Small sample electric aircraft composite material combined structure dynamics modeling method based on attention element learning

Abstract

The invention discloses a small sample electric aircraft composite material combined structure dynamics modeling method based on attention element learning. Comprising the following steps: step 1, a used data set comes from vibration experiments performed on a composite material laminated wing box structure, a composite material laminated air beam structure and a composite material conical-cylindrical combined shell under different temperature conditions; under the condition of limited data, various composite material structures and environment temperatures are divided into independent task samples, each task comprises vibration test data of the structure at the respective temperature, and the data are divided into a meta training set and a meta test set; the invention relates to the technical field of dynamic modeling, meta-learning and dynamic modeling of an electric aircraft composite material combined structure are creatively fused together, the AMLT model can utilize the rapid task adaptability and the multi-task learning ability of MAML to rapidly transfer and apply knowledge to a new task, and the modeling efficiency is improved. And the method has higher adaptability and generalization ability when facing new tasks.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of dynamics modeling, in particular to a small sample electric aircraft composite material combined structure dynamics modeling method based on attention meta learning. BACKGROUND

[0002] With green aviation becoming a focus, electric aircrafts provide a feasible technical solution for realizing green aviation due to the advantages of low carbon emissions. Composite material structures have many advantages such as high specific strength, specific stiffness, lightweight characteristics, fatigue resistance and corrosion resistance, etc., so the RX series general aviation electric aircraft adopts various composite material combined structures, such as all-composite fuselage, composite wing box and composite assembly structure, which accounts for more than 80% of the composition of the aircraft. Due to the lightweight development of electric aircrafts, the RX series electric aircraft requires these complex composite material combined structures to have excellent dynamic performance to reduce the probability of failure. However, traditional dynamics experiments of complex composite material combined structures are very complex and challenging, such as the need for special molds and fixtures for shaping test pieces, time-consuming and expensive vibration tests, and difficulty in controlling the experimental environment temperature, all of which limit their development. Therefore, an accurate dynamics modeling method is of great significance.

[0003] Dynamics modeling can be roughly divided into two categories, namely model-based methods and data-driven methods. Model-based methods largely depend on a large amount of professional knowledge, and the model usually shows weak generalization ability. With the development of artificial intelligence, deep learning and machine learning are widely used in dynamics modeling due to their strong feature extraction ability and generalization ability, but both of them depend on a large amount of data and have low accuracy and generalization ability under small sample data conditions. On the contrary, meta learning algorithms have attracted great interest due to their excellent generalization ability, and they are particularly effective in handling tasks with small sample size, complex systems and nonlinear characteristics. SUMMARY

[0004] In view of the shortcomings of the prior art, the present application provides a small sample electric aircraft composite material combined structure dynamics modeling method based on attention meta learning, which specifically introduces an attention meta learning Transformer model. This method can more effectively and accurately model the dynamics of small sample electric aircraft composite material combined structures, and solves the problem that existing model-based methods largely depend on a large amount of professional knowledge and the model usually shows weak generalization ability.

[0005] To achieve the above purpose, the present application is implemented by the following technical scheme: a small sample electric aircraft composite material combined structure dynamics modeling method based on attention meta learning, comprising the following steps: Step 1: The used dataset is derived from vibration experiments on composite laminated wing box (CLWBs) structures, composite laminated air beam (CLABs) structures, and composite conic-cylinder combined (CCCCs) shells under different temperature conditions. In the case of limited data, various composite structures and environmental temperatures are divided into independent task samples, each task including vibration test data of the structure at the respective temperature; the data is divided into meta-training and meta-testing sets; Step 2: Data set division, specifically, vibration test data of CLWBs structures at 20℃, 50℃, 100℃, and 125℃ is designated as test tasks; vibration test data of CLWBs structures under remaining temperature conditions, as well as CLABs and CCCC under all conditions, are designated as training tasks; each task is divided into a support set and a query set in a 4:1 ratio; Step 3: The support set data and labels in the meta-training set are imported into the model for training, and the AMLT model is trained to convergence. The Adam optimization algorithm is used during training, and the MSE loss function is used as the loss function. For each task, their respective support set data is used to quickly update the model parameters to achieve the optimal parameters for the current task environment; the update formula is: Secondly, the AMLT model uses a meta-learner to identify shared features in multiple temperature and multiple structure tasks, and dynamically adjusts the attention weight of the new task according to the similarity between the new task and the existing dataset; specifically, tasks that exhibit high similarity to the target task are given higher weights, while those with low similarity are given lower weights; The query set data and labels in the meta-training set are imported into the model for testing, and the meta-learner uses the query set from these different tasks to evaluate the performance of the base learner, and updates the initial model parameters again by calculating the error and performing backpropagation; the update formula is: This process is repeated until the training of all batch tasks is completed, until convergence is achieved in all structure and temperature tasks; Step 4: Meta-testing phase, the AMLT model is trained from the support set data divided from the test set, and the model parameters are updated to enhance adaptation to new tasks; subsequently, the query set divided from the test set is input into the model to evaluate the performance of the model on the new task, obtaining the dynamics model of CLWBs.

[0006] Step 5: Constructing model-agnostic meta-learning Transformer network, Transformer network, deep convolutional neural network, long short-term memory neural network, convolutional neural network-long short-term memory neural network, and bidirectional long short-term memory network respectively, input the same training set into the models for training, and input the test set and label into the trained models to obtain the amplitude-frequency response curves of CLWBs under these models respectively; Step 6: Compare the dynamic modeling results under the seven models, and use mean square error (MSE) and root mean square error (RMSE) to evaluate the prediction results. The calculation formulas of MSE and RMSE are as follows: In the formula, y is the predicted value, y is the actual value, and n is the length of the test sample. The accuracy of the model is evaluated by comparing MSE and RMSE.

[0007] Preferably, the step 1 normalizes the original vibration data. This method can convert the data to between [0, 1]. The operation of normalizing the original vibration data is represented as: Where x represents the initial data, x max and x min represent the maximum and minimum values of x, and y represents the normalized vector.

[0008] Preferably, the step 3 similarity is determined by the Pearson similarity coefficient between different structures and temperature characteristics. The formula for calculating the similarity coefficient is: Where r represents the similarity coefficient under one-dimensional characteristics, and n represents n-dimensional characteristics.

[0009] The application provides a small sample electric aircraft composite structure dynamics modeling method based on attention meta-learning, which has the following beneficial effects: meta-learning and electric aircraft composite structure dynamics modeling are innovatively combined together, the AMLT model can utilize the rapid task adaptability and multi-task learning ability of MAML to quickly transfer and apply knowledge to new tasks, which makes it have higher adaptability and generalization ability when facing new tasks. Since the AMLT model is trained on the current task dataset each time, it enables the network model to adjust its parameters according to the specific training task. At the same time, the AMLT model dynamically adjusts the influence weight for the target task through the attention mechanism during meta-training. Tasks that show higher feature consistency with the target task will obtain higher weights, so as to be able to retain prominent cross-task features. This reduces the significant prediction error caused by the compromise model when predicting the dynamics characteristics. By enabling the model to learn "meta-knowledge" across various tasks, optimal model initialization parameters for multiple tasks can be obtained. This helps to quickly iterate and converge the model with the least data in the actual prediction process, thereby reducing the requirement for training data in traditional neural networks. In addition, the model can quickly adapt to new tasks through a few gradient descent steps, thereby showing the best performance on these tasks. Compared with traditional deep learning models, the AMLT model has strong new task adaptability, faster calculation time and higher prediction accuracy in electric aircraft composite structure dynamics modeling. BRIEF DESCRIPTION OF DRAWINGS

[0010] Figure 1 is a flowchart of the small sample electric aircraft composite structure dynamics modeling method based on meta-learning of the application.

[0011] Figure 2 is a training flowchart of the AMLT model.

[0012] Figure 3 is a training rate diagram with and without meta-learning.

[0013] Figure 4 is a comparison diagram based on different deep learning dynamics modeling methods. DETAILED DESCRIPTION

[0014] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the application.

[0015] Please refer to Figures 1-4The application provides a technical scheme: a small sample electric aircraft composite material combined structure dynamics modeling method based on attention meta-learning, comprising the following steps: Step 1: The data used is derived from vibration experiments on composite material laminated wing box structures, composite material laminated air beam structures and composite material conical-cylindrical combined shell structures under different temperature conditions. In the case of limited data, various composite material structures and environmental temperatures are divided into independent task samples, each task including vibration test data of the structure at the respective temperature; the data is divided into a meta-training set and a meta-test set; Step 2: Data set division, specifically, vibration test data of the CLWBs structure at 20℃, 50℃, 100℃ and 125℃ is designated as a test task; vibration test data of the CLWBs structure under the remaining temperature conditions and vibration test data of the CLABs and CCCCCs under all conditions are designated as training tasks; each task is divided into a support set and a query set in a ratio of 4:1; Step 3: The support set data and labels in the meta-training set are imported into the model for training, and the AMLT model is trained to convergence, and the Adam optimization algorithm is used during training, and the MSE loss function is used as the loss function; for each task, the respective support set data is used to quickly update the model parameters to achieve the optimal parameters for the current task environment; the update formula is: Secondly, the AMLT model uses a meta-learner to identify shared features in multiple temperature and multiple structure tasks, and dynamically adjusts the attention weight of the new task according to the similarity between the new task and the existing data set; specifically, tasks that exhibit high similarity to the target task are given higher weights, while those with low similarity are given lower weights; The query set data and labels in the meta-training set are imported into the model for testing, and the meta-learner uses the query set from these different tasks to evaluate the performance of the base learner, and the initial model parameters are updated again through error calculation and back propagation; the update formula is: This process is repeated until the training of all batch tasks is completed, until convergence is achieved in all structure and temperature tasks; Step 4: Meta-test phase, the AMLT model is trained from the support set data divided from the test set, and the model parameters are updated to enhance adaptation to new tasks; subsequently, the query set divided from the test set is input into the model to evaluate the performance of the model on the new task, and the dynamics model of the CLWBs is obtained.

[0016] Step 5: Constructing model-agnostic meta-learning Transformer network, Transformer network, deep convolutional neural network, long short-term memory neural network, convolutional neural network-long short-term memory neural network, and bidirectional long short-term memory network respectively, input the same training set into the models for training, and input the test set and label into the trained models to obtain the amplitude-frequency response curves of CLWBs under these models respectively; Step 6: Compare the dynamic modeling results under the seven models, and use mean square error (MSE) and root mean square error (RMSE) to evaluate the prediction results. The calculation formulas of MSE and RMSE are as follows: In the formula, y is the predicted value, x is the actual value, and n is the length of the test sample. The accuracy of the model is evaluated by comparing MSE and RMSE.

[0017] Further, the step 1 normalizes the original vibration data. This method can convert the data to [0, 1]. The operation of normalizing the original vibration data is represented as: Where x represents the initial data, x max and x min represent the maximum and minimum values of x, and y represents the normalized vector.

[0018] Further, the similarity in step 3 is determined by the Pearson similarity coefficient between different structures and temperature characteristics. The formula for calculating the similarity coefficient is: Where r represents the similarity coefficient under one-dimensional characteristics, and n represents n-dimensional characteristics.

[0019] By the skilled person, the technology in the present case is operated in sequence. The specific operation sequence should refer to the working principle described below. The detailed connection means is a known technology in the art. The working principle and process are mainly introduced below.

[0020] Example: Step 1: The data used in this study was obtained from vibration experiments on composite laminated wing box (CLWBs) structures, composite laminated air beam (CLABs) structures, and composite conical-cylindrical combined shell (CCCCs) structures under different temperature conditions. In the case of limited data, various composite structures and environmental temperatures were divided into independent task samples, and each task included vibration test data of the structure at the respective temperature. The data was divided into a meta-training set and a meta-testing set. The original vibration data was normalized, which converted the data to between [0, 1]. The normalization operation on the original vibration data is represented as: where x represents the initial data, x max and x min represent the maximum and minimum values of x, and y represents the normalized vector.

[0021] Step 2: Data set division. Specifically, the vibration test data of CLWBs structures at 20°C, 50°C, 100°C, and 125°C was designated as the test task. The vibration test data of CLWBs structures under the remaining temperature conditions, as well as the vibration test data of CLABs and CCCC structures under all conditions, were designated as the training task. Each task was divided into a support set and a query set in a 4:1 ratio.

[0022] Step 3: The support set data and labels in the meta-training set were imported into the model for training, and the AMLT model was trained until convergence. The Adam optimization algorithm was used during training, and the MSE loss function was used as the loss function. For each task, their respective support set data was used to quickly update the model parameters to achieve the optimal parameters for the current task environment. The update formula is: Secondly, the AMLT model uses a meta-learner to identify shared features in multiple temperature and multiple structure tasks, and dynamically adjusts the attention weight of the new task according to the similarity between the new task and the existing data set. Specifically, tasks that exhibit high similarity to the target task are given higher weights, while those with low similarity are given lower weights. Similarity is determined by the Pearson correlation coefficient between different structure and temperature features. The formula for calculating the correlation coefficient is: where r represents the correlation coefficient under one-dimensional features, and n represents n-dimensional features.

[0023] The query set data and labels in the meta-training set were imported into the model for testing. The meta-learner uses the query set from these different tasks to evaluate the performance of the base learner, and updates the initial model parameters again by calculating the error and performing backpropagation. The update formula is: This process is repeated until all batch training tasks are completed, until convergence is achieved in all structure and temperature tasks.

[0024] Then the meta-testing phase is carried out, the AMLT model is trained from the support set data divided from the test set, and the model parameters are updated to enhance the adaptation to the new task; then, the query set divided from the test set is input into the model to evaluate the performance of the model on the new task, and the dynamic model of the CLWBs is obtained.

[0025] Then the model-agnostic meta-learning Transformer network, the Transformer network, the deep convolutional neural network, the long short-term memory neural network, the convolutional neural network-long short-term memory neural network, and the bidirectional long short-term memory network are constructed respectively, the same training set is input into the model for training, and the test set and the label are input into the trained model to obtain the amplitude-frequency response curve of the CLWBs under these models respectively; Finally, the dynamic modeling results under the seven models are compared, and the mean square error root mean square error is used to evaluate the prediction results, and the calculation formula of MSE and RMSE is: In the formula, The predicted value is y, The actual value is y, and n is the length of the test sample. By comparing MSE and RMSE, the accuracy of the model is evaluated It should be noted that in this text, relational terms such as first and second are used only to distinguish one entity or operation from another, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. Moreover, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or equipment. Without more limitations. The statement "including a limited element" does not exclude the existence of another identical element in the process, method, article or equipment including the element.

[0026] Although embodiments of the present application have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for dynamic modeling of composite material structures in small-sample electric aircraft based on attention meta-learning, characterized in that, Includes the following steps: Step 1: The dataset used comes from vibration experiments conducted on composite laminated wing box structures, composite laminated air beam structures, and composite conical-cylindrical composite shells under different temperature conditions. Given the limited data, various composite structures and ambient temperatures were divided into independent task samples, with each task including vibration test data of the structure at its respective temperature; the data was further divided into a meta-training set and a meta-test set. Step 2: Dataset partitioning. Specifically, vibration test data of CLWBs structures at 20℃, 50℃, 100℃ and 125℃ are designated as test tasks; vibration test data of CLWBs structures under the remaining temperature conditions and CLABs and CCCCs under all conditions are designated as training tasks; each task is divided into support set and query set in a 4:1 ratio. Step 3: Import the support set data and labels from the meta-training set into the model for training. Train the AMLT model until convergence, using the Adam optimization algorithm and the MSE loss function. For each task, quickly update the model parameters using their respective support set data to achieve optimal parameters for the current task environment; the update formula is: Secondly, the AMLT model uses a meta-learner to identify shared features in multi-temperature and multi-structure tasks, and dynamically adjusts the attention weights of the new task based on the similarity between the new task and the existing dataset. Specifically, tasks that show high similarity to the target task are given higher weights, while those with low similarity are given lower weights. The query set data and labels from the meta-training set are imported into the model for testing. The meta-learner uses query sets from these different tasks to evaluate the performance of the base learner, and updates the initial model parameters again by calculating the error and performing backpropagation; the update formula is: Repeat this process until training for all batch tasks is complete, and until convergence is achieved on all structure and temperature tasks. Step 4: In the meta-testing phase, the AMLT model is trained using support set data partitioned from the test set, and the model parameters are updated to enhance its adaptation to new tasks. Subsequently, the query set partitioned from the test set is input into the model to evaluate the model's performance on the new task, thus obtaining the dynamic model of CLWBs.

2. Step 5: Construct the following networks respectively: Model Agnostic Meta-Learning Transformer Network, Transformer Network, Deep Convolutional Neural Network, Long Short-Term Memory Neural Network, Convolutional Neural Network-Long Short-Term Memory Neural Network, and Bidirectional Long Short-Term Memory Network. Input the same training set into the model for training, and input the test set and labels into the trained model to obtain the amplitude-frequency response curves of CLWBs under these models respectively. Step 6: Compare the dynamic modeling results under the seven models. The root mean square error (RMSE) is used to evaluate the prediction results. The formulas for calculating MSE and RMSE are as follows: In the formula, For predicted values, is the actual value, and n is the test sample length. The accuracy of the model is evaluated by comparing MSE and RMSE.

3. The method for modeling the dynamics of composite material structures in small-sample electric aircraft based on attention meta-learning according to claim 1, characterized in that, Step 1 normalizes the original vibration data. This method transforms the data into the range [0, 1]. The normalization operation for the original vibration data is expressed as follows: Where x represents the initial data, x max and x min Let x represent the maximum and minimum values, and y represent the normalized vector.

4. The method for dynamic modeling of composite material structures of electric aircraft based on attention meta-learning according to claim 1, characterized in that, The similarity in step 3 is determined by the Pearson similarity coefficient between different structural and temperature characteristics. The formula for calculating the similarity coefficient is as follows: Where r represents the similarity coefficient under one-dimensional features; n represents n-dimensional features.

Images