Aero-engine digital twin modeling optimization method based on transfer learning

Abstract

The present application relates to the technical field of modeling, in particular to an aero-engine digital twin modeling optimization method based on transfer learning. The present application establishes a first digital twin model by combining random forest algorithm with environmental feature data and control feature data, so as to capture the performance and behavior changes of the aero-engine under different working conditions. The second digital twin model can be fine-tuned and optimized according to real-time data through transfer learning, thereby improving the generalization ability and adaptability of the model. The third digital twin model can adapt to the data distribution changes and offsets that may exist in the target task through transfer learning. The transfer learning of the source domain data helps to make the model more robust and effectively cope with these changes. Based on the transfer learning data of the third digital twin model, the parameters of the second digital twin model can be fine-tuned and optimized. This optimization not only improves the prediction accuracy of the model, but also helps to improve the generalization ability of the model on new data.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of modeling, in particular to an aviation engine digital twin modeling optimization method based on transfer learning. BACKGROUND

[0002] Transfer learning is an algorithm that uses the knowledge obtained by the source domain in solving the task to improve the target task, given the source domain and source task, target domain and target task.

[0003] The aviation engine is one of the core components of the aircraft, responsible for providing power to push the aircraft to fly, and the aviation engine digital twin modeling is to establish a simulation model in the virtual space for the real engine in the physical space. Through simulating and analyzing the running state of the aviation engine, engineers can optimize the design, monitor the operation and perform maintenance. The current aviation engine digital twin modeling is modeled by sample data, but there may be insufficient sample data, resulting in insufficient precision of the established model, so that the output exhaust temperature parameter is not accurate enough. SUMMARY

[0004] The main purpose of the present application is to provide an aviation engine digital twin modeling optimization method based on transfer learning, which aims to solve the technical problems in the prior art.

[0005] The present application provides an aviation engine digital twin modeling optimization method based on transfer learning, comprising:

[0006] Obtain historical target data and a plurality of real-time target data of the aviation engine, wherein the historical target data comprises environmental feature data and control feature data;

[0007] A random forest algorithm is used to construct a first digital twin model according to the environmental feature data and the control feature data;

[0008] The first digital twin model is trained according to a plurality of real-time target data to obtain a second digital twin model;

[0009] Obtain source domain data related to the historical target data;

[0010] Select a transfer learning model according to the source domain data, and input the source domain data into the transfer learning model for training to obtain a third digital twin model;

[0011] Obtain transfer learning data between the third digital twin model and the second digital twin model, and optimize the second digital twin model according to the transfer learning data to obtain a digital twin optimization model.

[0012] As preferred, the step of employing the random forest algorithm to construct the first digital twin model according to the environmental feature data and the control feature data comprises:

[0013] Atmospheric density data is acquired according to the environmental feature data, and fuel flow data is acquired according to the control feature data;

[0014] The cross-sectional area of the air intake of the aero-engine, the intake temperature of the air intake, the first total pressure of the air intake, and the second total pressure of the exhaust port are acquired;

[0015] The first digital twin model is constructed according to the cross-sectional area, the intake temperature, the first total pressure, the second total pressure, the fuel flow data, and the atmospheric density data, wherein the first digital twin model is:

[0016]

[0017] Wherein P(W) represents the predicted exhaust temperature, J(W) represents the intake temperature, R(L) represents the fuel flow data, J(M) represents the cross-sectional area, QY(1) represents the first total pressure, QY(2) represents the second total pressure, k represents the first empirical weight coefficient, n represents the second empirical weight coefficient, and m represents the third empirical weight coefficient.

[0018] As preferred, the step of acquiring atmospheric density data according to the environmental feature data and acquiring fuel flow data according to the control feature data comprises:

[0019] Temperature parameters, pressure parameters, and height parameters are acquired according to the environmental feature data;

[0020] Temperature gradients and gas constants in the environment where the aero-engine is located are acquired;

[0021] Atmospheric density data is calculated according to the temperature parameters, the pressure parameters, the height parameters, the temperature gradients, and the gas constants, wherein the calculation formula is:

[0022]

[0023] Wherein Q(M) represents the atmospheric density data, Y(C) represents the pressure parameters, W(C) represents the temperature parameters, Q(C) represents the gas constants, W(T) represents the temperature gradients, and H(C) represents the height parameters;

[0024] Throttle lever angle parameters, flight speed parameters, and throttle opening parameters are acquired according to the control feature data;

[0025] The basic fuel flow of the aero-engine is acquired;

[0026] The fuel flow data is calculated according to the base fuel flow, the throttle lever angle parameter, the flight speed parameter and the throttle opening degree parameter, and the calculation formula is:

[0027] R(L) = J(R) * [a * Y(J) + b * F(S) + c * Y(K)];

[0028] Wherein, R(L) represents the fuel flow data, J(R) represents the base fuel flow, a represents the weight of the throttle lever angle parameter, Y(J) represents the throttle lever angle parameter, b represents the weight of the flight speed parameter, F(S) represents the flight speed parameter, c represents the weight of the throttle opening degree parameter, and Y(K) represents the throttle opening degree parameter.

[0029] Preferably, the step of training the first digital twin model according to a plurality of real-time target data to obtain a second digital twin model comprises:

[0030] Any one of the real-time target data is selected as a training set and input into the first digital twin model for training, and the remaining real-time target data is input into the first digital twin model as a verification set for verification, to obtain a plurality of training results;

[0031] The model parameters of the first digital twin model are adjusted according to a plurality of the training results to obtain a second digital twin model.

[0032] Preferably, the step of selecting a transfer learning model according to the source domain data and inputting the source domain data into the transfer learning model for training to obtain a third digital twin model comprises:

[0033] The first data distribution information, the first data type information, the first data quality information and the first data label availability information are obtained according to the source domain data;

[0034] The second data distribution information, the second data type information, the second data quality information and the second data label availability information are obtained according to the historical target data;

[0035] The transfer learning model is determined according to the first data distribution information, the first data type information, the first data quality information and the first data label availability information and the second data distribution information, the second data type information, the second data quality information and the second data label availability information;

[0036] The model parameters in the transfer learning model are determined;

[0037] The source domain data is sequentially standardized and enhanced for preprocessing to obtain preprocessed data;

[0038] The preprocessed data is input into the transfer learning model for training to obtain a third digital twin model.

[0039] As preferred, the step of optimizing the second digital twin model according to the transfer learning data to obtain a digital twin optimization model comprises:

[0040] acquiring historical target data and drawing the historical target data into a first histogram;

[0041] acquiring transfer learning data and drawing the transfer learning data into a second histogram;

[0042] judging whether the overlap degree of the first histogram and the second histogram is greater than a preset value;

[0043] if the overlap degree of the first histogram and the second histogram is greater than the preset value, determining that the optimizer and the loss function of the second digital twin model are Adam optimizer and cross-entropy loss function respectively;

[0044] optimizing the second digital twin model according to the Adam optimizer and the cross-entropy loss function to obtain a digital twin optimization model.

[0045] The application also provides an aviation engine digital twin modeling optimization system based on transfer learning, comprising:

[0046] a first acquisition module configured to acquire historical target data and a plurality of real-time target data of an aviation engine, wherein the historical target data comprises environmental feature data and control feature data;

[0047] a construction module configured to construct a first digital twin model according to the environmental feature data and the control feature data by using a random forest algorithm;

[0048] a first training module configured to train the first digital twin model according to a plurality of the real-time target data to obtain a second digital twin model;

[0049] a second acquisition module configured to acquire source domain data related to the historical target data;

[0050] a second training module configured to select a transfer learning model according to the source domain data and train the source domain data in the transfer learning model to obtain a third digital twin model;

[0051] an optimization module configured to acquire transfer learning data between the third digital twin model and the second digital twin model and optimize the second digital twin model according to the transfer learning data to obtain a digital twin optimization model.

[0052] As preferred, the construction module comprises:

[0053] A first obtaining unit is configured to obtain atmospheric density data according to the environmental feature data and fuel flow data according to the control feature data;

[0054] A second obtaining unit is configured to obtain a sectional area of an air intake of the aero-engine, an intake temperature of the air intake, a first total pressure of airflow of the air intake, and a second total pressure of airflow of an exhaust port;

[0055] A constructing unit is configured to construct a first digital twin model according to the sectional area, the intake temperature, the first total pressure of airflow, the second total pressure of airflow, the fuel flow data, and the atmospheric density data, wherein the first digital twin model is:

[0056]

[0057] wherein P(W) represents the predicted exhaust temperature, J(W) represents the intake temperature, R(L) represents the fuel flow data, J(M) represents the sectional area, QY(1) represents the first total pressure of airflow, QY(2) represents the second total pressure of airflow, k represents a first empirical weight coefficient, n represents a second empirical weight coefficient, and m represents a third empirical weight coefficient.

[0058] The application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the aero-engine digital twin modeling optimization method based on the transfer learning when executing the computer program.

[0059] The application further provides a computer readable storage medium, which stores a computer program, and the computer program implements the steps of the aero-engine digital twin modeling optimization method based on the transfer learning when executed by a processor.

[0060] The beneficial effects of the present application are: the present application establishes a first digital twin model by combining environmental feature data and control feature data through a random forest algorithm, thereby being able to capture the performance and behavior changes of the aero-engine under different working conditions, through transfer learning, the second digital twin model can learn some general features and patterns from the first digital twin model, and then fine-tune and optimize according to real-time data, such a method can reduce the dependence on a large amount of real-time data, improve the generalization ability and adaptability of the model, the optimization of the second digital twin model can significantly improve the prediction accuracy and accuracy of the model, the third digital twin model can better adapt to the data distribution changes and offsets that may exist in the target task through transfer learning, the transfer learning of the source domain data helps to make the model more robust and can effectively cope with these changes, based on the transfer learning data of the third digital twin model, the parameters of the second digital twin model can be fine-tuned and optimized, such optimization not only improves the prediction accuracy of the model, but also helps to improve the generalization ability of the model on new data, the optimized second digital twin model can more accurately monitor the running state and health condition of the aero-engine in real time, the use of transfer learning data enables the model to quickly adapt when the real-time data is updated, and potential problems or abnormalities can be discovered in time, through transfer learning optimization of the second digital twin model, the risk of model prediction error can be reduced, thereby improving the safety and reliability of the aero-engine. BRIEF DESCRIPTION OF DRAWINGS

[0061] Fig. 1 The method flowchart of an embodiment of the present application.

[0062] Fig. 2 The device structure schematic diagram of an embodiment of the present application.

[0063] Fig. 3 The internal structure schematic diagram of a computer device of an embodiment of the present application.

[0064] The implementation, functional features and advantages of the present application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0065] It should be understood that the specific embodiments described herein are merely intended to explain the present application, and are not intended to limit the present application.

[0066] As shown in Figs. 1-3 The present application provides a transfer learning-based aero-engine digital twin modeling optimization method, which comprises:

[0067] S1, obtaining historical target data and a plurality of real-time target data of an aero-engine, wherein the historical target data comprises environmental feature data and control feature data;

[0068] S2, using a random forest algorithm to construct a first digital twin model according to the environmental feature data and the control feature data;

[0069] S3, training the first digital twin model according to a plurality of real-time target data to obtain a second digital twin model;

[0070] S4, obtaining source domain data related to historical target data;

[0071] S5, selecting a transfer learning model according to the source domain data, and inputting the source domain data into the transfer learning model for training to obtain a third digital twin model;

[0072] S6, obtaining transfer learning data between the third digital twin model and the second digital twin model, and optimizing the second digital twin model according to the transfer learning data to obtain a digital twin optimization model.

[0073] As described in steps S1-S6, the present application obtains historical target data and multiple real-time target data of the aero-engine, wherein the historical target data includes environmental feature data and control feature data, and adopts a random forest algorithm to construct a first digital twin model according to the environmental feature data and the control feature data, and trains the first digital twin model according to the multiple real-time target data to obtain a second digital twin model. Through the random forest algorithm, the first digital twin model is established in combination with the environmental feature data and the control feature data. Since the first digital twin model is based on historical target data, it can capture the performance and behavior changes of the aero-engine under different working conditions and provide a benchmark prediction model. Moreover, the random forest algorithm is suitable for processing complex nonlinear relationships and a large number of features, and can effectively capture potential patterns and rules in historical data. Since the second digital twin model is trained using multiple real-time target data, these real-time data may reflect the latest operating state of the current aero-engine, environmental changes and other factors, and can update model parameters in real time, thereby maintaining the accuracy and reliability of the model. Through transfer learning, the second digital twin model can learn some general features and patterns from the first digital twin model, and then fine-tune and optimize it according to real-time data. This method can reduce the dependence on a large amount of real-time data, improve the generalization ability and adaptability of the model, and the optimization of the second digital twin model can significantly improve the prediction accuracy and accuracy of the model. It can quickly adapt to changes in engine operating conditions and capture key features in real-time data, thereby more accurately predicting the performance and health status of the engine. The source domain data related to the historical target data is obtained, a transfer learning model is selected according to the source domain data, and the source domain data is input into the transfer learning model for training to obtain a third digital twin model. Since the source domain data usually contains features and patterns related to the target task but not exactly the same, through transfer learning, the third digital twin model can use the knowledge and features in the source domain data to help the model better understand the target task data, which can improve the generalization ability of the model. Transfer learning can reduce the labeling and collection cost on the target domain data. By using existing source domain data, a large amount of labeled data in the target task can be avoided, which is particularly important for high-cost fields such as aero-engines, because the acquisition and labeling of real-time data may be very expensive and difficult. Based on the source domain data, transfer learning can accelerate the training process of the model. Compared with training the model from scratch, transfer learning usually requires fewer iterations and tuning, so that the model can be quickly deployed to practical applications. Through transfer learning, the third digital twin model can better adapt to the data distribution changes and shifts that may exist in the target task. The transfer learning of the source domain data helps to make the model more robust and can effectively cope with these changes. The optimized third digital twin model can support real-time decision-making, and it can train a more accurate and reliable model based on historical data and source domain data.The third digital twin model can provide additional data features or feature adjustments that may not have been fully utilized or considered in the second digital twin model. Through transfer learning, these new features can be integrated into the second digital twin model, thereby improving its understanding and prediction ability of the working conditions of the aero-engine. Based on the transfer learning data of the third digital twin model, the parameters of the second digital twin model can be fine-tuned and optimized. This optimization not only improves the prediction accuracy of the model, but also helps to improve the generalization ability of the model on new data. The optimized second digital twin model can more accurately monitor the running state and health condition of the aero-engine in real time. The use of transfer learning data enables the model to quickly adapt to real-time data updates and timely detect potential problems or abnormalities. Through transfer learning optimization of the second digital twin model, the risk of model prediction error can be reduced, thereby improving the safety and reliability of the aero-engine.

[0074] In one embodiment, the step S2 of employing a random forest algorithm to construct a first digital twin model according to the environmental feature data and the control feature data comprises:

[0075] S21, obtaining atmospheric density data according to the environmental feature data, and obtaining fuel flow data according to the control feature data;

[0076] S22, obtaining the cross-sectional area of the aero-engine inlet, the inlet temperature of the inlet, the first total pressure of the inlet, and the second total pressure of the exhaust port;

[0077] S23, constructing a first digital twin model according to the cross-sectional area, the inlet temperature, the first total pressure, the second total pressure, the fuel flow data and the atmospheric density data, wherein the first digital twin model is:

[0078]

[0079] wherein P(W) represents the predicted exhaust temperature, J(W) represents the inlet temperature, R(L) represents the fuel flow data, J(M) represents the cross-sectional area, QY(1) represents the first total pressure, QY(2) represents the second total pressure, k represents a first empirical weight coefficient, n represents a second empirical weight coefficient, and m represents a third empirical weight coefficient.

[0080] As described in steps S21-S23 above, the present application obtains atmospheric density data from environmental characteristic data, obtains fuel flow data from control characteristic data, obtains the cross-sectional area of the aero-engine intake, the intake temperature of the intake, the first total pressure of the intake, and the second total pressure of the exhaust, and constructs a first digital twin model based on the cross-sectional area, the intake temperature, the first total pressure, the second total pressure, the fuel flow data, and the atmospheric density data. The first digital twin model can more comprehensively simulate the working state of the aero-engine by including data such as atmospheric density, fuel flow, intake, and exhaust physical parameters, which directly affect the combustion process, intake and exhaust performance of the engine, and are important factors for evaluating the performance and health state of the engine. The model constructed using these physical parameters can more accurately capture the complex physical and chemical processes inside the engine. For example, atmospheric density affects the air-fuel ratio, while fuel flow directly affects the combustion rate and engine output power. Considering environmental characteristic data such as atmospheric density, the model can better adapt to different environmental conditions, such as high-altitude areas or large temperature changes, which helps to improve the prediction accuracy and reliability of the model in practical applications. The first digital twin model can reflect the current working state and performance of the aero-engine in real time by combining real-time fuel flow, total pressure, and other data. The first digital twin model constructed using environmental characteristic data and control characteristic data can better adapt to new working conditions and data distribution in target tasks, and this enhanced adaptability helps the model to be applied in different flight tasks or different engine configurations.

[0081] In one embodiment, the step S21 of obtaining atmospheric density data from the environmental characteristic data and obtaining fuel flow data from the control characteristic data comprises:

[0082] S211, obtaining temperature parameters, pressure parameters, and height parameters from the environmental characteristic data;

[0083] S212, obtaining temperature gradient and gas constant in the environment where the aero-engine is located;

[0084] S213, calculating atmospheric density data according to the temperature parameters, pressure parameters, height parameters, temperature gradient, and gas constant, wherein the calculation formula is:

[0085]

[0086] wherein Q(M) represents atmospheric density data, Q(C) represents pressure parameters, W(C) represents temperature parameters, Q(C) represents gas constant, W(T) represents temperature gradient, and H(C) represents height parameters;

[0087] S214, obtaining a throttle lever angle parameter, a flight speed parameter and a throttle opening degree parameter according to the control feature data;

[0088] S215, obtaining a basic fuel flow of the aero-engine;

[0089] S216, calculating fuel flow data according to the basic fuel flow, the throttle lever angle parameter, the flight speed parameter and the throttle opening degree parameter, wherein a calculation formula is:

[0090] R(L) = J(R) * [a * Y(J) + b * F(S) + c * Y(K)];

[0091] wherein R(L) represents the fuel flow data, J(R) represents the basic fuel flow, a represents a weight of the throttle lever angle parameter, Y(J) represents the throttle lever angle parameter, b represents a weight of the flight speed parameter, F(S) represents the flight speed parameter, c represents a weight of the throttle opening degree parameter, and Y(K) represents the throttle opening degree parameter.

[0092] As described in steps S211-S216 above, the application obtains temperature parameters, pressure parameters, and altitude parameters, as well as temperature gradients and gas constants in the environment of the aero-engine, and calculates atmospheric density data based on the temperature parameters, pressure parameters, altitude parameters, temperature gradients, and gas constants. By considering environmental parameters such as temperature, pressure, and altitude, as well as temperature gradients and gas constants, the atmospheric density model constructed can more accurately reflect the physical state of the environment in which the aero-engine is located. This accuracy is crucial for the prediction and simulation of the model, and can improve the accuracy and practicality of the model. Considering different temperature, pressure, and altitude parameters, the model can adapt to various working conditions and environmental scenarios. This adaptability makes the model more robust and reliable when facing new data or changing working conditions. Accurate calculation of atmospheric density data can help optimize the model's prediction of aero-engine performance and combustion process. Atmospheric density directly affects the air-fuel ratio and combustion efficiency, which is crucial for evaluating aero-engine output power and fuel efficiency. By considering multiple environmental parameters and physical characteristics, especially the accurate calculation of atmospheric density, the prediction error of the model can be effectively reduced, and the prediction ability, adaptability, and real-time monitoring support ability of the model can be significantly improved, thereby effectively optimizing the performance management and health monitoring of the aero-engine. By controlling the throttle lever angle parameter, flight speed parameter, and throttle opening parameter, and obtaining the basic fuel flow of the aero-engine, and calculating the fuel flow data based on the basic fuel flow, throttle lever angle parameter, flight speed parameter, and throttle opening parameter, the model can more comprehensively understand and simulate the complex process of engine fuel consumption by combining throttle lever angle, flight speed, throttle opening, and basic fuel flow. This comprehensive analysis can improve the accuracy and precision of fuel flow prediction. By accurately calculating and predicting fuel flow, the model can help optimize engine fuel consumption and operating efficiency. Based on accurate fuel flow data, the model can monitor the performance state of the engine in real time.

[0093] In one embodiment, the step S3 of training the first digital twin model according to a plurality of real-time target data to obtain a second digital twin model comprises:

[0094] S31, select any one real-time target data as a training set input to the first digital twin model for training, and input the remaining real-time target data as a validation set to the first digital twin model for validation, to obtain a plurality of training results;

[0095] S32, adjust the model parameters of the first digital twin model according to a plurality of training results to obtain a second digital twin model.

[0096] As described in steps S31-S32 above, the present application can train and validate the first digital twin model under multiple different data distributions by selecting different real-time target data as the training set and validation set. This diversified training helps the model better adapt to different working conditions and operating scenarios, improving its generalization ability and robustness. Based on multiple training results, the model parameters of the first digital twin model are adjusted, and the structure and parameter settings of the model are optimized by comparing different training effects. This method can help find a model configuration that is more suitable for the current task and data characteristics, improving the performance and effectiveness of the model. The optimized second digital twin model can process real-time data more quickly and make accurate predictions or decisions, which is crucial for real-time monitoring, health management, and performance optimization of aero-engines. By effectively utilizing training data and optimizing model structure, the cost of model training and deployment can be reduced, which not only saves time and resources but also improves the actual application efficiency and cost-effectiveness of the model. By comparing and verifying multiple training results, the model's explainability and trustworthiness can be enhanced.

[0097] In one embodiment, the step S5 of selecting a transfer learning model according to the source domain data and inputting the source domain data into the transfer learning model for training to obtain a third digital twin model comprises:

[0098] S51, obtaining first data distribution information, first data type information, first data quality information, and first data label availability information according to the source domain data;

[0099] S52, obtaining second data distribution information, second data type information, second data quality information, and second data label availability information according to the historical target data;

[0100] S53, determining a transfer learning model according to the first data distribution information, first data type information, first data quality information, and first data label availability information, and the second data distribution information, second data type information, second data quality information, and second data label availability information;

[0101] S54, determining model parameters in the transfer learning model;

[0102] S55, sequentially performing standardization and enhancement preprocessing on the source domain data to obtain preprocessed data;

[0103] S56, inputting the preprocessed data into the transfer learning model for training to obtain a third digital twin model.

[0104] As described in steps S51-S56, the present application obtains first data distribution information, first data type information, first data quality information, and first data label availability information from source domain data, and obtains second data distribution information, second data type information, second data quality information, and second data label availability information from historical target data, and determines a transfer learning model according to the first data distribution information, the first data type information, the first data quality information, and the first data label availability information, and the second data distribution information, the second data type information, the second data quality information, and the second data label availability information. By comparing the distribution information of the source domain data and the historical target data, the similarity or difference in the feature space can be evaluated, which helps to determine when the source domain data can be directly applied, and when adaptive adjustment or specific data preprocessing is needed to ensure the consistency and availability of the data. Analyzing the types (such as numerical, textual, etc.) and quality (such as data integrity, missing values, etc.) of the first data and the second data helps to understand the applicability and processing needs of the data. For example, if the historical target data quality is low, additional data cleaning or supplementation may be needed to improve the stability and accuracy of the model. Evaluating the label availability of the first data and the second data, i.e., whether the data has been labeled and the quality of the labeling, is particularly important for supervised learning models, which can guide the model to choose the appropriate learning method and training strategy to maximize the use of available labeled data. Based on the distribution, type, quality, and label availability information of the data, appropriate transfer learning methods and strategies can be selected. For example, if the source domain data and the target data are similar in data distribution, simple transfer learning techniques can be directly applied, and if the difference is large, more complex domain adaptation methods are needed to adjust the model. Based on comprehensive data analysis and evaluation, the transfer learning model constructed can provide more reliable and accurate prediction results, which is of great significance to improving engine performance, reducing maintenance costs, and enhancing flight safety. By determining the model parameters in the transfer learning model, the source domain data is sequentially standardized and enhanced for preprocessing to obtain preprocessed data, and then the preprocessed data is input into the transfer learning model for training to obtain a third digital twin model. Through standardization and enhancement preprocessing, the consistency and quality of the source domain data can be improved to ensure that the data has similar data distribution and feature representation before being input into the model, which helps to reduce the risk of model training difficulty or performance degradation caused by inconsistent data, and enhances the stability and reliability of the model. By training the third digital twin model through the transfer learning model, the model can further adapt to new target tasks and data environments based on the source domain data and historical target data. This adaptability enables the model to better generalize to different operating conditions and environments, improving the model's practicality and adaptability. By determining appropriate transfer learning model parameters, the performance and efficiency of the model can be optimized to better adapt to the needs of aviation engine digital twin modeling. The optimized data is standardized and enhanced for preprocessing,The efficiency and speed of model training can be improved, and the third digital twin model optimized through training and preprocessing can provide more accurate and reliable prediction results and decision support, which is of great significance for aviation engine performance optimization, fault prediction and safety management, and helps to improve the efficiency and safety of aviation operations.

[0105] In one embodiment, the step S6 of optimizing the second digital twin model according to the transfer learning data to obtain a digital twin optimization model comprises:

[0106] S61, obtaining historical target data and drawing the historical target data into a first histogram;

[0107] S62, obtaining transfer learning data and drawing the transfer learning data into a second histogram;

[0108] S63, judging whether the overlap degree of the first histogram and the second histogram is greater than a preset value;

[0109] If the overlap degree of the first histogram and the second histogram is greater than the preset value, it is determined that the optimizer and the loss function of the second digital twin model are Adam optimizer and cross-entropy loss function respectively;

[0110] S64, optimizing the second digital twin model according to the Adam optimizer and the cross-entropy loss function to obtain a digital twin optimization model.

[0111] As described in steps S101-S103, the present application obtains historical target data and transfer learning data, and draws first and second histograms of the historical target data and the transfer learning data, respectively. By judging whether the overlap of the first and second histograms is greater than a preset value, if it is greater than the preset value, the optimizer and loss function of the second digital twin model are determined to be Adam optimizer and cross-entropy loss function, respectively. Then, the second digital twin model is optimized according to the Adam optimizer and the cross-entropy loss function to obtain a digital twin optimization model. By drawing histograms of the historical target data and the transfer learning data and comparing their overlap, it can be directly evaluated whether the data distribution is similar. If the overlap of the histograms of the two is higher than the preset value, it indicates that the data is consistent in the feature space and is suitable for direct application in the transfer learning model. This helps to ensure the generalization ability and stability of the model on new data. If it is determined that the overlap of the histograms meets the preset condition, the Adam optimizer and the cross-entropy loss function can be selected to optimize the second digital twin model. The Adam optimizer can usually effectively adjust the learning rate and adaptively adjust the gradient of different features, which is suitable for processing variable aero-engine data. The cross-entropy loss function performs well in classification tasks and can help the model effectively learn and optimize the prediction results, especially for tasks such as anomaly detection and fault prediction, which have high sensitivity and accuracy. Using appropriate optimizers and loss functions to optimize the second digital twin model can improve the performance and prediction accuracy of the model in aero-engine digital twin modeling, which is crucial for real-time monitoring of engine state, fault prediction and optimization of maintenance plan, and helps to improve aviation safety and operational efficiency. By fully utilizing histogram analysis of historical target data and transfer learning data, additional data processing costs and model adjustment time can be reduced. This method not only improves the efficiency of model construction, but also quickly responds to different data environments and task requirements, reducing technical risks and costs in the modeling process.

[0112] The present application also provides an aero-engine digital twin modeling optimization system based on transfer learning, comprising:

[0113] A first obtaining module is configured to obtain historical target data and a plurality of real-time target data of an aero-engine, wherein the historical target data comprises environmental feature data and control feature data.

[0114] A construction module is configured to construct a first digital twin model according to the environmental feature data and the control feature data using a random forest algorithm.

[0115] A first training module is configured to train the first digital twin model according to a plurality of real-time target data to obtain a second digital twin model.

[0116] The second acquisition module is configured to acquire source domain data related to historical target data.

[0117] The second training module is configured to select a transfer learning model according to the source domain data, and input the source domain data into the transfer learning model for training to obtain a third digital twin model.

[0118] The optimization module is configured to acquire transfer learning data between the third digital twin model and the second digital twin model, and optimize the second digital twin model according to the transfer learning data to obtain a digital twin optimization model.

[0119] In one embodiment, the construction module comprises:

[0120] The first acquisition unit is configured to acquire atmospheric density data according to the environmental feature data, and acquire fuel flow data according to the control feature data.

[0121] The second acquisition unit is configured to acquire a sectional area of an air intake of the aero-engine, an intake temperature of the air intake, a first total air flow pressure of the air intake, and a second total air flow pressure of an exhaust port.

[0122] The construction unit is configured to construct a first digital twin model according to the sectional area, the intake temperature, the first total air flow pressure, the second total air flow pressure, the fuel flow data and the atmospheric density data, wherein the first digital twin model is:

[0123]

[0124] wherein P(W) represents a predicted exhaust temperature, J(W) represents the intake temperature, R(L) represents the fuel flow data, J(M) represents the sectional area, QY(1) represents the first total air flow pressure, QY(2) represents the second total air flow pressure, k represents a first empirical weight coefficient, n represents a second empirical weight coefficient, and m represents a third empirical weight coefficient.

[0125] The application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the aero-engine digital twin modeling optimization method based on transfer learning when executing the computer program.

[0126] The application further provides a computer readable storage medium, which stores a computer program, and the computer program implements the steps of the aero-engine digital twin modeling optimization method based on transfer learning when executed by a processor.

[0127] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when the computer program is executed, the processes of the above-mentioned embodiment methods can be included. Any reference to memory, storage, databases, or other media in this application and in examples used herein, unless specifically stated otherwise, can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in many forms such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0128] It should be noted that in this document, the terms "comprising", "including", or any other variant thereof are intended to cover a non-exclusive inclusion, such that a process, device, article, or method that comprises a list of elements does not only include those elements, but can also include other elements not expressly listed or inherent to such process, device, article, or method. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, device, article, or method that includes the element.

[0129] The above description is only the preferred embodiment of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation, or direct or indirect application in other related technical fields, based on the content of the present application specification and drawings, is also included in the patent protection scope of the present application.

Claims

1. A method for modeling and optimizing an aero-engine digital twin based on transfer learning, characterized in that, The method comprises the following steps: acquiring historical target data and a plurality of real-time target data of an aero-engine, wherein the historical target data comprises environmental characteristic data and control characteristic data; acquiring atmospheric density data according to the environmental characteristic data and fuel flow data according to the control characteristic data; acquiring a cross-sectional area of an air inlet of the aero-engine, an air inlet temperature of the air inlet, a first total air flow pressure of the air inlet, and a second total air flow pressure of an exhaust port; constructing a first digital twin model according to the cross-sectional area, the air inlet temperature, the first total air flow pressure, the second total air flow pressure, the fuel flow data, and the atmospheric density data, wherein the first digital twin model is: ； wherein, P(W) represents a predicted exhaust gas temperature, J(W) represents an intake air temperature, R(L) represents fuel flow data, J(M) represents a cross-sectional area, QY (1) represents a first air flow total pressure, QY (2) represents a second air flow total pressure, k represents a first empirical weight coefficient, n represents a second empirical weight coefficient, m represents a third empirical weight coefficient; training the first digital twin model according to a plurality of the real-time target data to obtain a second digital twin model; acquiring source domain data related to the historical target data; selecting a transfer learning model according to the source domain data and inputting the source domain data into the transfer learning model to train the third digital twin model; acquiring transfer learning data between the third digital twin model and the second digital twin model, and optimizing the second digital twin model according to the transfer learning data to obtain a digital twin optimization model.

2. The method of claim 1, wherein, The step of acquiring atmospheric density data according to the environmental characteristic data and fuel flow data according to the control characteristic data comprises: acquiring temperature parameters, pressure parameters, and height parameters according to the environmental characteristic data; acquiring a temperature gradient and a gas constant in an environment where the aero-engine is located; calculating the atmospheric density data according to the temperature parameters, the pressure parameters, the height parameters, the temperature gradient, and the gas constant, wherein the calculation formula is: ； wherein, Q(M) denotes atmospheric density data, Y(C) denotes pressure parameter, W (C) denotes temperature parameter, Q(C) denotes gas constant, W(T) denotes temperature gradient, H(C) denotes altitude parameter; acquiring throttle lever angle parameters, flight speed parameters, and throttle opening degree parameters according to the control characteristic data; acquiring a basic fuel flow of the aero-engine; calculating the fuel flow data according to the basic fuel flow, the throttle lever angle parameters, the flight speed parameters, and the throttle opening degree parameters, wherein the calculation formula is: ； wherein, R(L) represents fuel flow data, J(R) represents base fuel flow, a represents a weight of a throttle lever angle parameter, Y(J) represents a throttle lever angle parameter, b represents a weight of a flight speed parameter, F(S) represents a flight speed parameter, c represents a weight of a throttle opening parameter, Y(K) represents a throttle opening parameter.

3. The method of claim 1, wherein, The step of training the first digital twin model according to a plurality of the real-time target data to obtain a second digital twin model comprises: selecting any one of the real-time target data as a training set input into the first digital twin model for training, and inputting the remaining real-time target data as a validation set into the first digital twin model for validation to obtain a plurality of training results; adjusting model parameters of the first digital twin model according to a plurality of the training results to obtain a second digital twin model.

4. The method of claim 1, wherein, The step of selecting a transfer learning model according to the source domain data and inputting the source domain data into the transfer learning model to train the third digital twin model comprises: acquiring first data distribution information, first data type information, first data quality information, and first data label availability information according to the source domain data; acquiring second data distribution information, second data type information, second data quality information, and second data label availability information according to the historical target data; determine a transfer learning model according to the first data distribution information, the first data type information, the first data quality information, and the first data label availability information and the second data distribution information, the second data type information, the second data quality information, and the second data label availability information; determine model parameters in the transfer learning model; perform standardization and enhancement preprocessing on the source domain data in sequence to obtain preprocessed data; input the preprocessed data into the transfer learning model for training to obtain a third digital twin model.

5. The method of claim 1, wherein, The step of optimizing the second digital twin model according to the transfer learning data to obtain a digital twin optimization model comprises: obtaining historical target data and drawing the historical target data into a first histogram; obtaining transfer learning data and drawing the transfer learning data into a second histogram; determining whether the overlap degree of the first histogram and the second histogram is greater than a preset value; if the overlap degree of the first histogram and the second histogram is greater than the preset value, determining that the optimizer and the loss function of the second digital twin model are an Adam optimizer and a cross-entropy loss function, respectively; optimizing the second digital twin model according to the Adam optimizer and the cross-entropy loss function to obtain a digital twin optimization model.

6. An aero-engine digital twin modeling and optimization system based on transfer learning, characterized in that, comprise: a first obtaining module configured to obtain historical target data and a plurality of real-time target data of an aero-engine, wherein the historical target data comprises environmental feature data and control feature data; a constructing module configured to obtain atmospheric density data according to the environmental feature data and fuel flow data according to the control feature data; obtain a cross-sectional area of an air inlet of the aero-engine, an inlet temperature of the air inlet, a first total air flow pressure of the air inlet, and a second total air flow pressure of an exhaust port; construct a first digital twin model according to the cross-sectional area, the inlet temperature, the first total air flow pressure, the second total air flow pressure, the fuel flow data, and the atmospheric density data, wherein the first digital twin model is: ； wherein, P(W) represents a predicted exhaust gas temperature, J(W) represents an intake air temperature, R(L) represents fuel flow data, J(M) represents a cross-sectional area, QY (1) represents a first air flow total pressure, QY (2) represents a second air flow total pressure, k represents a first empirical weight coefficient, n represents a second empirical weight coefficient, m represents a third empirical weight coefficient; a first training module configured to train the first digital twin model according to the plurality of real-time target data to obtain a second digital twin model; a second obtaining module configured to obtain source domain data related to the historical target data; a second training module configured to select a transfer learning model according to the source domain data and input the source domain data into the transfer learning model for training to obtain a third digital twin model; an optimization module configured to obtain transfer learning data between the third digital twin model and the second digital twin model and optimize the second digital twin model according to the transfer learning data to obtain a digital twin optimization model. 7.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-6 when the computer program is executed by the processor. The processor executes the computer program to implement the steps of the method of any one of claims 1 to 5.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 5.

Images