates. It is a smoothing term used to avoid division by zero errors, in this embodiment. for ;

[0137] In this embodiment, the optimizer is the Adam optimizer, the learning rate is 0.001, the gradient clipping rate is 1.0, the LSTM output layer dimension is 100, and ReLU is used as the activation function.

[0138] Figure 4 This demonstrates the comparison of validation and training loss rates between LSTM neural networks and other models: fully connected feedforward neural networks (MLP, Multi-Layer Perceptron), simple RNNs (Recurrent Neural Networks), and 1D convolutional neural networks (CNNs). Figure 4As can be seen, the LSTM neural network can capture long-term dependencies in time series data, and its training and validation losses are lower than those of other models. In this embodiment, the number of training repetitions was set to 200. After training with the LSTM neural network, both the training loss and validation loss were effectively reduced. When the number of training epochs was 100, the training and validation losses of the LSTM model were both effectively below 0.1, while the training and validation losses of other models that did not use the LSTM model were between 0.1 and 0.4. This indicates that the LSTM neural network model performed well throughout the training process. The stability of the LSTM neural network was significantly improved after adding the Adam optimizer, and it could effectively adapt to the validation set without obvious overfitting or underfitting, thus avoiding the problem of output errors caused by model instability.

[0139] S5: Based on the initialization variables described in step S1, solve for the dynamic variables of the launch vehicle using the dynamic equations. The dynamic variables of the launch vehicle include the total thrust F and acceleration. Quality changing over time The derivative of height h ,gravity Attitude adjustment force and launch vehicle thrust vector control torque;

[0140] The expression for the dynamic equation is shown in formula (9):

[0141] (9)

[0142] In formula (9), The total mass of the rocket. yes The total mass of the rocket and the mass of its fuel at any given moment, and the thrust. air resistance ,gravity Total thrust formed by random disturbances Determines acceleration ; It is the instantaneous combustion rate of the fuel; It is the effective exhaust speed. It is the fuel consumption efficiency coefficient; F is the rocket's acceleration, and F is the total thrust. It's speed. It's the attitude angle. It refers to altitude;

[0143] Furthermore, the expressions for the attitude adjustment force and the launch vehicle thrust vector control torque are shown in formula (10):

[0144] (10)

[0145] In formula (10), This refers to attitude adjustment force (the thrust of a single engine varies with altitude). It is the proportional gain coefficient, which controls the thrust compensation amount for lateral wind speed. For wind speed, For the thrust vector control torque of the launch vehicle, For attitude adjustment angle (i.e. ).

[0146] In this embodiment, an end time is set. Posture adjustment force The range is [100N, 10000N]. The rocket's altitude increases continuously over time, and the attitude adjustment angle... The range is [85°, 92°], and the thrust vector control torque of the launch vehicle is... [100N] m, 2000N [m], the thrust compensation for controlling the lateral wind speed is 0.7. The attitude adjustment angle is ;

[0147] S6: Based on the environmental parameters initialized in step 2, solve for the environmental variables during rocket flight using the environmental parameter expressions. These environmental variables include atmospheric density. Atmospheric pressure ,temperature Water vapor pressure Wind speed ;

[0148] The environmental parameter expression is shown in formula (11):

[0149] (11)

[0150] In formula (11), , These represent atmospheric density and atmospheric pressure at the Earth's surface, respectively. It is the von Kármán constant, which is 0.4. It is the friction speed. It is height The average wind speed at that location; This represents the vertical temperature lapse rate. Sea level temperature; tropopause (approximately 11...) Afterwards, the temperature becomes constant; It is height The water vapor pressure at that location; It is the water vapor pressure at sea level; It is the molar mass of water vapor; It is the gas constant of water vapor; It is atmospheric temperature. The initial atmospheric temperature. It is the product of the vertical temperature lapse rate and the variable. It is the molar mass of water vapor. It is the gas constant of water vapor.

[0151] In this embodiment, the environmental variables change with altitude. Atmospheric density The range is [0.003 kg / 1.225 kg / Atmospheric pressure The range is [0.2 kPa, 101.325 kPa], temperature. The range is [50.65K, 288.15K], and the water vapor pressure is... The range is [0.0000001kPa, 3kPa], wind speed The range is ;

[0152] Step 7: Input the dynamic variables and environmental variables calculated in Step S5 and Step S6 respectively into the LSTM neural network trained in Step S4 to simulate the attitude adjustment force and thrust torque of the launch vehicle, and obtain the flight simulation results of the launch vehicle to ensure that the attitude and thrust distribution of the rocket are reasonable under complex environmental changes.

[0153] Figure 5 The diagram shows the altitude change curve simulated using an LSTM neural network, which is more consistent with physical characteristics. In the initial stage, the altitude rises slowly, affected by air resistance and gravity. In the middle stage, the altitude increase slows down significantly, simulating the transition region where thrust weakens. In the later stage, the altitude rises rapidly and tends to stabilize, simulating the effect of reduced air resistance and increased thrust. Figure 6 The diagram shows the altitude change curves simulated without using an LSTM neural network. It only simulates the changes in the initial and intermediate stages. The intermediate stage is similar to a linear case and cannot simulate the real flight process. Figure 7 The study demonstrates that adding an LSTM neural network reduces fuel consumption and effectively saves fuel. Figure 8 The speed change curve using the LSTM neural network takes into account dynamic environmental factors and achieves adaptive adjustment, demonstrating better adaptability and anti-interference ability. Due to environmental disturbances, the attitude control force will change continuously. The attitude control force with the addition of the LSTM neural network is smoother and more stable, with the control amplitude not exceeding ±5°. Figure 9 , Figure 10Attitude control angles include pitch, yaw, and roll, which are controlled by... Figure 9 , Figure 10 The comparison reveals that the pose control angle changes without the LSTM neural network are very chaotic, while the pose control with the LSTM neural network is smoother and less prone to drastic fluctuations.

[0154] Secondly, a multi-morphological launch vehicle simulation system based on an LSTM neural network is provided. Applying the multi-morphological launch vehicle simulation method, the simulation system includes an initial parameter acquisition module, an LSTM neural network training module, an instantaneous parameter acquisition module, and a rocket flight simulation module.

[0155] Initial parameter acquisition module: sets the initialization variables of the launch vehicle, including the initial mass of the launch vehicle. Thrust per engine Initial fuel combustion rate ,altitude Launch angle Configuration Load Total mass of the rocket ,speed ,time Initial fuel mass Set the initialization environment parameters for the launch vehicle, including turbulence parameters. Initial air pressure Wind speed initial temperature air resistance turbulent disturbance Initial atmospheric density Total mass of the rocket ;

[0156] LSTM Neural Network Training Module: An LSTM neural network for rocket thrust compensation is constructed. The LSTM neural network includes an optimizer, learning rate, gradient clipping rate, LSTM output layer dimension, and activation function. The learning rate and gradient clipping rate of the LSTM neural network are optimized using a loss function optimization algorithm to obtain an optimized LSTM neural network. Then, the Adam optimizer is used to train the optimized LSTM neural network to obtain the trained LSTM neural network.

[0157] Instantaneous parameter acquisition module: Based on the initialization variables, solves for the dynamic variables of the launch vehicle through dynamic equations. The dynamic variables of the launch vehicle include total thrust F, acceleration, etc. Quality changing over time The derivative of height h ,gravity Attitude adjustment force and launch vehicle thrust vector control torque; based on the initialized environmental parameters, solve for the environmental variables during rocket flight using environmental parameter expressions, the environmental variables including atmospheric density. Atmospheric pressure ,temperature Water vapor pressure Wind speed ;

[0158] Rocket flight simulation module: The dynamic variables and environmental variables are input into the trained LSTM neural network to simulate the launch vehicle's attitude adjustment force and thrust torque, and the launch vehicle flight simulation results are obtained.

[0159] Thirdly, an electronic device including a memory and a processor:

[0160] Memory: Used to store the computer program that implements the LSTM neural network-based multi-morphological launch vehicle simulation method;

[0161] Processor: Used to implement the LSTM neural network-based multi-morphological launch vehicle simulation method when executing the computer program.

[0162] Fourthly, a computer-readable storage medium stores a computer program, which, when executed by a processor, implements the multi-morphological launch vehicle simulation method based on an LSTM neural network; the computer-readable storage medium includes various media capable of storing program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard drive, magnetic disk, or optical disk.

[0163] The working principle of this invention is as follows:

[0164] This method constructs an LSTM neural network by initially setting basic parameters (i.e., rocket initialization variables and initialization environment parameters in steps 1 and 2), optimizes the LSTM neural network parameters through a loss function optimization algorithm, iteratively trains the neural network based on the basic parameters input to the LSTM neural network, and trains the neural network through the Adam optimizer. Finally, the parameters during rocket flight are simulated in the trained LSTM neural network, and the simulation results are output in the form of charts.

Claims

1. A multi-morphological launch vehicle simulation method based on LSTM neural network, characterized in that, Includes the following steps: S1: Set the initialization variables for the launch vehicle, including the initial mass of the launch vehicle. Thrust per engine Initial fuel combustion rate ,altitude Launch angle Configuration Load Total mass of the rocket Current speed Current time Initial fuel mass Total mass of the rocket ; S2: Set the initialization environment parameters for the launch vehicle, including turbulence parameters. Initial air pressure Wind speed Initial atmospheric temperature air resistance turbulent disturbance Initial atmospheric density ; S3: Construct an LSTM neural network for rocket thrust compensation, wherein the LSTM neural network is configured with an optimizer, learning rate, gradient clipping rate, LSTM output layer dimension, and activation function; S4: The learning rate and gradient clipping rate of the LSTM neural network described in step S3 are optimized using a loss function optimization algorithm to obtain an optimized LSTM neural network. The Adam optimizer is then used to train the optimized LSTM neural network to obtain a trained LSTM neural network. S5: Based on the initialization variables described in step S1, solve for the dynamic variables of the launch vehicle using the dynamic equations. The dynamic variables of the launch vehicle include the total thrust F and acceleration. Quality changing over time The derivative of height h ,gravity Attitude adjustment force and launch vehicle thrust vector control torque; S6: Based on the environmental parameters initialized in step S2, solve for the environmental variables during rocket flight using the environmental parameter expressions. These environmental variables include atmospheric density. Atmospheric pressure ,temperature Water vapor pressure Wind speed ; S7: Input the dynamic variables and environmental variables calculated in steps S5 and S6 respectively into the LSTM neural network trained in step S4 to simulate the launch vehicle's attitude adjustment force and thrust torque, and obtain the launch vehicle flight simulation results; wherein, the LSTM neural network construction process in step S3 is as follows: S3.1: Obtain time series data, including the current time, from the initialization variables described in step S1 and the initialization environment parameters described in step S2. Current speed Initial atmospheric density Thrust per engine Initial fuel mass ; S3.2: Normalize the time series data described in step S3.

1. As shown in formula (1): （1） In formula (1), x represents unnormalized data. It is the mean. It is the standard deviation; S3.3: Utilize the normalized data described in step S3.2 The construction of the LSTM neural network described in step S3 includes an input gate, a forget gate, and an output gate.

2. The multi-morphological launch vehicle simulation method based on LSTM neural network as described in claim 1, characterized in that, The parameters in the loss function optimization algorithm described in step S4 are as follows: Mean square error As shown in formula (2): （2） In formula (2), N is the total number of data points. It is the actual value at time t. It is the predicted value at time t; Angle change penalty As shown in formula (3): （3） In formula (3), It represents the change in the predicted value between time t and time t+1; It represents the change in the true value between the t-th and t+1-th time points; Angular velocity constraints As shown in formula (4): （4） If the rate of angle change exceeds a set maximum value, an angle rate change penalty is applied. ,in ; Total loss function As shown in formula (5): （5） In formula (5), and These are the weighting coefficients for change_penalty and rate_penalty, respectively, which control the degree of influence of the penalty term.

3. The multi-morphological launch vehicle simulation method based on LSTM neural network as described in claim 1, characterized in that, The momentum update, second-order moment estimation, and parameter update in the Adam optimizer described in step S4 are shown in equations (6), (7), and (8), respectively: （6） In formula (6), It is a momentum estimate, representing the cumulative information of the direction and magnitude of the gradient. It is a first-moment estimate of the gradient. It is the decay rate of the first moment. It is the gradient at the current time; （7） In formula (7), It is a second-order moment estimate, representing the weighted sum of squares of the gradient magnitude, used to adjust the update step size. It is a second-moment estimate of the gradient. It is the decay rate of the second moment; （8） In formula (8), These are the model parameters at the current time step. These are the model parameters for the next time step. It is the global learning rate, which controls the step size of parameter updates. It is a smoothing term used to avoid division by zero errors.

4. The multi-morphological launch vehicle simulation method based on LSTM neural network as described in claim 1, characterized in that, The expression for the dynamic equation in step S5 is shown in formula (9): （9） In formula (9), The total mass of the rocket. yes The total mass of the rocket and the mass of its fuel at any given moment, and the thrust. air resistance ,gravity Total thrust formed by random disturbances Determines acceleration ; It is the instantaneous combustion rate of the fuel; It is the effective exhaust speed. It is the fuel consumption efficiency coefficient; F is the rocket's acceleration, and F is the total thrust. It's speed. It's the attitude angle. It refers to altitude.

5. The multi-morphological launch vehicle simulation method based on LSTM neural network as described in claim 1, characterized in that, The expressions for the attitude adjustment force and the launch vehicle thrust vector control torque mentioned in step S5 are shown in formula (10): （10） In formula (10), For attitude adjustment force, It is the proportional gain coefficient, which controls the thrust compensation amount for lateral wind speed. For wind speed, For the thrust vector control torque of the launch vehicle, Adjust the attitude angle.

6. The multi-morphological launch vehicle simulation method based on LSTM neural network as described in claim 1, characterized in that, The environmental parameter expression in step S6 is shown in formula (11): （11） In formula (11), , These represent atmospheric density and atmospheric pressure at the Earth's surface, respectively. It is the von Kármán constant. It is the friction speed. It is height The average wind speed at that location; This represents the vertical temperature lapse rate. The temperature is the sea level temperature; the tropopause temperature becomes a constant. It is height The water vapor pressure at that location; It is the water vapor pressure at sea level; It is the molar mass of water vapor; It is the gas constant of water vapor; It is atmospheric temperature. The initial atmospheric temperature. It is the product of the vertical temperature lapse rate and the variable. It is the molar mass of water vapor. It is the gas constant of water vapor.

7. A multi-morphological launch vehicle simulation system based on LSTM neural network, characterized in that, The multi-morphological launch vehicle simulation method based on LSTM neural network as described in any one of claims 1 to 6 is used, wherein the multi-morphological launch vehicle simulation system based on LSTM neural network includes an initial parameter acquisition module, an LSTM neural network training module, an instantaneous parameter acquisition module, and a rocket flight simulation module. Initial parameter acquisition module: sets the initialization variables of the launch vehicle, including the initial mass of the launch vehicle. Thrust per engine Initial fuel combustion rate ,altitude Launch angle Configuration Load Total mass of the rocket ,speed ,time Initial fuel mass Set the initialization environment parameters for the launch vehicle, including turbulence parameters. Initial air pressure Wind speed initial temperature air resistance turbulent disturbance Initial atmospheric density Total mass of the rocket ; LSTM Neural Network Training Module: An LSTM neural network for rocket thrust compensation is constructed. The LSTM neural network includes an optimizer, learning rate, gradient clipping rate, LSTM output layer dimension, and activation function. The learning rate and gradient clipping rate of the LSTM neural network are optimized using a loss function optimization algorithm to obtain an optimized LSTM neural network. Then, the Adam optimizer is used to train the optimized LSTM neural network to obtain the trained LSTM neural network. Instantaneous parameter acquisition module: Based on the initialization variables, solves for the dynamic variables of the launch vehicle through dynamic equations. The dynamic variables of the launch vehicle include total thrust F, acceleration, etc. Quality changing over time The derivative of height h ,gravity Attitude adjustment force and launch vehicle thrust vector control torque; based on the initialized environmental parameters, solve for the environmental variables during rocket flight using environmental parameter expressions, the environmental variables including atmospheric density. Atmospheric pressure ,temperature Water vapor pressure Wind speed ; Rocket flight simulation module: The dynamic variables and environmental variables are input into the trained LSTM neural network to simulate the launch vehicle's attitude adjustment force and thrust torque, and the launch vehicle flight simulation results are obtained.

8. An electronic device comprising a memory and a processor, characterized in that: Memory: Used to store computer programs that implement the LSTM neural network-based multi-morphological launch vehicle simulation method as described in any one of claims 1 to 6; Processor: Used to implement the multi-morphological launch vehicle simulation method based on LSTM neural network as described in any one of claims 1 to 6 when executing the computer program.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the multi-morphological launch vehicle simulation method based on an LSTM neural network as described in any one of claims 1 to 6.

Images