Rocket attitude data processing method and device and storage medium

Abstract

The invention discloses a rocket attitude data processing method and device and a storage medium, and belongs to the field of rocket attitude data processing, and the method comprises the steps: obtaining the real attitude data of a rocket in a motion state, and carrying out the preprocessing of the obtained real attitude data; constructing a generative adversarial network model, and defining a loss function of the generative adversarial network model; training the generative adversarial network model, and evaluating and iteratively updating the trained generative adversarial network model to obtain an optimal generative adversarial network model; using the optimal generative adversarial network model to generate rocket flight attitude data; wherein the real attitude data comprises an attitude angle, a navigation speed, a position and a timestamp. According to the method, the high-quality complementation data of the lost part can be generated according to the known data, the integrity of the data is ensured, and a guarantee is provided for subsequent data processing.

Description

Technical Field

[0001] The present invention relates to the field of rocket attitude data processing, and in particular to a rocket attitude data processing method, device and storage medium. Background Art

[0002] Flight attitude angle, navigation speed, position and timestamp are the core data describing the motion state of the aircraft, which are directly related to the accuracy and reliability of the flight control, navigation and guidance systems. In the field of aerospace, the accuracy and real-time performance of data processing are the key to ensuring the safe and efficient operation of aircraft. However, complex environmental interference (such as turbulence, atmospheric disturbances) and system limitations (such as sensor noise, data loss, etc.) during flight pose major challenges to data processing.

[0003] Traditional methods rely on physical models and filtering algorithms, and perform well under specific conditions, but they are insufficient in correlation modeling of multi-dimensional dynamic data, handling of nonlinear problems, and adaptability to diverse scenarios.

[0004] In particular, existing solutions lack generation capabilities when processing high-dimensional data, and the data augmentation methods are simple, which may not cover the diversity of real posture data and make it difficult to generate high-quality synthetic data.

[0005] In view of this, it is necessary to provide a new technical solution to solve the above problems. Summary of the invention

[0006] In order to solve the above technical problems, the present application provides a rocket attitude data processing method, device and storage medium, which can generate high-quality completion data of the missing part based on known data, ensure the integrity of the data, and provide a guarantee for subsequent data processing.

[0007] A rocket attitude data processing method, comprising: Acquire the real attitude data of the rocket in motion and pre-process the acquired real attitude data; Build a generative adversarial network model and define the loss function of the generative adversarial network model; Train the generative adversarial network model, and evaluate and iteratively update the trained generative adversarial network model to obtain the optimal generative adversarial network model; Generate rocket flight attitude data using the optimal generative adversarial network model; The real posture data includes posture angle, navigation speed, position and timestamp.

[0008] Preferably, the preprocessed real posture data must meet the following conditions: The pre-processed attitude angle changes cyclically within the range of -180° to 180°, with continuity and smoothness, avoiding sudden changes or unreasonable oscillations; The pre-processed navigation speed complies with the flight physics including the maximum acceleration and deceleration range and is consistent with the flight mission phase; The preprocessed position satisfies the physical trajectory constraint; the preprocessed position includes longitude, latitude and altitude, and the altitude is consistent with the attitude angle and navigation speed; The preprocessed timestamps are time series data with the correct time interval and time point order.

[0009] Preferably, defining the loss function of the generative adversarial network model includes defining a loss function of a discriminator and a loss function of a generator in the generative adversarial network model; Among them, the loss function of the discriminator is used to evaluate the discriminator's ability to distinguish between real posture data and generated posture data; the loss function of the generator is used to evaluate the generator's ability to deceive the discriminator.

[0010] Preferably, the discriminator loss function is expressed as: ; In the formula, The average log probability of the discriminator correctly classifying the real pose data; The real posture data for input discriminator; is the output value of the discriminator for the true posture data; is the average log probability of the generated pose data being correctly classified as false by the discriminator; is the discriminant output value of the generated posture data by the discriminator; The posture data is generated by the generator; is the random noise of the input.

[0011] Preferably, the expression of the generator loss function is: ; In the formula, Indicates that the discriminator believes that the generated posture data is The logarithm of the true probability.

[0012] Preferably, in the evaluation of the trained generative adversarial network model, the verification of the attitude angle data includes: Check whether the generated posture angle is within a reasonable range; Calculate the rate of change of the angle sequence, use Fourier transform to analyze the frequency domain characteristics of the generated angles, and check whether there is high-frequency noise; Through autocorrelation analysis, it is evaluated whether the generated angle sequence is consistent with the real angle sequence; Apply the attitude angle data to the flight simulation model to check whether the generated data supports a stable flight attitude.

[0013] Preferably, in the evaluation of the trained generative adversarial network model, the verification of the navigation speed data includes: Check if the generated speed value is within a reasonable range; Analyze the speed direction to ensure there is no abnormal direction; Calculate the derivatives of the velocity series to check whether the acceleration complies with the physical constraints of the aircraft; The correlation coefficient is used to evaluate the relationship between speed and attitude angle to avoid excessive angle changes when the navigation speed is too high.

[0014] Preferably, in the evaluation of the trained generative adversarial network model, the verification of the position data includes: Visualize the generated longitude, latitude, and altitude data on the map to see if the trajectory distribution is reasonable; Check if the trajectory points meet the aircraft trajectory constraints; Calculate the distance change between trajectory points and check whether there is a mutation; Use kinematic formulas to verify that position, velocity, and angle match.

[0015] According to another aspect of the present application, a computing device is also provided, including: a processor and a memory storing a computer program, wherein when the computer program is run by the processor, the rocket attitude data processing method is executed.

[0016] According to another aspect of the present application, a computer-readable storage medium is also provided, on which computer instructions are stored. When the instructions are executed on a computer, the computer executes the rocket attitude data processing method.

[0017] Compared with the prior art, this application has at least the following beneficial effects: 1. The present invention can generate high-quality complementary data of the missing part based on known data to ensure the integrity of the data. By learning the distribution laws between multiple variables, the generated complementary data is more in line with the actual dynamics.

[0018] 2. The present invention can generate diversified flight data through data enhancement technology, thereby expanding the data set, and can provide a richer sample space to improve the generalization ability of the model in unseen scenarios.

[0019] 3. The present invention can learn the joint distribution between multiple variables and generate high-dimensional dynamic data that conforms to physical laws.

[0020] 4. The present invention does not need to rely on a specific physical model, can learn nonlinear relationships, and adapt to complex dynamic environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Hereinafter, some specific embodiments of the present invention will be described in detail in an exemplary and non-limiting manner with reference to the accompanying drawings. The same reference numerals in the accompanying drawings indicate the same or similar components or parts. It should be understood by those skilled in the art that these drawings are not necessarily drawn to scale. In the accompanying drawings: Figure 1 It is a schematic diagram of the overall process of the present invention. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solution and advantages of the present application clearer, the technical solution of the present application will be clearly and completely described below in combination with the specific embodiments of the present application and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present application.

[0023] like Figure 1 As shown, a method for processing rocket attitude data comprises the following steps: S1. Acquire the real attitude data of the rocket in motion state, and preprocess the acquired real attitude data.

[0024] The real attitude data is obtained from flight test records, simulation experiment data, and historical flight test data, including attitude angle, navigation speed, position, and timestamp.

[0025] When preprocessing the acquired real posture data, it includes data cleaning and data standardization.

[0026] Specifically, data cleaning includes: removing missing values, outliers and other noise data, unifying the data format, and providing high-quality data for the training of the Generative Adversarial Network (GAN) model. At the same time, filters such as Kalman filtering are used to filter out sensor noise.

[0027] When performing data standardization, the continuous time series of the same flight phase is divided into segments of a fixed length of 1 second, with one frame of data being 10 milliseconds. Each segment is used as the input of the generative adversarial network model to make the data generated by the generative adversarial network model more consistent. Attention should be paid to data boundary issues.

[0028] The Min-Max normalization formula for data is: ; The processed real posture data should meet the corresponding conditions: The attitude angles include yaw, pitch, and roll, which indicate the spatial orientation of the aircraft. They usually change cyclically within the range of [-180°, 180°] and should be continuous and smooth to avoid sudden changes or unreasonable oscillations.

[0029] The navigation speed includes a three-dimensional velocity vector, which describes the magnitude and direction of the aircraft's speed in three-dimensional space. It should comply with the laws of flight physics, such as the maximum acceleration and deceleration range, and be consistent with the flight mission phase.

[0030] Position includes longitude, latitude, and altitude, which indicates the specific location of the aircraft on the earth's surface. It is affected by the curvature of the earth and usually needs to meet physical trajectory constraints. Altitude information is strongly related to time and should be consistent with attitude angle and speed.

[0031] Timestamps are time series data, and it is necessary to ensure that the generated data has the correct time interval and time point sequence.

[0032] S2. Build a generative adversarial network model and define the loss function of the generative adversarial network model.

[0033] The generative adversarial network model consists of a generator and a discriminator, which are trained using an adversarial training method, where the goal of the generator is to "fool" the discriminator so that it cannot distinguish between real data and generated posture data, while the discriminator tries its best to distinguish them correctly.

[0034] The generator uses real data plus random noise to generate fake data as generated pose data. In order to make the generated data look as much like real data as possible, the generator usually includes a series of fully connected layers, convolutional layers, and deconvolutional layers to gradually transform random noise into data.

[0035] Add a low-dimensional Gaussian random noise vector to the input , the dimension is the length and feature number of the time series. After being processed by the model, pseudo data with the same dimension as the real posture data is output as the generated posture data. The LSTM model is used to generate continuous data in combination with the characteristics of the time series to capture the long-term and short-term dependencies of the time series. The dimension of the posture data generated by the output layer is consistent with the amplitude and number of channels of the real posture.

[0036] The discriminator is a binary classification network used to distinguish between generated pose data and real data. It takes real data and generated pose data as input, and its dimension is consistent with the generator output.

[0037] The input data is classified into "real pose data" or "generated pose data". The goal is to identify the fake data generated by the generator. At the same time, multiple convolutional layers and pooling layers are used, and each layer has an activation function Leaky ReLU to enhance nonlinearity and avoid overfitting. The Sigmoid function is used to limit the output value between 0 and 1 to judge the authenticity of the data. In this implementation, 1 represents real data and 0 represents generated pose data.

[0038] In order to make the generator deceive the discriminator as much as possible and make the generated data look real, a loss function needs to be used to process it.

[0039] The discriminator loss function is used to evaluate the discriminator's ability to distinguish between real pose data and generated pose data. The goal is to maximize the correctness of the judgment of the real pose data, while trying to judge the generated pose data as false. The expression is: ; in, is the true pose data input to the discriminator, is the output value of the discriminator for the real posture data, indicating that the discriminator believes is the true probability, ranging from [0, 1]. The average log probability of the discriminator correctly classifying the real pose data, ensuring that the discriminator can identify the authenticity of the data. It is the generated posture data generated by the generator, which inputs random noise z and outputs generated posture data similar to the real posture data, including simulated time series posture angles and dynamic change data. The output value of the discriminator for the generated posture data indicates the probability that the discriminator believes that these fake posture data are real posture data. The discriminator attempts to detect whether the generated pseudo posture data is consistent with the real posture data, and whether the generated posture angle, speed, and position changes over time are consistent with the real flight trajectory. is the average log probability that the generated pose data is correctly classified as fake by the discriminator, and the discriminator attempts to distinguish the generated pose data from the real pose data.

[0040] Optimize the discriminator loss function Closer to 1. Closer to 0, the discriminator determines the real pose data as real and identifies the generated pose data as fake.

[0041] The generator's loss function is used to evaluate the generator's ability to deceive the discriminator. The goal is to maximize the generation of data that conforms to the laws of real rocket flight attitude, making it impossible for the discriminator to distinguish between real and generated attitude data.

[0042] Generator loss function The expression is: ; in, Indicates that the discriminator believes that the generated posture data is The logarithm of the true probability. The generator hopes to maximize this value so that the discriminator "thinks" The generator attempts to make the generated time series of attitude angle, position, and speed change data mistakenly regarded by the discriminator as real flight data.

[0043] The loss function of the generator makes the generator generate Data, let the discriminator judge , optimize the generator so that its output data is judged as true by the discriminator, that is Close to 1.

[0044] S3. Train the generative adversarial network model, and evaluate and iteratively update the trained generative adversarial network model to obtain the optimal generative adversarial network model.

[0045] Specifically, training the generative adversarial network model includes training the discriminator and training the generator.

[0046] When training the discriminator, a batch of real posture data samples are extracted from the real posture data set, and at the same time, a batch of fake data samples are generated by the generator by inputting random noise into the generator. The real posture data samples are marked as "true" (with a probability of 1) and the generated data samples are marked as "false" (with a probability of 0).

[0047] These labeled real posture data and generated posture data samples are input into the discriminator for training, and the parameters of the discriminator are updated using the back propagation algorithm. The purpose is to enable the discriminator to better distinguish between real posture data and generated posture data, that is, to make the probability of the discriminator output for real posture data closer to 1, and the probability of the output for generated posture data closer to 0.

[0048] When training the generator, fix the parameters of the discriminator (in actual implementation, the parameters of the discriminator are not updated temporarily to train the generator first), and let the generator generate a batch of fake data samples as generated posture data samples again, and input these generated posture data samples into the discriminator.

[0049] In this way, we hope that the probability of the discriminator outputting the data generated by the generator is as high as possible (close to 1), because this means that the data generated by the generator can "fool" the discriminator into thinking it is real posture data. The parameters of the generator are updated according to the output of the discriminator through the back propagation algorithm.

[0050] The steps of training the discriminator and generator are continuously implemented, and in each round of training, the training data is iterated multiple times, and the parameters of the generator and discriminator are updated in each iteration.

[0051] In this process, the performance of the generator and the discriminator will continue to improve. The data generated by the generator will become more and more realistic, and the discriminator will be more and more capable of distinguishing between real and generated pose data. Until certain training stop conditions are reached, such as reaching the preset number of training rounds, the loss of the generator and the discriminator converges, etc.

[0052] During the training process, the rocket attitude data including attitude angle data, navigation speed data and position data need to be verified in order to evaluate the trained generative adversarial network model.

[0053] Among them, the verification of attitude angle data includes: verifying whether the generated data meets the physical laws of the rocket attitude. In the flight attitude data, the model evaluation of attitude angle, navigation speed, position and timestamp needs to comprehensively consider the physical rationality, dynamic consistency and specific indicators of the mission requirements of the data.

[0054] Verification of navigation velocity data includes: evaluating the physical range, checking whether the generated attitude angle is within a reasonable range; calculating the rate of change of the angle sequence , the frequency domain characteristics of the generated angles are analyzed using Fourier transform to check whether there is high-frequency noise; through autocorrelation analysis, it is evaluated whether the generated angle sequence is consistent with the real angle sequence; finally, the attitude angle data is applied to the flight simulation model to check whether the generated data supports a stable flight attitude.

[0055] Verification of position data includes: visualizing the generated longitude, latitude, and altitude data on a map to see if the trajectory distribution is reasonable; checking if the trajectory points meet the aircraft trajectory restrictions; calculating the distance change between trajectory points to check if there are mutations (such as jump points); and using kinematic formulas to verify if the position matches the speed and angle.

[0056] S4. Use the optimal generative adversarial network model to generate rocket flight attitude data.

[0057] When the training of the generative adversarial network model is completed and the optimal generative adversarial network model is obtained, the generator in the optimal generative adversarial network model can be used to generate new rocket posture data. By inputting a random latent vector into the generator, the generator will output new generated posture data based on the patterns and features learned during the training process.

[0058] Through the above process, the generative adversarial network model can be used to efficiently generate rocket flight attitude data, providing strong data support for the design, testing and control system of the rocket. The generated attitude data can be used to simulate the complex environment in flight, and generate attitude data under special conditions such as high wind disturbance and uneven fuel to verify the robustness of the control algorithm. Use the normal data generated by the generative adversarial network model as a benchmark, compare it with the actual flight data, and identify anomalies. The attitude data generated by the generative adversarial network model can also be used to supplement the original data set to reversely verify and improve the training effect of the attitude control algorithm.

[0059] Based on the same inventive concept, the present invention also provides a computing device, including: a processor and a memory storing a computer program, and when the computer program is run by the processor, the rocket attitude data processing method is executed.

[0060] Based on the same inventive concept, the present invention also provides a computer-readable storage medium, on which computer instructions are stored. When the instructions are executed on a computer, the computer executes the rocket attitude data processing method.

[0061] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0062] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein.

[0063] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A rocket attitude data processing method, characterized in that: include: Acquire the real attitude data of the rocket in motion and pre-process the acquired real attitude data; Build a generative adversarial network model and define the loss function of the generative adversarial network model; Train the generative adversarial network model, and evaluate and iteratively update the trained generative adversarial network model to obtain the optimal generative adversarial network model; Generate rocket flight attitude data using the optimal generative adversarial network model; The real posture data includes posture angle, navigation speed, position and timestamp.

2. The rocket attitude data processing method according to claim 1, characterized in that: The preprocessed real posture data must meet the following conditions: The pre-processed attitude angle changes cyclically within the range of -180° to 180°, with continuity and smoothness, avoiding sudden changes or unreasonable oscillations; The pre-processed navigation speed complies with the flight physics including the maximum acceleration and deceleration range and is consistent with the flight mission phase; The preprocessed position satisfies the physical trajectory constraint; the preprocessed position includes longitude, latitude and altitude, and the altitude is consistent with the attitude angle and navigation speed; The preprocessed timestamps are time series data with the correct time interval and time point order.

3. The rocket attitude data processing method as claimed in claim 2, characterized in that: Defining the loss function of the generative adversarial network model includes defining a loss function of the discriminator and a loss function of the generator in the generative adversarial network model; Among them, the loss function of the discriminator is used to evaluate the discriminator's ability to distinguish between real posture data and generated posture data; the loss function of the generator is used to evaluate the generator's ability to deceive the discriminator.

4. The rocket attitude data processing method according to claim 3, characterized in that: The expression of the discriminator loss function is: ； In the formula, The average log probability of the discriminator correctly classifying the real pose data; The real posture data is input to the discriminator; is the output value of the discriminator for the true posture data; is the average log probability of the generated pose data being correctly classified as false by the discriminator; is the discriminant output value of the generated posture data by the discriminator; The posture data is generated by the generator; is the random noise of the input.

5. The rocket attitude data processing method according to claim 4, characterized in that: The expression of the generator loss function is: ； In the formula, Indicates that the discriminator believes that the generated posture data is The logarithm of the true probability.

6. The rocket attitude data processing method according to claim 5, characterized in that: In the evaluation of the trained generative adversarial network model, the verification of the attitude angle data includes: Check whether the generated posture angle is within a reasonable range; Calculate the rate of change of the angle sequence, use Fourier transform to analyze the frequency domain characteristics of the generated angles, and check whether there is high-frequency noise; Through autocorrelation analysis, it is evaluated whether the generated angle sequence is consistent with the real angle sequence; Apply the attitude angle data to the flight simulation model to check whether the generated data supports a stable flight attitude.

7. The rocket attitude data processing method according to claim 6, characterized in that: In the evaluation of the trained generative adversarial network model, the verification of the navigation speed data includes: Check if the generated speed value is within a reasonable range; Analyze the speed direction to ensure there is no abnormal direction; Calculate the derivatives of the velocity series to check whether the acceleration complies with the physical constraints of the aircraft; The correlation coefficient is used to evaluate the relationship between speed and attitude angle to avoid excessive angle changes when the navigation speed is too high.

8. The rocket attitude data processing method according to claim 6, characterized in that: In the evaluation of the trained generative adversarial network model, the verification of the location data includes: Visualize the generated longitude, latitude, and altitude data on the map to see if the trajectory distribution is reasonable; Check if the trajectory points meet the aircraft trajectory constraints; Calculate the distance change between trajectory points and check whether there is a mutation; Use kinematic formulas to verify that position, velocity, and angle match.

9. A computing device, characterized in that include: A processor and a memory storing a computer program, wherein when the computer program is executed by the processor, the rocket attitude data processing method as described in any one of claims 1 to 8 is executed.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and when the instructions are executed on a computer, the computer executes the rocket attitude data processing method as described in any one of claims 1 to 8.

Images