Vertical tail bending moment prediction method based on maneuvering data classification
Through the vertical tail bending moment prediction method based on maneuver data classification, a maneuver fly parametric-vertical tail load model is constructed, which solves the problem of low load prediction accuracy in the existing technology, and achieves higher prediction accuracy and reliability.
Patent Information
- Application Number
- CN202510109017.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-23
- Publication Date
- 2025-05-02
AI Technical Summary
The prior art is difficult to effectively predict the vertical tail bending moment, especially in different flight states, resulting in low load prediction accuracy.
The vertical tail moment prediction method based on maneuver data classification is adopted, and the maneuver action data is classified by obtaining flight parameters and vertical tail moments, and a maneuver action data is constructed to improve the prediction accuracy.
The prediction accuracy of the vertical tail bending moment is significantly improved, and the impact of different flight states on load prediction is taken into account, which enhances the reliability of the prediction.
Smart Images
Figure CN119918418A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of aircraft structure load prediction, and in particular relates to a vertical tail bending moment prediction method based on maneuvering data classification. Background Art
[0002] Aircraft structures are constantly subjected to complex alternating loads during their service, which can cause fatigue damage to the structure and lead to fatigue cracks and even fatigue failure. As the flight time increases, this fatigue accumulation effect poses a serious threat to the structural integrity of the aircraft. In order to ensure the reliability, availability and safe fatigue life of the aircraft structure, the monitoring of aircraft structural loads becomes particularly important. Structural load monitoring is a key technology for life monitoring and structural health management of in-service aircraft. By monitoring and analyzing the load data on the aircraft structure, potential fatigue damage can be detected in a timely manner, and the health status and fatigue life of the structure can be evaluated. This not only helps to take preventive maintenance measures in advance to avoid structural failure, but also optimizes maintenance strategies and extends the service life of the aircraft.
[0003] As a key component of aircraft flight control and stability, the vertical tail is subjected to complex and variable aerodynamic loads and control loads. Especially with the improvement of fighter performance, its loads are often highly random and sudden. This complex force environment increases the risk of fatigue damage of the vertical tail, so its load monitoring is crucial. Since the force characteristics of the vertical tail are complex and difficult to predict, its load prediction faces significant challenges, and more sophisticated monitoring technology and advanced data analysis models are needed to ensure the accuracy and reliability of the prediction.
[0004] The prediction accuracy depends not only on the selected model and training method, but also on the quality and processing method of the data. At present, the load is predicted based on the maneuvering action of the aircraft. However, for the same maneuvering action, different flight states will have different effects on the structural load. How to explore the impact of different flight states on the structural load and thus improve the load prediction accuracy is a difficult problem that the existing technology needs to solve urgently. Summary of the invention
[0005] In order to solve the above technical problems, the present invention proposes a vertical tail bending moment prediction method based on maneuvering data classification to solve the above problems existing in the prior art.
[0006] To achieve the above object, the present invention provides a method for predicting the vertical tail bending moment based on maneuvering data classification, comprising:
[0007] Acquiring flight parameters and vertical tail bending moment; and obtaining different maneuvering action data according to the flight parameters;
[0008] Classifying the maneuvering action data to obtain category labels, classifying the training samples of the flight parameters according to the category labels, and training the classification model according to the classification results, and training the regression model according to the output results of the trained classification model, wherein the classification model uses the maneuvering action as output, and the regression model uses the predicted vertical tail bending moment as output;
[0009] The current flight parameters are classified and predicted according to the trained classification model and regression model to obtain the prediction results of the vertical tail bending moment of the current flight parameters.
[0010] Optionally, the process of acquiring the maneuvering action data includes:
[0011] Obtaining flight process data according to the flight parameters, wherein the flight process data includes a flight trajectory and a flight attitude;
[0012] Visualizing the flight process data to obtain a flight trajectory diagram and a flight attitude diagram;
[0013] Different maneuvering action data are obtained according to the division criteria and the flight trajectory diagram and the flight attitude diagram, wherein the maneuvering action data include level flight, descent, ascending turn, descending turn, circling descent, ground taxiing and ground turn.
[0014] Optionally, the process of acquiring the flight trajectory in the flight process data includes:
[0015] The spatial coordinates of the flight are obtained, wherein the flight parameters include the altitude, and the coordinates of the x-axis and the y-axis are obtained as follows:
[0016]
[0017] X i =X i-1 +ΔX i
[0018] Y i =Y i-1 +ΔY i
[0019] Among them, α is the angle of attack, u is the pitch angle, ψ is the heading angle, W is the wind speed, η is the angle between the wind speed and the north direction, V is the speed, Δt is the time step, X i and Y i is the position of the aircraft relative to the origin, X represents the x-axis coordinate, Y represents the y-axis coordinate, and Δ represents the change.
[0020] Optionally, before classifying the flight parameters, the method further includes:
[0021] The flight parameters are screened according to the correlation between the flight parameters and between the flight parameters and the vertical tail bending moment; a linear correlation analysis is performed between the flight parameters, and if the correlation between the flight parameters is greater than 0.9, one of the corresponding flight parameters is removed; a nonlinear correlation analysis is performed between the flight parameters and the vertical tail bending moment, and if the correlation between the flight parameters and the vertical tail bending moment is less than 0.2, the corresponding flight parameters are removed.
[0022] Optionally, based on the Pearson correlation coefficient, pairwise correlations between flight parameters are calculated to obtain a correlation coefficient matrix, and the correlation coefficients in the correlation coefficient matrix are used as the correlations between the flight parameters.
[0023] Optionally, the process of obtaining the correlation between the flight parameters and the vertical tail bending moment includes:
[0024] Under different maneuvering data, the flight parameter data and the vertical tail bending moment are calculated:
[0025]
[0026] Wherein, p(x,y) is the joint probability density function of X and Y, p(x) and p(y) are the marginal probability distribution functions of X and Y respectively, where x represents the flight parameter, y represents the vertical tail bending moment, X and Y represent the set of a single flight parameter and the vertical tail bending moment respectively, and I(X;Y) represents the mutual information value of the flight parameter and the vertical tail bending moment;
[0027] The mutual information value between the flight parameter and the vertical tail bending moment is used as the correlation between the flight parameter and the vertical tail bending moment.
[0028] Optionally, the process of classifying the flight parameters includes:
[0029] The maneuvering action data are classified to obtain category labels, the flight parameters are divided into a training set and a test set, the training set is classified by a clustering algorithm, and the optimal number of clusters of the clustering algorithm is obtained by an elbow method;
[0030] The training set is processed by using a clustering algorithm with an optimal number of clusters to obtain category labels corresponding to the training set. The classification model is trained based on the training set and the corresponding category labels. The test set is processed by the trained classification model to obtain category labels corresponding to the test set.
[0031] Optionally, the process of training a classification model includes:
[0032] The classification model is trained according to the training set and the corresponding category labels, and the test set is classified by the trained classification model to obtain the category labels corresponding to the test set, wherein the classification model adopts the GBM classifier.
[0033] Optionally, the process of training a regression model includes:
[0034] The training set and the test set under different category labels are trained and tested by the regression model to obtain an optimized regression model, wherein the regression model adopts a GBM regressor.
[0035] On the other hand, the present invention provides a vertical tail bending moment prediction system based on maneuvering data classification, which is used to execute the above method.
[0036] Compared with the prior art, the present invention has the following advantages and technical effects:
[0037] The present invention collects different flight parameters and vertical tail bending moments based on different maneuvering actions, and considers the correlation between flight parameters and vertical tail bending moments according to different maneuvering action pairs. On this basis, a maneuvering flight parameter-vertical tail load model is constructed. The above scheme takes into account the impact of different flight states on load prediction, and by classifying the maneuvering data, the prediction accuracy of the vertical tail bending moment is significantly improved. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:
[0039] Figure 1 The vertical tail bending moment and the corresponding flight parameter time history of the embodiment of the present invention;
[0040] Figure 2 A schematic diagram of the division of maneuvering actions according to an embodiment of the present invention;
[0041] Figure 3 This is a schematic diagram of an aircraft circling and descending according to an embodiment of the present invention;
[0042] Figure 4 Schematic diagram of correlation coefficients between various flight parameters in an embodiment of the present invention;
[0043] Figure 5 A schematic diagram of calculation results of mutual information values of some actions according to an embodiment of the present invention;
[0044] Figure 6 A schematic diagram of calculation results of mutual information values of another part of actions in an embodiment of the present invention;
[0045] Figure 7A flow chart of a method for classifying mobile data according to an embodiment of the present invention;
[0046] Figure 8 1 is the SSE variation curve of each action in the embodiment of the present invention. DETAILED DESCRIPTION
[0047] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.
[0048] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.
[0049] In this embodiment, a method for predicting the vertical tail bending moment based on maneuvering data classification is provided, and the method includes the following contents:
[0050] 1. For the vertical tail structure, its main load is bending moment. The present invention selects the measured bending moment of the vertical tail of a certain type of aircraft as the research object and establishes a flight parameter-vertical tail bending moment model. The data source includes the flight parameters and vertical tail bending moment recorded synchronously during 10 complete flight takeoffs and landings of a certain subject.
[0051] The flight parameters include altitude, speed, angle of attack, roll angle, heading angle, pitch angle, roll angular velocity, yaw angular velocity, pitch angular velocity, total fuel volume, rear fuselage side load coefficient, rudder position voting value, left horizontal tail position voting value, right horizontal tail position voting value, speed brake position, sideslip angle and flight parameter normal load coefficient.
[0052] In the data collection during the maneuvering action, the maneuvering action is defined as starting from the 1g level flight state, performing a series of actions, and then returning to the 1g level flight state as a complete maneuver. The specific division method is as follows: First, based on the peak point of the normal overload of the center of gravity, search forward and backward along the time axis to find the time point when the normal overload of the center of gravity first approaches 1g and the aircraft state parameters (such as roll angle, pitch angle, roll angular velocity, etc.) are close to zero, which are respectively used as the start and end time of the maneuvering action. According to this method, several different maneuvers are divided, and the multi-parameter time history of each type of maneuvering action, that is, the time series of different flight parameters of different maneuvers, can be extracted from the data of each flight takeoff and landing.
[0053] In order to improve the accuracy of maneuvering action division, the method of visualizing flight parameters is adopted to assist in identifying and distinguishing different maneuvers by drawing the flight trajectory and attitude changes of the aircraft.
[0054] The only flight parameter about the aircraft position is the height, which is the coordinate of the Z axis. The coordinates of the X and Y axes need to be calculated based on the relevant parameters. The following formula is used for calculation:
[0055]
[0056] X i =X i-1 +ΔX i
[0057] Y i =Y i-1 +ΔY i
[0058] Where: α is the angle of attack, u is the pitch angle, ψ is the heading angle, W is the wind speed, η is the angle between the wind speed and the north direction, V is the speed, Δt is the time step, X i and Y i is the position of the aircraft relative to the origin.
[0059] By determining the aircraft's spatial coordinates X, Y, Z, the aircraft's velocity roll angle γ, track pitch angle θ, and heading angle ψ, the aircraft's flight trajectory and flight attitude can be accurately depicted. These parameters are visualized using MATLAB functions to obtain the trajectory and attitude diagram of the aircraft's entire flight takeoff and landing process.
[0060] The maneuvering segments are determined according to the division criteria, and the maneuvers are judged based on the aircraft trajectory and attitude diagram through the experience of relevant experts. A certain subject is divided into several maneuvers such as level flight, descent, ascending turn, descending turn, circling descent, ground taxiing and ground turn.
[0061] 2. Conduct correlation analysis between the above flight parameters and between the flight parameters and the vertical tail bending moment:
[0062] Correlation analysis between flight parameters:
[0063] In order to simplify the model structure, improve computational efficiency, and avoid model instability caused by multicollinearity between parameters, the linear correlation between flight parameters is first analyzed to determine which parameters can be deleted and simplified, thereby reducing model instability.
[0064] (1) Perform pairwise correlation analysis on all parameters in all flight data, calculate the Pearson correlation coefficients between various flight parameters, and form a correlation coefficient matrix. These correlation coefficients can reflect the strength of the linear relationship between the parameters.
[0065] (2) According to the correlation coefficient matrix, for the flight parameter pairs with the absolute value of the correlation coefficient greater than 0.9, one of the parameters is selectively eliminated, and random elimination can be selected. These highly correlated parameters contain redundant information.
[0066] Correlation analysis between flight parameters and vertical tail bending moment
[0067] (1) Analysis method
[0068] In the prediction of vertical tail bending moment load, there is a complex nonlinear relationship between flight parameters and vertical tail bending moment. In order to fully explore the potential correlation between flight parameters and vertical tail bending moment and identify the key flight parameters that have an important impact on load changes, the mutual information method is used for screening. By calculating the mutual information value between flight parameters and vertical tail bending moment, the contribution (correlation) of each flight parameter to the vertical tail bending moment is evaluated, and the flight parameters with higher contribution to load prediction are screened out.
[0069] The mutual information value of two discrete random variables X and Y is calculated as follows:
[0070]
[0071] Wherein, p(x,y) is the joint probability density function of X and Y, p(x) and p(y) are the marginal probability distribution functions of X and Y, respectively, where x represents the flight parameter, y represents the vertical tail bending moment, X and Y represent the set of single flight parameters and vertical tail bending moment, respectively, and I(X;Y) represents the mutual information value of X and Y.
[0072] In the case of continuous random variables, the sum is replaced by a double definite integral:
[0073]
[0074] The flight parameter x and the vertical tail bending moment y are considered as continuous random variables, and Python is used to calculate the mutual information value between the bending moments of each flight in each maneuver.
[0075] (2) Screening principles
[0076] Under different maneuvers, the mutual information between each flight parameter and the vertical tail bending moment is calculated according to the above formula. Since the mutual information values between the flight parameters and the vertical tail bending moment of different maneuvers are different, in order to make an effective comparison, the mutual information values between the flight parameters and the vertical tail bending moment of each maneuver are normalized, and the values with mutual information values lower than 0.2 are removed.
[0077] 3. Mobile data classification method:
[0078] The same maneuver will have different effects on the structural load under different flight parameters (such as altitude, speed, etc.). According to the range of these parameters, the same maneuver can be divided into different flight states, and different flight states will have different effects on the load. Mining the impact of maneuver data classification on prediction accuracy.
[0079] Classify the training set to obtain the maneuver labels:
[0080] The collected flight parameters and corresponding maneuvers are divided into training set and test set, and the K-means method is used to classify the training set. To determine the optimal clustering number K, the elbow method is used to draw a curve of the sum of squared errors (SSE) as K changes, as shown in the following formula:
[0081]
[0082] Among them, S i is the set of data points of the ith class, c i is the i-th cluster center, and x represents the data points of different flight parameters. As K increases, SSE gradually decreases, but when K reaches the actual number of clusters, the decline of SSE slows down significantly, forming an "elbow" inflection point, and then determining the optimal number of clusters K.
[0083] After determining the value of K, the training set is classified as follows:
[0084] (1) Randomly initialize K cluster centers;
[0085] (2) Calculate the Euclidean distance between the data point of each flight parameter and the cluster center, and assign the data to the nearest cluster center:
[0086]
[0087] Where x is the data point, c i is the i-th cluster center, d is the dimension of the data, x j and c ij are x and c respectively i The value in the jth dimension, dist(x,c i ) represents the distance between the data sample and the cluster center.
[0088] (3) Update each cluster center to the mean of all data points in the current cluster:
[0089]
[0090] Among them, S i is the set of data points of the ith cluster, |S i | is the number of data points in the set.
[0091] (4) Repeat the above steps until the cluster centers converge or the maximum number of iterations is reached.
[0092] Test set classification and prediction:
[0093] After obtaining the category labels of the maneuvers in the training set through the K-means method, the flight parameters are used as input data and the maneuvers are used as labels to train the classification model (the GBM classifier is used in this invention), and the test set data is classified into the category of the corresponding maneuvers.
[0094] On this basis, the GBM regressor is used to train and predict the training set and test set data of each category to generate an optimized regression model. The input data of the GBM regressor is the flight parameters, and the output data is the vertical tail bending moment data. Each type corresponds to a regressor, and the classifier is used to classify the data to be tested and indicate which regressor to use.
[0095] Prediction results of each maneuver model
[0096] The maneuvering action classification method is used to establish the maneuvering flight parameter-vertical tail load model. In order to ensure the accuracy and effectiveness of the conclusion, the hyperparameters of the model are optimized before and after classification to ensure that the results obtained before and after classification are optimal.
[0097] Next, GBM is used to train and predict the data before and after classification, and the correlation coefficient r and mean square error MSE are used to quantify the model accuracy. Based on the model accuracy, it is determined whether the optimized regressor meets the usage criteria.
[0098] The flight parameters acquired in real time are classified and identified through optimized classification model and regression model to obtain the corresponding vertical tail bending moment data.
[0099] The above technical solution of the present invention is described in detail with reference to the relevant drawings and specific data:
[0100] 1. Data processing
[0101] 1.1 Flight parameters and load samples
[0102] For the vertical tail structure, the main load is the bending moment. This paper selects the measured bending moment of the vertical tail of a certain type of aircraft as the research object and establishes a flight parameter-vertical tail bending moment model. The data source includes the flight parameters and vertical tail bending moment recorded synchronously during 10 complete flight takeoffs and landings of a certain subject.
[0103] To facilitate subsequent discussion, each flight parameter is numbered, and the corresponding relationship is shown in Table 1. Table 1 shows the flight parameter number and meaning.
[0104] Table 1
[0105]
[0106]
[0107] 1.2 Motor Action Recognition
[0108] A maneuver is defined as starting from a 1g level flight state, performing a series of actions, and then returning to a 1g level flight state as a complete maneuver. The specific division method is as follows: First, taking the peak point of the normal overload at the center of gravity as the reference, search forward and backward along the time axis to find the time point when the normal overload at the center of gravity approaches 1g for the first time and the aircraft state parameters (such as roll angle, pitch angle, roll angular velocity, etc.) are close to zero, which are used as the start and end time of the maneuver respectively. According to this method, the multi-parameter time history of various maneuvers can be extracted from the data of each flight takeoff and landing. Figure 1-2 As shown, Figure 1 is the vertical tail bending moment of a certain takeoff and landing and the corresponding typical flight parameter time history change, Figure 2 Schematic diagram of the division of maneuvers in multi-parameter time history.
[0109] In order to improve the accuracy of maneuvering action division, the method of visualizing flight parameters is adopted to assist in identifying and distinguishing different maneuvers by drawing the flight trajectory and attitude changes of the aircraft.
[0110] The original parameters about the aircraft position are only the height, which is the coordinate of the Z axis. The coordinates of the X and Y axes need to be calculated based on the parameters. The following formula is used for calculation:
[0111]
[0112] X i =X i-1 +ΔX i #(2-3)
[0113] Y i =Y i-1 +ΔY i #(2-4)
[0114] Where: α is the angle of attack, u is the pitch angle, ψ is the heading angle, W is the wind speed, η is the angle between the wind speed and the north direction, V is the speed, Δt is the time step, X i and Y i is the position of the aircraft relative to the origin.
[0115] By determining the aircraft's spatial coordinates X, Y, Z, the aircraft's velocity roll angle γ, track pitch angle θ, and heading angle ψ, the aircraft's flight trajectory and flight attitude can be accurately depicted. These parameters are visualized using MATLAB functions to obtain the trajectory and attitude diagram of the aircraft's entire flight takeoff and landing process.
[0116] According to the classification criteria and the aircraft trajectory and attitude diagram, a subject is divided into several maneuvers such as level flight, descent, ascending turn, descending turn, circling descent, ground taxiing and ground turn. Some of the maneuvers are demonstrated, such as ground turn. Figure 3 The ground taxiing and turning are the same as the level flight and turning actions, but the height is different.
[0117] 1.3 Correlation Analysis
[0118] 1.3.1 Correlation analysis between flight parameters
[0119] In order to simplify the model structure, improve computational efficiency, and avoid model instability caused by multicollinearity among parameters, the linear correlation among flight parameters is first analyzed to determine which parameters can be deleted and simplified.
[0120] (1) Perform pairwise correlation analysis on all parameters in the flight data, calculate the Pearson correlation coefficients between the various flight parameters, and form a correlation coefficient matrix. These correlation coefficients can reflect the strength of the linear relationship between the parameters.
[0121] (2) According to the correlation coefficient matrix, for flight parameter pairs with an absolute value of correlation coefficient greater than 0.9, one of the parameters is selectively eliminated. These highly correlated parameters contain redundant information.
[0122] The correlation coefficients among the parameters are Figure 4 As shown:
[0123] Depend on Figure 4 It can be seen that the correlation coefficients between 18 and 19, 22, 28 and 12, 23, 29 and 13, and 30 and 14 are greater than 0.9, so only one of them needs to be retained. Therefore, the flight parameters numbered 18, 22, 23, 28, 29 and 30 are screened out.
[0124] 1.3.2 Correlation analysis between flight performance and vertical tail bending moment
[0125] (1) Analysis method
[0126] In the prediction of vertical tail bending moment load, there is a complex nonlinear relationship between flight parameters and vertical tail bending moment. In order to fully explore the potential correlation between flight parameters and vertical tail bending moment and identify the key flight parameters that have an important impact on load changes, the mutual information method is used for screening. By calculating the mutual information value between flight parameters and vertical tail bending moment, the contribution (correlation) of each flight parameter to the vertical tail bending moment is evaluated, and the flight parameters with higher contribution to load prediction are screened out.
[0127] The mutual information value of two discrete random variables X and Y is calculated as follows:
[0128]
[0129] Where p(x,y) is the joint probability density function of X and Y, and p(x) and p(u) are the marginal probability distribution functions of X and Y, respectively.
[0130] In the case of continuous random variables, the sum is replaced by a double definite integral:
[0131]
[0132] The flight parameters p(x) and bending moments p(y) are considered as continuous random variables, and Python is used to calculate the mutual information value between each flight parameter and bending moment of each maneuver.
[0133] (2) Screening principles
[0134] Since the mutual information values of different maneuvering flight parameters and the vertical tail bending moment are different, in order to make an effective comparison, the mutual information values of the maneuvering flight parameters and the vertical tail bending moment are normalized, and the values with a mutual information value lower than 0.2 are screened.
[0135] (3) Mutual information value calculation results
[0136] The calculation results of the normalized mutual information value between the bending moments of level flight, turning, ascending, descending, ascending turn, descending turn, and circling descent are as follows: Figure 5-6 shown.
[0137] (4) Filter results
[0138] The flight parameter screening results of each maneuver are shown in Table 2, where Table 2 is the flight parameter screening results of each action.
[0139] Table 2
[0140]
[0141] 3.1 Mobile data classification method
[0142] The same maneuver will have different effects on the structural load under different flight parameters (such as altitude, speed, etc.). According to the range of these parameters, the same maneuver can be divided into different flight states, and the impact of different flight states on the load will also be different. This section mainly discusses the impact of maneuver data classification on prediction accuracy. The overall process is as follows Figure 7 shown.
[0143] 3.1.1 Training set classification to obtain labels
[0144] The K-means method is used to classify the training set. To determine the optimal number of clusters, the elbow method is used to draw a curve of the sum of squared errors (SSE) versus the number of clusters, as shown in the following formula:
[0145]
[0146] Among them, S i is the set of data points of the ith class, c i is the i-th cluster center. As K increases, SSE gradually decreases, but when K reaches the actual number of clusters, the decline of SSE slows down significantly, forming an "elbow" inflection point, and then determining the optimal number of clusters K.
[0147] After determining the value of K, the training set is classified as follows:
[0148] (1) Randomly initialize K cluster centers;
[0149] (2) Calculate the Euclidean distance between each data point and the cluster center, and assign the data to the nearest cluster center:
[0150]
[0151] Where x is the data point, c i is the i-th cluster center, d is the dimension of the data, x j and c ij are x and c respectively i The value in the jth dimension.
[0152] (3) Update each cluster center to the mean of all data points in the current cluster:
[0153]
[0154] Among them, S i is the set of data points of the ith cluster, |S i | is the number of data points in the set.
[0155] (4) Repeat the above steps until the cluster centers converge or the maximum number of iterations is reached.
[0156] 3.1.2 Test set classification and prediction
[0157] After obtaining the category labels of the training set through the K-means method, the classification model is trained based on the classification results (GBM classifier is used in this paper), and the test set data is classified into the corresponding category.
[0158] On this basis, the GBM regressor is used to train and predict the training set and test set data of each category respectively.
[0159] 3.1.3 Prediction results of each maneuver model
[0160] The maneuvering action classification method is used to establish the maneuvering flight parameter-vertical tail load model. In order to ensure the accuracy and effectiveness of the conclusion, the hyperparameters of the model are optimized before and after classification to ensure that the results obtained before and after classification are optimal.
[0161] First, through the SSE (sum of squared errors) change curves of each action (see Figure 8 ,in, Figure 8 The optimal cluster number K of each maneuver is determined by (a) corresponding to ascending turn; (b) corresponding to descent; (c) corresponding to descending turn; (d) corresponding to level flight; (e) corresponding to circling descent. The results show that the optimal cluster number of different maneuvers is different, reflecting the complexity of different maneuvers in data distribution. Among them, the cluster number of ascending turn, descending turn and circling descent is relatively high, indicating that these actions contain more state changes, while the cluster number of level flight and descent is relatively small and the state is relatively single. The classification results are summarized in Table 3, and Table 4 is the classification results.
[0162] Table 3
[0163]
[0164] Next, GBM is used to train and predict the data before and after classification, and the correlation coefficient r and mean square error MSE are used to quantify the model accuracy. As shown in Table 4, the accuracy of the model after classification is generally better than that before classification, indicating that data classification can effectively improve the accuracy of load prediction. Table 5 shows the prediction effect before and after classification.
[0165] Table 4
[0166]
[0167] Rising turn GBM r = 0.91, MSE = 0.289 r = 0.93, MSE = 0.235
[0168] Downward GBM r = 0.88, MSE = 0.154 r = 0.89, MSE = 0.141
[0169] Descending turn GBM r = 0.9, MSE = 0.125 r = 0.932, MSE = 0.09
[0170] Level flight GBM r = 0.857, MSE = 0.0813r = 0.873, MSE = 0.073
[0171] Spiraling down GBM r = 0.93, MSE = 0.9748 r = 0.952, MSE = 0.85
[0172] The present invention takes into account the influence of different flight states on load prediction and significantly improves the prediction accuracy of the vertical tail bending moment by classifying maneuvering data.
[0173] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application 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 predicting the vertical tail bending moment based on maneuvering data classification, characterized in that: include: Obtain flight parameters and vertical tail bending moment; According to the flight parameters, different maneuvering action data are obtained; Classifying the maneuvering action data to obtain category labels, classifying the training samples of the flight parameters according to the category labels, and training the classification model according to the classification results, and training the regression model according to the output results of the trained classification model, wherein the classification model uses the maneuvering action as output, and the regression model uses the predicted vertical tail bending moment as output; The current flight parameters are classified and predicted according to the trained classification model and regression model to obtain the prediction results of the vertical tail bending moment of the current flight parameters.
2. The method according to claim 1, characterized in that The process of acquiring maneuvering action data includes: Obtaining flight process data according to the flight parameters, wherein the flight process data includes a flight trajectory and a flight attitude; Visualizing the flight process data to obtain a flight trajectory diagram and a flight attitude diagram; Different maneuvering action data are obtained according to the division criteria and the flight trajectory diagram and the flight attitude diagram, wherein the maneuvering action data include level flight, descent, ascending turn, descending turn, circling descent, ground taxiing and ground turn.
3. The method according to claim 2, characterized in that The process of obtaining the flight trajectory in the flight process data includes: The spatial coordinates of the flight are obtained, wherein the flight parameters include the altitude, and the coordinates of the x-axis and the y-axis are obtained as follows: X i =X i-1 +ΔX i Y i =Y i-1 +ΔY i Among them, α is the angle of attack, u is the pitch angle, ψ is the heading angle, W is the wind speed, η is the angle between the wind speed and the north direction, V is the speed, Δt is the time step, X i and Y i is the position of the aircraft relative to the origin, X represents the x-axis coordinate, Y represents the y-axis coordinate, and Δ represents the change.
4. The method according to claim 1, characterized in that: Prior to classifying the flight parameters, the following steps are also included: The flight parameters are screened according to the correlation between the flight parameters and between the flight parameters and the vertical tail bending moment; a linear correlation analysis is performed between the flight parameters, and if the correlation between the flight parameters is greater than 0.9, one of the corresponding flight parameters is removed; a nonlinear correlation analysis is performed between the flight parameters and the vertical tail bending moment, and if the correlation between the flight parameters and the vertical tail bending moment is less than 0.2, the corresponding flight parameters are removed.
5. The method according to claim 4, characterized in that According to the Pearson correlation coefficient, the pairwise correlations between the flight parameters are calculated to obtain a correlation coefficient matrix, and the correlation coefficients in the correlation coefficient matrix are used as the correlations between the flight parameters.
6. The method according to claim 4, characterized in that The process of obtaining the correlation between flight parameters and vertical tail bending moment includes: Under different maneuvering data, the flight parameter data and the vertical tail bending moment are calculated: Wherein, p(x,y) is the joint probability density function of X and Y, p(x) and p(y) are the marginal probability distribution functions of X and Y respectively, where x represents the flight parameter, y represents the vertical tail bending moment, X and Y represent the set of a single flight parameter and the vertical tail bending moment respectively, and I(X;Y) represents the mutual information value of the flight parameter and the vertical tail bending moment; The mutual information value between the flight parameter and the vertical tail bending moment is used as the correlation between the flight parameter and the vertical tail bending moment.
7. The method according to claim 1, characterized in that The process of classifying the flight parameters includes: The maneuvering action data are classified to obtain category labels, the flight parameters are divided into a training set and a test set, the training set is classified by a clustering algorithm, and the optimal number of clusters of the clustering algorithm is obtained by an elbow method; The training set is processed by using a clustering algorithm with an optimal number of clusters to obtain category labels corresponding to the training set. The classification model is trained based on the training set and the corresponding category labels. The test set is processed by the trained classification model to obtain category labels corresponding to the test set.
8. The method according to claim 7, characterized in that The process of training a classification model includes: The classification model is trained according to the training set and the corresponding category labels, and the test set is classified by the trained classification model to obtain the category labels corresponding to the test set, wherein the classification model adopts the GBM classifier.
9. The method according to claim 8, characterized in that The process of training a regression model includes: The training set and the test set under different category labels are trained and tested by the regression model to obtain an optimized regression model, wherein the regression model adopts a GBM regressor.
10. The vertical tail bending moment prediction system based on maneuvering data classification is characterized by: Used to execute the method according to any one of claims 1 to 9.