A method and system for identifying aerodynamic parameters of aircraft based on neural network

Abstract

The invention discloses a method and system for aircraft aerodynamic parameter identification based on a neural network, which specifically relates to the field of aircraft aerodynamic parameter identification and is used to solve data processing and accuracy problems. The method and system avoid data mixing errors by constructing a time series data grid storage database based on flight state segmentation; calculate the nonlinear complexity index by phase space reconstruction to reveal the dynamic behavior and nonlinear characteristics of flight data; calculate the energy spectrum density by time-frequency analysis and construct a recursive graph, calculate the recursive graph complexity index, comprehensively evaluate the dynamic complexity of data, select the most suitable processing mode, flexibly decide to adopt an adaptive recursive feature elimination and generative adversarial network feature enhancement mode or an adaptive time-frequency analysis and variational autoencoder feature enhancement mode to improve the accuracy and reliability of aerodynamic parameter identification; format, segment and train the processed data with a neural network, predict aerodynamic parameters, and improve aircraft performance and safety.

Description

Technical Field

[0001] The present invention relates to the field of aircraft aerodynamic parameter identification, and more specifically, to an aircraft aerodynamic parameter identification method and system based on a neural network. Background Art

[0002] With the rapid development of aviation technology, aerodynamic parameter identification of aircraft plays a vital role in optimizing flight performance and improving safety. Aerodynamic parameter identification methods accurately obtain aerodynamic characteristic parameters of aircraft in different states by analyzing and processing flight data, thereby providing data support for the design, control and optimization of aircraft. In practical applications, flight data usually contains rich dynamic information and complex nonlinear characteristics, which makes aerodynamic parameter identification particularly complicated. In order to meet this challenge, scientific researchers have proposed a variety of data processing and analysis methods to improve the accuracy and reliability of aerodynamic parameter identification.

[0003] However, existing technologies have significant deficiencies in processing flight data. Traditional methods are mostly based on simple storage and analysis of time series data, and fail to fully consider the changes in flight status and the dynamic complexity of data, resulting in data confusion and error accumulation. In addition, existing methods lack flexibility in feature extraction and processing mode selection, and cannot effectively deal with data characteristics under different flight conditions, thus affecting the accuracy of aerodynamic parameter identification.

[0004] In order to solve the above problems, a technical solution is now provided. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a method and system for identifying aerodynamic parameters of an aircraft based on a neural network. By collecting and constructing a grid storage database of time-series data based on flight state segmentation, using real-time flight parameters to judge the flight state, and storing data in segments to unit grids when the state changes, efficient organization and management of data are achieved, ensuring independent analysis of data under different flight states, and avoiding the complexity and errors caused by data mixing; phase space reconstruction is performed on the data in each grid unit, and the nonlinear complexity index is calculated to reveal the complex dynamic behavior and nonlinear characteristics of the flight data, ensuring accurate evaluation of highly dynamic and complex data; energy spectral density is calculated through time-frequency analysis and a recursive graph is constructed, and the recursive rate and diagonal length are calculated using cosine similarity. degree distribution, calculate the recursive graph complexity index, and comprehensively evaluate the dynamic complexity of the data, which helps to select the most suitable processing mode under different dynamic conditions; by comprehensively processing the nonlinear feature complexity index and the recursive graph complexity index, the dynamic complexity evaluation coefficient is obtained, and the adaptive recursive feature elimination and generative adversarial network feature enhancement mode or the adaptive time-frequency analysis and variational autoencoder feature enhancement mode are flexibly decided to achieve efficient feature enhancement and optimization processing of the data, significantly improving the accuracy and reliability of aerodynamic parameter identification; for the unit grid data processed as above, the data is formatted and segmented, and the neural network model is trained to predict and analyze aerodynamic parameters, providing an efficient and accurate aerodynamic parameter identification method, and improving the performance and safety of the aircraft under different flight conditions. By flexibly selecting the processing mode that adapts to different feature data, the present invention effectively responds to complex flight data, optimizes the aerodynamic parameter identification process, and improves the overall performance and safety of the aircraft to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: The present invention discloses a method for identifying aerodynamic parameters of an aircraft based on a neural network, comprising the following steps:

[0007] S1, collects and builds a grid storage database for time series data based on flight status segmentation, determines the flight status through real-time flight parameters, and stores data in segments to unit grids when the status changes;

[0008] S2, reconstruct the phase space of the preprocessed data in each grid cell, and calculate the nonlinear complexity index based on the Lyapunov index and Shannon entropy to evaluate the nonlinear complexity of the data;

[0009] S3, perform time-frequency analysis on the preprocessed data in each grid cell, calculate the energy spectral density and construct a recurrence graph, calculate the recurrence rate and diagonal length distribution through cosine similarity, and obtain the recurrence graph complexity index to evaluate the dynamic complexity of the data and assist in selecting the processing mode;

[0010] S4, the nonlinear feature complexity index and the recursive graph complexity index are processed comprehensively to obtain the dynamic complexity evaluation coefficient, and the decision is made to adopt the adaptive recursive feature elimination and generative adversarial network feature enhancement mode or the adaptive time-frequency analysis and variational autoencoder feature enhancement mode;

[0011] S5, formatting and segmenting the unit grid data after being processed in step S4, and training the neural network model to predict and analyze the aerodynamic parameters of the new data.

[0012] In a preferred embodiment, step S1 includes the following contents:

[0013] When collecting flight data and building a database, the data is stored in segments according to the flight status; different flight states of the aircraft are defined, and the current flight state is determined in real time based on the flight parameters; in the time series data, whenever the flight state changes, the data within the current flight state time period is stored in a grid unit, and then the state of the subsequent flight time is continued to be analyzed, and the data within the next flight state time period is stored in the next grid unit; and so on, a data grid storage form based on continuous identical flight states is constructed to facilitate subsequent data analysis and feature extraction.

[0014] In a preferred embodiment, the logic for obtaining the nonlinear complexity index is as follows:

[0015] A1, for the data in each network unit, the delayed embedding method is used to reconstruct the phase space, using a fixed delay time and embedding dimension to map the time series data into a high-dimensional space;

[0016] A2, calculate the divergence rate of adjacent points and obtain the Lyapunov index;

[0017] A3, calculate the probability distribution of the reconstructed phase space vector, and calculate the Shannon entropy based on this probability distribution;

[0018] A4, calculate the entropy values ​​of the Lyapunov index and the Shannon entropy, then determine their respective weights according to their entropy values, and then use their respective weights to perform weighted summation on the entropy values ​​of the Lyapunov index and the Shannon entropy to obtain the nonlinear feature complexity index.

[0019] In a preferred embodiment, the logic for obtaining the recursive graph complexity index is as follows:

[0020] B1, preprocessing the raw data in each grid cell;

[0021] B2, use short-time Fourier transform to perform time-frequency analysis on the preprocessed time series data, obtain the time-frequency domain representation, and calculate the energy spectrum density at each time window and frequency to reveal the changing characteristics of the data at different frequencies and times;

[0022] B3, by calculating the similarity between the energy spectral density in the time-frequency domain, constructing a time-frequency recurrence diagram to display the recurrence pattern of the data in the time-frequency domain, including using cosine similarity to calculate the similarity between different time windows, and selecting a similarity threshold according to data characteristics to construct a recurrence diagram;

[0023] B4, calculate the recursion rate, that is, the density of recurring points in the recursion graph, by counting the number of points with a value of 1 in the recursion graph and calculating the recursion rate formula;

[0024] B5, calculate the distribution of diagonal lengths by counting the frequency of diagonal lengths in the recurrence graph and calculating the certainty;

[0025] B6, combining the recursion rate and diagonal length distribution, calculates the complexity index of the recursion graph, which is the product of the recursion rate and the determinism.

[0026] In a preferred embodiment, step S4 includes the following contents:

[0027] The dynamic complexity evaluation coefficient is calculated by weighted summation of the nonlinear feature complexity index and the recursive graph complexity index; the dynamic complexity evaluation coefficient is compared with the degree threshold. If the dynamic complexity evaluation coefficient is greater than or equal to the degree threshold, the adaptive recursive feature elimination and generative adversarial network feature enhancement mode is used to process the data; otherwise, if the dynamic complexity evaluation coefficient is less than the degree threshold, the adaptive time-frequency analysis and variational autoencoder feature enhancement mode is used to process the data.

[0028] The present invention also discloses an aircraft aerodynamic parameter identification system based on a neural network, comprising: a data collection module, a feature reconstruction module, a time-frequency analysis module, a mode selection module, a feature enhancement module, a model training module, and a parameter prediction module;

[0029] The data collection module collects and constructs a grid storage database of time series data based on flight status segmentation, determines the flight status through real-time flight parameters, stores data in segments to unit grids when the status changes, and outputs the segmented grid data to the feature reconstruction module and the time-frequency analysis module;

[0030] The feature reconstruction module reconstructs the phase space of the preprocessed data in each grid cell, calculates the nonlinear complexity index, evaluates the nonlinear complexity of the data based on the Lyapunov index and Shannon entropy, and outputs the nonlinear complexity index to the mode selection module;

[0031] The time-frequency analysis module performs time-frequency analysis on the preprocessed data in each grid cell, calculates the energy spectrum density and constructs a recurrence graph, calculates the recurrence rate and diagonal length distribution through cosine similarity, calculates the recurrence graph complexity index to evaluate the dynamic complexity of the data, and outputs the dynamic complexity index to the mode selection module;

[0032] The mode selection module comprehensively processes the nonlinear feature complexity index and the recursive graph complexity index to obtain the dynamic complexity evaluation coefficient, and decides to use the adaptive recursive feature elimination and generative adversarial network feature enhancement mode or the adaptive time-frequency analysis and variational autoencoder feature enhancement mode, and outputs the dynamic complexity evaluation coefficient to the feature enhancement module;

[0033] The feature enhancement module applies the adaptive recursive feature elimination and generative adversarial network feature enhancement mode or the adaptive time-frequency analysis and variational autoencoder feature enhancement mode according to the result of the mode selection, enhances and optimizes the grid data, and outputs the enhanced grid data to the model training module;

[0034] The model training module formats and segments the unit grid data after feature enhancement, trains the neural network model, and outputs the trained model to the parameter prediction module;

[0035] The parameter prediction module uses the trained neural network model to predict and analyze aerodynamic parameters of new data.

[0036] The technical effects and advantages of the neural network-based aircraft aerodynamic parameter identification method and system provided by the present invention are as follows:

[0037] The present invention collects and constructs a grid storage database of time-series data based on flight status segmentation, uses real-time flight parameters to judge the flight status, and stores data in unit grids in segments when the status changes, thereby realizing efficient organization and management of data, ensuring independent analysis of data under different flight states, and avoiding complexity and errors caused by data mixing; reconstructing the phase space of the data in each grid unit, calculating the nonlinear complexity index, revealing the complex dynamic behavior and nonlinear characteristics of the flight data, and ensuring accurate evaluation of highly dynamic and complex data; calculating the energy spectral density through time-frequency analysis and constructing a recursive graph, calculating the recursive rate and diagonal length distribution using cosine similarity, calculating the recursive graph complexity index, and comprehensively evaluating the data Dynamic complexity helps to select the most suitable processing mode under different dynamic conditions; by comprehensively processing the nonlinear feature complexity index and the recursive graph complexity index, the dynamic complexity evaluation coefficient is obtained, and the adaptive recursive feature elimination and generative adversarial network feature enhancement mode or the adaptive time-frequency analysis and variational autoencoder feature enhancement mode are flexibly decided to achieve efficient feature enhancement and optimization processing of data, significantly improving the accuracy and reliability of aerodynamic parameter identification; for the unit grid data processed as above, the data is formatted and segmented, and the neural network model is trained to predict and analyze aerodynamic parameters, providing an efficient and accurate aerodynamic parameter identification method, improving the performance and safety of aircraft under different flight conditions. By flexibly selecting the processing mode that adapts to different feature data, the present invention effectively responds to complex flight data, optimizes the aerodynamic parameter identification process, and improves the overall performance and safety of the aircraft. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 A schematic diagram of a flow chart of a method for identifying aerodynamic parameters of an aircraft based on a neural network according to the present invention;

[0039] Figure 2 The figure is a schematic diagram of the structure of an aircraft aerodynamic parameter identification system based on a neural network according to the present invention. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] Embodiment 1: Figure 1 The present invention provides a method for identifying aerodynamic parameters of an aircraft based on a neural network, comprising:

[0042] S1, collects and builds a grid storage database for time series data based on flight status segmentation, determines the flight status through real-time flight parameters, and stores data in segments to unit grids when the status changes;

[0043] S2, reconstruct the phase space of the preprocessed data in each grid cell, and calculate the nonlinear complexity index based on the Lyapunov index and Shannon entropy to evaluate the nonlinear complexity of the data;

[0044] S3, perform time-frequency analysis on the preprocessed data in each grid cell, calculate the energy spectral density and construct a recurrence graph, calculate the recurrence rate and diagonal length distribution through cosine similarity, and obtain the recurrence graph complexity index to evaluate the dynamic complexity of the data and assist in selecting the processing mode;

[0045] S4, the nonlinear feature complexity index and the recursive graph complexity index are processed comprehensively to obtain the dynamic complexity evaluation coefficient, and the decision is made to adopt the adaptive recursive feature elimination and generative adversarial network feature enhancement mode or the adaptive time-frequency analysis and variational autoencoder feature enhancement mode;

[0046] S5, formatting and segmenting the unit grid data after being processed in step S4, and training the neural network model to predict and analyze the aerodynamic parameters of the new data.

[0047] In the process of flight data analysis and feature extraction, accurate data organization and management are crucial. The flight parameters of aircraft in different states are significantly different. The mixed storage of data in different states will lead to the complexity and accuracy of subsequent analysis. Existing methods based only on time series data storage often mix data in different flight states, which requires a lot of time and computing resources to separate and process data in each state during data analysis. This not only increases the complexity of data processing, but also easily introduces errors and reduces the accuracy of analysis. Therefore, segmented storage of data according to flight status helps to improve the efficiency and accuracy of data processing. By classifying data in the same continuous flight state into the same grid unit, not only can the consistency of data be maintained, but also a clearer basis can be provided for subsequent feature extraction and pattern recognition. In addition, this segmented storage method facilitates the rapid retrieval and analysis of data in a specific flight state, supports more detailed and targeted research, and improves the overall efficiency and quality of flight data analysis. Therefore, implementing segmented storage of data based on flight status is a key step in optimizing flight data processing and analysis, and provides a solid foundation for accurate aerodynamic parameter identification and flight performance optimization.

[0048] Step S1 includes the following contents:

[0049] S1.1, first, define the different flight states of the aircraft, which may include climbing, cruising, turning and descending. The definition of flight state is based on parameters such as the aircraft's altitude, speed and acceleration. For example, the climbing state can be defined as a stage in which the aircraft's altitude continues to increase, and the speed and acceleration change within a specific range; the cruising state is a stage in which the aircraft flies at a relatively constant altitude and speed. Through clear parameter standards and thresholds, the current state of the aircraft can be judged in real time, laying the foundation for segmented data storage.

[0050] S1.2, initialize the flight data storage structure and create an empty grid cell list to store the relevant data under each flight state. When traversing the time series data, segment the data according to the change of flight state. Specifically, first create an empty grid cell and use it as the current storage unit. Then, starting from the start time of the data, read the flight parameters at each time point step by step, and judge the current flight state based on these flight parameters. When a change in flight state is detected, the data of the current grid cell is stored in the database, and a new grid cell is created to store the data under the new flight state.

[0051] S1.3, the data segmentation storage process starts with initializing the grid cell list, creating an empty grid cell list, and defining the current flight state and the current grid cell. When traversing the time series data, the flight parameters at each time point are obtained, and the current flight state is determined based on these flight parameters. If the current flight state is the same as the previous flight state, the data is added to the current grid cell; if the current flight state is different from the previous flight state, the current grid cell is stored in the grid cell list, a new grid cell is created, and the current state is updated to the new flight state.

[0052] S1.4, after traversing the time series data, the last current grid cell is stored in the grid cell list. At this point, all the data of the same continuous flight state has been segmented and stored in different grid cells. Next, all the grid cells in the grid cell list are stored in the database. Each grid cell contains all the data under a flight state. This storage method can improve the efficiency of data organization and management, and facilitate subsequent data analysis and feature extraction.

[0053] When collecting flight data and building a database, the data is stored in segments according to the flight status; different flight states of the aircraft are defined, and the current flight state is determined in real time based on the flight parameters; in the time series data, whenever the flight state changes, the data within the current flight state time period is stored in a grid unit, and then the state of the subsequent flight time is continued to be analyzed, and the data within the next flight state time period is stored in the next grid unit; and so on, a data grid storage form based on continuous identical flight states is constructed to facilitate subsequent data analysis and feature extraction.

[0054] In the process of aircraft aerodynamic parameter identification, accurate data organization and management are crucial. The flight parameters of an aircraft in different states are significantly different. Therefore, segmenting the data according to the flight state and analyzing the data in each grid unit separately can significantly improve the accuracy and effectiveness of data processing.

[0055] The traditional method based on time series data storage stores all data together in chronological order without considering the differences in flight status. This method has multiple shortcomings. First, data of different flight status are mixed together, which increases the difficulty of data processing and leads to poor results in feature extraction and pattern recognition. Second, when performing aerodynamic parameter identification, the noise and interference in the mixed data will reduce the accuracy and reliability of identification. Finally, this storage method is not convenient for rapid retrieval and analysis of data under specific flight status, which limits the in-depth study and optimization of aircraft performance.

[0056] By storing the data in segments according to the flight status and analyzing and evaluating the data of each grid unit separately, the data characteristics under each flight status can be captured more accurately. Specifically, by evaluating the nonlinear complexity and dynamic characteristics of the data of each grid unit separately, the complexity and change trend of the data can be accurately identified, and the most appropriate preprocessing mode can be selected, thereby improving the efficiency and accuracy of data processing.

[0057] This approach has many benefits: first, it avoids the interference caused by the mixing of data from different flight states, ensuring that the data for each flight state is highly coherent and consistent; second, by accurately evaluating the data characteristics of each grid cell, it can select the most suitable preprocessing mode in a targeted manner to further improve data quality and analysis accuracy; finally, this method facilitates rapid retrieval and analysis of data under specific flight states, which helps to conduct in-depth research on aircraft performance and optimization strategies.

[0058] Step S2 includes the following contents:

[0059] When analyzing each grid cell separately, it is necessary to calculate and analyze the complexity of nonlinear features, because flight data has significant nonlinear characteristics under different flight states, which are crucial for accurate aerodynamic parameter identification. By calculating the complexity of nonlinear features, complex dynamic behaviors and potential patterns in the data can be captured, which helps to identify and understand the performance and behavior of the aircraft under specific states. Existing methods based on time series data storage and analysis cannot effectively separate and process these complex nonlinear features, resulting in reduced accuracy and reliability when performing aerodynamic parameter identification. By analyzing the data in each grid cell separately and evaluating the complexity of nonlinear features, it can ensure that the data of each flight state is preprocessed and analyzed in the most appropriate way, improving the accuracy and effectiveness of data processing, thereby providing high-quality data support for subsequent aerodynamic parameter identification. This method not only avoids the interference caused by the mixing of data from different flight states, but also can more accurately select and apply the most appropriate preprocessing mode, further enhancing the research depth of aircraft performance optimization and safety.

[0060] The logic for obtaining the nonlinear complexity index is as follows:

[0061] A1, for the data in each network unit, use the delayed embedding method to reconstruct the phase space. Select a fixed delay time and embedding dimension to map the time series data into a high-dimensional space;

[0062] ;

[0063] in, Indicates at time The purpose is to embed the one-dimensional time series data into a high-dimensional space in order to better capture its dynamic characteristics and complex behaviors. It is a multidimensional vector containing data of the current time point and several time points before and after it;

[0064] Indicates at different time points The data value is the original one-dimensional time series data;

[0065] is the delay time, which is used to reconstruct the time interval of the phase space. It is used to delay the original time series data for a period of time before sampling, which helps to capture the dynamic characteristics of the time series;

[0066] is the embedding dimension, the dimension of the reconstructed phase space, which indicates how many sampling points with delayed time are used when reconstructing the phase space.

[0067] Through the delayed embedding method, the one-dimensional time series data is converted into a multi-dimensional vector. The purpose of this process is to embed the time series data into a high-dimensional space in order to reveal the dynamic characteristics and potential patterns of the data.

[0068] Phase space reconstruction is the basis for subsequent steps by converting one-dimensional data into multi-dimensional data, providing the necessary input for the subsequent Lyapunov exponent and Shannon entropy calculations.

[0069] A2, calculates the divergence rate of adjacent points and obtains the Lyapunov index to measure the nonlinear dynamic behavior of the system.

[0070] ;

[0071] in, Represents the Lyapunov index, which is used to measure the complexity of data; is the total number of data points; It is Data points at time distance; is the time interval used to calculate the divergence rate between adjacent points.

[0072] The calculation of Lyapunov exponents relies on the vectors reconstructed in phase space. By analyzing the divergence of these vectors in time, the chaos and nonlinear characteristics of the system can be quantified.

[0073] A3, calculate the probability distribution of the reconstructed phase space vector , and calculate the Shannon entropy based on this probability distribution . Shannon entropy is used to measure the complexity of data.

[0074] ;

[0075] The calculation of Shannon entropy also relies on the vectors reconstructed in phase space. By analyzing the probability distribution of these vectors, the complexity of the data can be evaluated.

[0076] A4, calculate the entropy values ​​of the Lyapunov index and the Shannon entropy, then determine their respective weights according to their entropy values, and then use their respective weights to perform weighted summation on the entropy values ​​of the Lyapunov index and the Shannon entropy to obtain the nonlinear feature complexity index.

[0077] The nonlinear characteristic complexity index is used to represent and reflect the complex dynamic behavior and data complexity of flight data under different flight states. A larger nonlinear characteristic complexity index indicates that the flight data has more complex dynamic characteristics and higher uncertainty in this state, and may involve more nonlinear relationships and chaotic behaviors, which means that more complex and sophisticated analysis and processing methods are required in the aerodynamic parameter identification process. On the contrary, a smaller nonlinear characteristic complexity index indicates that the dynamic behavior of the flight data is relatively simple, the data complexity is low, and there are fewer nonlinear relationships and uncertainties, which is suitable for a relatively simple preprocessing mode. By evaluating and comparing the nonlinear characteristic complexity index within different grid cells, it can help select the most appropriate preprocessing mode, thereby improving the accuracy and reliability of aerodynamic parameter identification, and ultimately optimizing the performance and safety of the aircraft.

[0078] When analyzing each grid cell separately, it is necessary to calculate and analyze the recursive graph complexity index, because the flight data has significant time-frequency dynamic characteristics under different flight states, which are crucial for accurate aerodynamic parameter identification. By calculating the recursive graph complexity index, the recurrence pattern and dynamic complexity of the data in the time-frequency domain can be captured, which helps to identify and understand the performance and behavior of the aircraft in a specific state. The existing methods based on time series data storage and analysis cannot effectively separate and process these complex time-frequency dynamic characteristics, resulting in reduced accuracy and reliability when performing aerodynamic parameter identification. By analyzing the data in each grid cell separately and evaluating the recursive graph complexity index, it can ensure that the data of each flight state is preprocessed and analyzed in the most suitable way, improving the accuracy and effectiveness of data processing, thereby providing high-quality data support for subsequent aerodynamic parameter identification. This method not only avoids the interference caused by the mixing of data from different flight states, but also can more accurately select and apply the most suitable preprocessing mode, further enhancing the research depth of aircraft performance optimization and safety.

[0079] Step S3 includes the following contents:

[0080] The logic for obtaining the recursive graph complexity index is as follows:

[0081] B1, preprocess the raw data within each grid cell to eliminate noise and unnecessary fluctuations and make the data smoother and more stable.

[0082] A low-pass filter was used to remove high-frequency noise from the data to ensure that the main features of the data were not disturbed by noise in subsequent analysis.

[0083] The data is standardized so that its mean is 0 and its variance is 1, thereby eliminating the influence of dimension and facilitating the comparison and processing of different features.

[0084] B2, use short-time Fourier transform (STFT) to perform time-frequency analysis on the preprocessed time series data, obtain the time-frequency domain representation, and calculate the energy spectrum density in each time window and frequency, revealing the changing characteristics of the data at different frequencies and times, and providing richer information for recursive graph construction.

[0085] Perform short-time Fourier transform on the time series data of each grid cell to obtain the time-frequency domain representation.

[0086] For each time window and frequency, the energy spectral density is calculated.

[0087] B3, by calculating the similarity between the energy spectral density in the time-frequency domain, construct a time-frequency recurrence diagram to display the recurrence pattern of the data in the time-frequency domain. The specific method includes using cosine similarity to calculate the similarity between different time windows, and selecting the similarity threshold according to the data characteristics to construct the recurrence diagram.

[0088] Use cosine similarity to calculate the similarity of energy spectral density between different time windows. The formula is: ;

[0089] in, Indicates time and The similarities between and Respectively represent the signal at time and and the frequencies are The energy distribution on .

[0090] Select the similarity threshold according to the data characteristics and construct the recursive graph. The formula is: ;

[0091] in, Indicates time and Are they similar? If they are similar, it is 1, otherwise it is 0. is the Heaviside step function.

[0092] B4, calculate the recurrence rate (RR), that is, the density of recurring points in the recurrence graph, which reflects the recurrence frequency of the system state. This is achieved by counting the number of points with a value of 1 in the recurrence graph and calculating the recurrence rate formula;

[0093] B5, calculate the diagonal length distribution (DET), which reflects the duration and certainty of the system state, by counting the frequency of the diagonal length in the recursive graph and calculating the certainty: ;

[0094] in, Indicates the length of the diagonal line; The length of the diagonal is frequency.

[0095] B6, integrates the recursion rate and diagonal length distribution, calculates the recursion graph complexity index (RPC), which is the product of the recursion rate and the certainty, comprehensively evaluates the dynamic complexity of the data, and provides a reliable evaluation index for mode selection, thereby optimizing aerodynamic parameter identification and aircraft performance;

[0096] The recursive graph complexity index is used to represent and reflect the recurrence pattern and dynamic complexity of flight data in the time-frequency domain. A larger recursive graph complexity index indicates that the data has a higher recurrence frequency and certainty, reflecting that the dynamic behavior of the data is more complex and has more long-term dependencies; on the contrary, a smaller recursive graph complexity index indicates that the data has a lower recurrence frequency and certainty, reflecting that the dynamic behavior of the data is relatively simple and has strong short-term dependencies. In this scenario, the recursive graph complexity index can assist in selecting the most appropriate processing mode: when the recursive graph complexity index is large, it means that the flight data is complex and has high dynamic characteristics. At this time, it is suitable to adopt mode 1, that is, adaptive recursive feature elimination and generative adversarial network (GAN) feature enhancement mode, to cope with complex nonlinear relationships and dynamic characteristics; when the recursive graph complexity index is small, it means that the flight data is relatively simple and the dynamic characteristics are not strong. At this time, it is suitable to adopt mode 2, that is, adaptive time-frequency analysis and variational autoencoder (VAE) feature enhancement mode, to more efficiently process simpler data features, thereby optimizing aerodynamic parameter identification and aircraft performance.

[0097] Step S4 includes the following contents:

[0098] The nonlinear feature complexity index and the recursive graph complexity index are weighted and summed to obtain the dynamic complexity evaluation coefficient. For example, the following calculation formula can be used: ;

[0099] in, Represents the dynamic complexity evaluation coefficient; and They are the nonlinear feature complexity index and the recursive graph complexity index respectively; and They are the preset proportional coefficients of the nonlinear feature complexity index and the recursive graph complexity index, respectively, and both are greater than zero.

[0100] The dynamic complexity evaluation coefficient is used to represent and reflect the comprehensive dynamic characteristics and complexity of flight data, combining the nonlinear feature complexity index and the recursive graph complexity index. A larger dynamic complexity evaluation coefficient indicates that the flight data has higher nonlinear dynamic behavior and complex recurrence pattern, reflecting that the data has strong complexity and dynamic characteristics, and is suitable for processing complex data features using adaptive recursive feature elimination and generative adversarial network (GAN) feature enhancement mode; on the contrary, a smaller dynamic complexity evaluation coefficient indicates that the nonlinear dynamic behavior and recurrence pattern of flight data are relatively simple, and the dynamic characteristics are relatively low, and is suitable for using adaptive time-frequency analysis and variational autoencoder (VAE) feature enhancement mode to more efficiently process simpler data features, thereby optimizing aerodynamic parameter identification and aircraft performance.

[0101] Compare the dynamic complexity evaluation coefficient with the degree threshold. If the dynamic complexity evaluation coefficient is greater than or equal to the degree threshold, it means that the flight data in the unit grid has high nonlinear characteristics and complex dynamic behavior, the data has high recurrence frequency and certainty, and the dynamic characteristics of the system are complex and have long-term dependencies. In this case, it is suitable to use adaptive recursive feature elimination and generative adversarial network (GAN) feature enhancement mode to process the data to cope with complex nonlinear relationships and dynamic characteristics, and ensure the accuracy and reliability of aerodynamic parameter identification.

[0102] On the contrary, if the dynamic complexity evaluation coefficient is less than the degree threshold, it means that the nonlinear characteristics and dynamic complexity of the flight data in the unit grid are low, the data recurrence frequency and certainty are weak, the dynamic behavior of the system is relatively simple and the short-term dependence is strong. In this case, it is suitable to use adaptive time-frequency analysis and variational autoencoder (VAE) feature enhancement mode to process the data in order to more efficiently extract features and optimize aerodynamic parameter identification, thereby improving the performance and stability of the aircraft.

[0103] The Adaptive Recursive Feature Elimination and Generative Adversarial Network (GAN) feature enhancement mode is a technology that recursively eliminates redundant and irrelevant features through an adaptive method, and uses a generative adversarial network (GAN) to enhance and generate more representative feature data. In the aircraft aerodynamic parameter identification method, this mode can effectively process complex and highly dynamic data, reduce data redundancy and improve data purity through recursive feature elimination, and then use GAN to enhance features and generate more high-quality data, thereby improving the accuracy and reliability of aerodynamic parameter identification, especially for data with high nonlinear characteristics and complex dynamic behavior.

[0104] The adaptive time-frequency analysis and variational autoencoder (VAE) feature enhancement mode is a feature enhancement method that combines time-frequency analysis technology with variational autoencoder (VAE). Time-frequency analysis is used to decompose and analyze the characteristics of data at different times and frequencies, while VAE is used to extract and enhance the nonlinear characteristics of the data. In the aircraft aerodynamic parameter identification method, this mode can efficiently process relatively simple and low-dynamic data, capture the main features of the data through time-frequency analysis, and use VAE to extract and enhance features, thereby improving the efficiency and stability of aerodynamic parameter identification. It is suitable for data with weak dynamic characteristics and low nonlinear characteristics.

[0105] These two modes have a significant impact on the processing process in the aerodynamic parameter identification method of the aircraft. The adaptive recursive feature elimination and GAN feature enhancement mode improves the processing capability of highly dynamic and complex data through complex data preprocessing and feature enhancement, and enhances the accuracy and robustness of aerodynamic parameter identification. The adaptive time-frequency analysis and VAE feature enhancement mode improves the efficiency and stability of processing simple data through efficient data decomposition and feature extraction, and optimizes the process and results of aerodynamic parameter identification. The selection of these two modes is based on the size of the dynamic complexity evaluation coefficient to adapt to the needs of different data characteristics, thereby optimizing the performance and safety of the aircraft.

[0106] Step S5 includes the following contents:

[0107] S5.1, format the processed data into a neural network input format, generate label data, and divide the data set into a training set, a validation set, and a test set;

[0108] The cell grid data after preprocessing and feature enhancement is organized into an input format acceptable to the neural network.

[0109] The processed data is organized into a two-dimensional array, where each row represents a sample and each column represents a feature.

[0110] Generate corresponding label data according to actual needs, usually the aerodynamic parameters of the aircraft (such as lift coefficient, drag coefficient, etc.).

[0111] The dataset is divided into training set, validation set and test set. The common ratio is 70% training set, 15% validation set and 15% test set.

[0112] S5.2, build a neural network architecture consisting of input layer, convolution layer, LSTM layer, fully connected layer and output layer to effectively extract and learn spatial and temporal features in the data.

[0113] Define the input layer, the number of input nodes is equal to the number of features.

[0114] Design multiple convolutional layers (such as Conv1D) to extract the spatial features of the data. Each convolutional layer is followed by a maximum pooling layer (MaxPooling1D) to reduce the feature dimension.

[0115] Add LSTM layers to capture the time series characteristics of the data. There can be multiple LSTM layers to increase the learning ability of the model.

[0116] A fully connected layer (Dense) is added after the LSTM layer to further extract high-level features.

[0117] Define the output layer, the number of output nodes is equal to the number of aerodynamic parameters, and use a linear activation function.

[0118] Specific architecture:

[0119] Import tensorflow as tf

[0120] From tensorflow.keras.modelsimportSequential

[0121] from tensorflow.keras.layers Import Dense,Conv1D,MaxPooling1D,LSTM,Flatten

[0122] model=Sequential([

[0123] Conv1D(filters=64,kernel_size=3,activation='relu',input_shape=(input_shape)),

[0124] MaxPooling1D(pool_size=2),

[0125] Conv1D(filters=128,kernel_size=3,activation='relu'),

[0126] MaxPooling1D(pool_size=2),

[0127] LSTM(50,return_sequences=True),

[0128] LSTM(50),

[0129] Flatten(),

[0130] Dense(50,activation='relu'),

[0131] Dense(output_shape,activation='linear') ])

[0133] S5.3, select mean square error as the loss function, use Adam optimizer, set evaluation indicators, and compile the neural network model.

[0134] The mean squared error (MSE) is selected as the loss function for regression problems.

[0135] Select Adam optimizer and set hyperparameters such as learning rate.

[0136] The mean square error (MSE) is chosen as the evaluation metric.

[0137] S5.4, use the training data to train the model, optimize the network parameters through forward propagation, error calculation and back propagation, and evaluate and adjust on the validation data to prevent overfitting.

[0138] The training set is input into the neural network, and forward propagation, error calculation and back propagation are performed to adjust the network weights and biases.

[0139] After each training cycle, the validation set is used to evaluate the model performance, monitor the loss value and evaluation indicators, and prevent overfitting.

[0140] According to the performance of the validation set, adjust the model's hyperparameters (such as learning rate, batch size, etc.) to optimize the model performance.

[0141] S5.5, use the test data to evaluate the generalization ability of the model, verify the model performance by calculating the loss value and evaluation indicators, and fine-tune the model based on the results.

[0142] S5.6, preprocess and feature enhance the new data, input it into the trained model for aerodynamic parameter prediction, and analyze the prediction results to evaluate the actual application effect.

[0143] The new data is subjected to the same preprocessing and feature augmentation as the training data.

[0144] The preprocessed new data is input into the trained neural network model and the predicted aerodynamic parameters are output.

[0145] Analyze the prediction results, evaluate the actual application effect of the model, and make corrections and optimizations as needed.

[0146] The present invention collects and constructs a grid storage database of time-series data based on flight status segmentation, uses real-time flight parameters to judge the flight status, and stores data in unit grids in segments when the status changes, thereby realizing efficient organization and management of data, ensuring independent analysis of data under different flight states, and avoiding complexity and errors caused by data mixing; reconstructing the phase space of the data in each grid unit, calculating the nonlinear complexity index, revealing the complex dynamic behavior and nonlinear characteristics of the flight data, and ensuring accurate evaluation of highly dynamic and complex data; calculating the energy spectral density through time-frequency analysis and constructing a recursive graph, calculating the recursive rate and diagonal length distribution using cosine similarity, calculating the recursive graph complexity index, and comprehensively evaluating the data Dynamic complexity helps to select the most suitable processing mode under different dynamic conditions; by comprehensively processing the nonlinear feature complexity index and the recursive graph complexity index, the dynamic complexity evaluation coefficient is obtained, and the adaptive recursive feature elimination and generative adversarial network feature enhancement mode or the adaptive time-frequency analysis and variational autoencoder feature enhancement mode are flexibly decided to achieve efficient feature enhancement and optimization processing of data, significantly improving the accuracy and reliability of aerodynamic parameter identification; for the unit grid data processed as above, the data is formatted and segmented, and the neural network model is trained to predict and analyze aerodynamic parameters, providing an efficient and accurate aerodynamic parameter identification method, improving the performance and safety of aircraft under different flight conditions. By flexibly selecting the processing mode that adapts to different feature data, the present invention effectively responds to complex flight data, optimizes the aerodynamic parameter identification process, and improves the overall performance and safety of the aircraft.

[0147] Embodiment 2: Figure 2 The invention provides an aircraft aerodynamic parameter identification system based on a neural network, comprising: a data collection module, a feature reconstruction module, a time-frequency analysis module, a mode selection module, a feature enhancement module, a model training module, and a parameter prediction module;

[0148] The data collection module collects and constructs a grid storage database of time series data based on flight status segmentation, determines the flight status through real-time flight parameters, stores data in segments to unit grids when the status changes, and outputs the segmented grid data to the feature reconstruction module and the time-frequency analysis module;

[0149] The feature reconstruction module reconstructs the phase space of the preprocessed data in each grid cell, calculates the nonlinear complexity index, evaluates the nonlinear complexity of the data based on the Lyapunov index and Shannon entropy, and outputs the nonlinear complexity index to the mode selection module;

[0150] The time-frequency analysis module performs time-frequency analysis on the preprocessed data in each grid cell, calculates the energy spectrum density and constructs a recurrence graph, calculates the recurrence rate and diagonal length distribution through cosine similarity, calculates the recurrence graph complexity index to evaluate the dynamic complexity of the data, and outputs the dynamic complexity index to the mode selection module;

[0151] The mode selection module comprehensively processes the nonlinear feature complexity index and the recursive graph complexity index to obtain the dynamic complexity evaluation coefficient, and decides to use the adaptive recursive feature elimination and generative adversarial network feature enhancement mode or the adaptive time-frequency analysis and variational autoencoder feature enhancement mode, and outputs the dynamic complexity evaluation coefficient to the feature enhancement module;

[0152] The feature enhancement module applies the adaptive recursive feature elimination and generative adversarial network feature enhancement mode or the adaptive time-frequency analysis and variational autoencoder feature enhancement mode according to the result of the mode selection, enhances and optimizes the grid data, and outputs the enhanced grid data to the model training module;

[0153] The model training module formats and segments the unit grid data after feature enhancement, trains the neural network model, and outputs the trained model to the parameter prediction module;

[0154] The parameter prediction module uses the trained neural network model to predict and analyze aerodynamic parameters of new data.

[0155] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0156] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0157] It should be noted that, in this article, if there are relational terms such as first and second, etc., they are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or still includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "including a..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0158] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for identifying aerodynamic parameters of aircraft based on neural network. Features: S1, collects and builds a grid storage database for time series data based on flight status segmentation, determines the flight status through real-time flight parameters, and stores data in segments to unit grids when the status changes; S2, reconstruct the phase space of the preprocessed data in each grid cell, and calculate the nonlinear complexity index based on the Lyapunov index and Shannon entropy to evaluate the nonlinear complexity of the data; S3, perform time-frequency analysis on the preprocessed data in each grid cell, calculate the energy spectral density and construct a recurrence graph, calculate the recurrence rate and diagonal length distribution through cosine similarity, and obtain the recurrence graph complexity index to evaluate the dynamic complexity of the data and assist in selecting the processing mode; S4, the nonlinear complexity index and the recursive graph complexity index are processed comprehensively to obtain the dynamic complexity evaluation coefficient, and the decision is made to adopt the adaptive recursive feature elimination and generative adversarial network feature enhancement mode or the adaptive time-frequency analysis and variational autoencoder feature enhancement mode; S5, formatting and segmenting the unit grid data after being processed in step S4, and training a neural network model to predict and analyze aerodynamic parameters of the new data; Step S1 includes the following contents: When collecting flight data and building a database, data is stored in segments according to the flight status; different flight statuses of the aircraft are defined, and the current flight status is determined in real time based on the flight parameters; in time series data, whenever the flight status changes, the data in the current flight status time period is stored in a grid unit, and then the status of the subsequent flight time is analyzed, and the data in the next flight status time period is stored in the next grid unit; and so on, a data grid storage form based on continuous identical flight status is constructed to facilitate subsequent data analysis and feature extraction; The logic for obtaining the nonlinear complexity index is as follows: A1, for the data in each network unit, the delayed embedding method is used to reconstruct the phase space, using a fixed delay time and embedding dimension to map the time series data into a high-dimensional space; A2, calculate the divergence rate of adjacent points and obtain the Lyapunov index; A3, calculate the probability distribution of the reconstructed phase space vector, and calculate the Shannon entropy based on this probability distribution; A4, calculate the entropy values ​​of the Lyapunov index and the Shannon entropy, then determine their respective weights based on their entropy values, and then use their respective weights to perform weighted summation on the entropy values ​​of the Lyapunov index and the Shannon entropy to obtain the nonlinear complexity index.

2. The method for identifying aerodynamic parameters of an aircraft based on a neural network according to claim 1, Features: The logic for obtaining the recursive graph complexity index is as follows: B1, preprocessing the raw data in each grid cell; B2, use short-time Fourier transform to perform time-frequency analysis on the preprocessed time series data, obtain the time-frequency domain representation, and calculate the energy spectrum density at each time window and frequency to reveal the changing characteristics of the data at different frequencies and times; B3, by calculating the similarity between the energy spectral density in the time-frequency domain, constructing a time-frequency recurrence diagram to display the recurrence pattern of the data in the time-frequency domain, including using cosine similarity to calculate the similarity between different time windows, and selecting a similarity threshold according to data characteristics to construct a recurrence diagram; B4, calculate the recursion rate, that is, the density of recurring points in the recursion graph, by counting the number of points with a value of 1 in the recursion graph and calculating the recursion rate formula; B5, calculate the distribution of diagonal lengths by counting the frequency of diagonal lengths in the recurrence graph and calculating the certainty; B6, combining the recursion rate and diagonal length distribution, calculates the complexity index of the recursion graph, which is the product of the recursion rate and the determinism.

3. The method for identifying aerodynamic parameters of an aircraft based on a neural network according to claim 2, Features: Step S4 includes the following contents: The nonlinear complexity index and the recursive graph complexity index are weighted and summed to obtain the dynamic complexity evaluation coefficient; The dynamic complexity evaluation coefficient is compared with the degree threshold. If the dynamic complexity evaluation coefficient is greater than or equal to the degree threshold, the adaptive recursive feature elimination and generative adversarial network feature enhancement mode are used to process the data; otherwise, if the dynamic complexity evaluation coefficient is less than the degree threshold, the adaptive time-frequency analysis and variational autoencoder feature enhancement mode are used to process the data.

4. An aircraft aerodynamic parameter identification system based on a neural network, used to implement the aircraft aerodynamic parameter identification method according to any one of claims 1 to 3, It is characterized in that include: Data collection module, feature reconstruction module, time-frequency analysis module, mode selection module, feature enhancement module, model training module, parameter prediction module; The data collection module collects and constructs a grid storage database of time series data based on flight status segmentation, determines the flight status through real-time flight parameters, stores data in segments to unit grids when the status changes, and outputs the segmented grid data to the feature reconstruction module and the time-frequency analysis module; The feature reconstruction module reconstructs the phase space of the preprocessed data in each grid cell, calculates the nonlinear complexity index, evaluates the nonlinear complexity of the data based on the Lyapunov index and Shannon entropy, and outputs the nonlinear complexity index to the mode selection module; The time-frequency analysis module performs time-frequency analysis on the preprocessed data in each grid cell, calculates the energy spectrum density and constructs a recurrence graph, calculates the recurrence rate and diagonal length distribution through cosine similarity, calculates the recurrence graph complexity index to evaluate the dynamic complexity of the data, and outputs the dynamic complexity index to the mode selection module; The mode selection module comprehensively processes the nonlinear complexity index and the recursive graph complexity index to obtain the dynamic complexity evaluation coefficient, and decides to adopt the adaptive recursive feature elimination and generative adversarial network feature enhancement mode or the adaptive time-frequency analysis and variational autoencoder feature enhancement mode, and outputs the dynamic complexity evaluation coefficient to the feature enhancement module; The feature enhancement module applies the adaptive recursive feature elimination and generative adversarial network feature enhancement mode or the adaptive time-frequency analysis and variational autoencoder feature enhancement mode according to the result of the mode selection, enhances and optimizes the grid data, and outputs the enhanced grid data to the model training module; The model training module formats and segments the unit grid data after feature enhancement, trains the neural network model, and outputs the trained model to the parameter prediction module; The parameter prediction module uses the trained neural network model to predict and analyze aerodynamic parameters of new data.

Images