A method for predicting longitudinal available overload of aircraft under fault conditions based on ensemble learning algorithm

Abstract

The present application discloses a method for predicting the longitudinal available overload of an aircraft under fault conditions based on an integrated learning algorithm, and relates to the field of aviation aircraft. The method first constructs a theoretical model of the longitudinal available overload of the aircraft, and then performs feature screening on each group of motion simulation data of the aircraft under different flight conditions, and uses the theoretical model to calculate the difference between the longitudinal available overload under each flight condition and the longitudinal available overload under a fault-free condition based on the time series data of the screened feature parameters as the longitudinal available overload variation, and then obtains a longitudinal available overload variation prediction model based on the integrated learning algorithm and the simulation data training. The method combines the longitudinal available overload variation prediction model obtained by the integrated learning algorithm with the longitudinal available overload theoretical model obtained by theoretical analysis, and can accurately and reliably obtain the longitudinal available overload prediction result when a control surface failure occurs in the aircraft.

Description

Technical Field

[0001] The present application relates to the field of aviation aircraft, and in particular to a method for predicting the longitudinal available overload of an aircraft under fault conditions based on an integrated learning algorithm. Background Art

[0002] For aircraft with a long development cycle and long mission range, quickly and accurately predicting their flight performance during flight can provide information support for strategic planning adjustments and mission execution capabilities. The flight performance of aircraft mainly includes control performance, range capability, maneuverability and other aspects.

[0003] The maneuverability of an aircraft refers to the ability of an aircraft to change its flight state within a certain time interval. It is an extremely important tactical and technical indicator, and is also one of the important indicators to measure the level of aircraft design and manufacturing. The maneuverability of an aircraft depends to a large extent on the overload it can generate. The greater the overload, the greater the speed or direction the aircraft can change in the same time, thus having better maneuverability. Therefore, overload is an important indicator for evaluating the maneuverability of an aircraft, among which longitudinal overload is the most critical. Longitudinal overload focuses on the force applied to an aircraft in the vertical direction, and refers to the ratio of the sum of external forces applied to the aircraft other than gravity to the weight of the aircraft. Therefore, the industry currently often predicts and evaluates the maneuverability of an aircraft by predicting its longitudinal overload.

[0004] At present, there are relatively few literatures on the prediction and evaluation of the maneuverability of aircraft, which mainly focus on analytical methods and machine learning-based methods: (1) When evaluating the maneuverability of aircraft based on analytical methods, the parameters under certain conditions are fitted into algebraic functions in the motion equation set with the aircraft as a particle, and then the maneuverability of the aircraft is solved. The analytical method relies on a specific and accurate aircraft model, so it is easy to produce large errors when the model changes or contains interference and uncertainty. (2) When using machine learning methods to evaluate the maneuverability of aircraft, a statistical distribution model or a neural network model is established based on flight test data for predictive evaluation. Machine learning methods rely on a large amount of highly reliable flight test data. Most of the above research methods are based on definite analytical functions of performance indicators, and conduct qualitative analysis of maneuverability performance in the verification phase before the aircraft is put into use, or predict performance indicators before flight based on probability distribution fitting of flight experiment data in the existing test flight phase. In actual flight, it is often necessary to provide prediction results of definite maneuverability performance indicators under the current flight state. Currently, neither analytical methods nor machine learning methods can meet this demand well, especially after an aircraft malfunctions during a flight mission, it is even more difficult to predict maneuverability performance. Summary of the invention

[0005] In response to the above problems and technical requirements, this application proposes a method for predicting the longitudinal available overload of an aircraft under fault conditions based on an ensemble learning algorithm. The technical solution of this application is as follows:

[0006] A method for predicting the longitudinal overload of an aircraft under fault conditions based on an integrated learning algorithm, the method comprising:

[0007] Based on the kinematic model of the aircraft and the instantaneous balance assumption, a theoretical model of the longitudinal available overload of the aircraft is constructed. The longitudinal available overload theoretical model represents the longitudinal available overload of the aircraft. Relationship with flight status parameters;

[0008] Acquire motion simulation data obtained by motion simulation of the aircraft under different flight states, the flight state of the aircraft includes a fault-free state and a plurality of different control surface fault states, and each set of motion simulation data of the aircraft includes time series data of a plurality of motion parameters of the aircraft under the current flight state;

[0009] Performing feature screening on the motion parameters to obtain feature parameters, the screened feature parameters include flight state parameters;

[0010] The longitudinal available overload theoretical model is used to calculate the longitudinal available overload under the flight state according to the characteristic parameter time series data under each flight state; and the difference between the longitudinal available overload under each flight state and the longitudinal available overload under the fault-free state is calculated to obtain the longitudinal available overload change under the flight state;

[0011] Taking the characteristic parameter time series data and flight state in each flight state as input and the longitudinal available overload change in the flight state as output, a longitudinal available overload change prediction model is obtained based on ensemble learning algorithm training;

[0012] By utilizing the longitudinal available overload theoretical model and the trained longitudinal available overload change prediction model, the longitudinal available overload prediction result of the aircraft in the control surface failure state is obtained based on the actual motion data of the aircraft in the actual flight state.

[0013] A further technical solution is to construct a theoretical model of longitudinal available overload of an aircraft, including:

[0014] The longitudinal overload model of the aircraft is constructed based on the kinematic model of the aircraft and the instantaneous balance assumption. The longitudinal overload model represents the longitudinal overload n of the aircraft. y Relationship with flight state parameters and rudder deflection angle;

[0015] The rudder deflection angle in the longitudinal overload model is taken to the extreme position to obtain the longitudinal available overload theoretical model.

[0016] A further technical solution is that the obtained longitudinal available overload theoretical model is:

[0017]

[0018] Where Q is the dynamic pressure and ρ is the air density, V is the flight speed of the aircraft, S is the reference area of ​​the rudder, m is the mass of the aircraft, g is the acceleration of gravity, P is the thrust of the aircraft, and C Y0 is the lift coefficient when the angle of attack and elevator deflection are zero, C Yα is the stability derivative of the lift coefficient with respect to the angle of attack, is the pitch moment coefficient when the angle of attack, sideslip angle and multiple deviation terms are zero, is the stability derivative of the pitching moment with respect to the rudder angle, is the stability derivative of the pitch moment coefficient with respect to the angle of attack; (δ1-δ2-δ3+δ4) limit It indicates that the rudder deflection angle δ1-δ2-δ3+δ4 is taken to the extreme position, δ1 is the rudder deflection angle of rudder No. 1 of the aircraft, δ2 is the rudder deflection angle of rudder No. 2 of the aircraft, δ3 is the rudder deflection angle of rudder No. 3 of the aircraft, δ4 is the rudder deflection angle of rudder No. 4 of the aircraft, and rudder No. 1, rudder No. 2, rudder No. 3, and rudder No. 4 are arranged in clockwise order from the upper left corner along the flight direction as viewed from the tail of the aircraft to form an X-shaped structure.

[0019] A further technical solution is to construct a longitudinal overload model of an aircraft including:

[0020] According to the kinematic model of the aircraft, the longitudinal overload n is obtained. y The calculation formula is:

[0021]

[0022] Where α is the angle of attack, is the stability derivative of the lift coefficient with respect to the rudder deflection angle;

[0023] Determined based on the instantaneous equilibrium assumption of the aircraft:

[0024]

[0025] in, is the pitching moment coefficient;

[0026] Obtain the expression for the angle of attack α and substitute it into the longitudinal overload n y The calculation formula of the longitudinal overload model is:

[0027]

[0028] A further technical solution is to obtain the longitudinal overload n according to the kinematic model of the aircraft. y The calculation formula includes:

[0029] The lift coefficient C is obtained by fitting Y The expression is:

[0030]

[0031] According to the lift coefficient C Y The expression of the lift Y on the aircraft is:

[0032]

[0033] According to the lift Y on the aircraft, we can get the total external force N on the aircraft except gravity. y The expression is:

[0034] N y =Psinα+Y

[0035] According to the resultant external force N y Get the longitudinal overload n y The expression is:

[0036]

[0037] Within the allowable error range, the sinα of the aircraft during cruising flight is equivalent to α, and the longitudinal overload n is obtained. y The calculation formula for .

[0038] A further technical solution is to perform feature screening on the motion parameters to obtain feature parameters including:

[0039] Calculate the Spearman correlation coefficient between different motion parameters to perform correlation analysis, and eliminate one of the two motion parameters whose Spearman correlation coefficient reaches the correlation threshold;

[0040] The importance of the remaining motion parameters is analyzed to obtain the importance result of each motion parameter, and the motion parameters whose importance results are lower than the importance threshold are eliminated, and the remaining motion parameters are regarded as feature parameters.

[0041] A further technical solution is to perform importance analysis on the remaining motion parameters to obtain the importance result of each motion parameter. For each of the remaining motion parameters:

[0042] The importance sub-results of motion parameters are calculated using gradient boosting decision tree and random forest algorithm respectively, and the importance result of motion parameters is obtained by weighted average of the two importance sub-results.

[0043] A further technical solution is to obtain a longitudinal available overload change prediction model based on ensemble learning algorithm training, including:

[0044] A model structure of a longitudinal available overload change prediction model is built based on an integrated learning algorithm, and a longitudinal available overload change prediction model is obtained by training based on the built model structure. The model structure includes a base learner and a meta learner connected in sequence, and the base learner is obtained by fusing multiple models.

[0045] A further technical solution is that the base learner in the model structure fuses the extremely random tree model, the LightGBM model and the CatBoost model.

[0046] A further technical solution is to obtain a longitudinal available overload prediction result of an aircraft when the aircraft is in a control surface failure state, including:

[0047] Using the longitudinal available overload theoretical model, according to the actual motion data of the aircraft in a fault-free state, a reference longitudinal available overload prediction value of the aircraft is obtained;

[0048] The longitudinal available overload variation prediction model is used to obtain the predicted value of the longitudinal available overload variation of the aircraft under the current control surface failure state according to the actual motion data and flight state of the aircraft under the control surface failure state;

[0049] The sum of the longitudinal available overload change prediction value and the reference longitudinal available overload prediction value is calculated to obtain the longitudinal available overload prediction result of the aircraft under the current control surface failure state.

[0050] The beneficial technical effects of this application are:

[0051] The present application discloses a method for predicting the longitudinal available overload of an aircraft under fault based on an ensemble learning algorithm. The method combines the ensemble learning algorithm with a theoretical model. A theoretical model of the longitudinal available overload is derived based on the mathematical model of the aircraft to estimate a reference value. A prediction model for the change in the longitudinal available overload under fault is constructed based on the ensemble learning algorithm to predict the deviation relative to the reference value. Combining the two, an accurate and reliable longitudinal available overload prediction result can be obtained when a control surface fault occurs in an aircraft, thus filling in the technical gap.

[0052] When training the longitudinal available overload change prediction model, this method uses a feature optimizer that combines the Spearman correlation coefficient, random forest model and gradient boosting decision tree to screen out feature parameters for model training to eliminate redundant motion parameters, thereby improving the prediction performance of the longitudinal available overload change prediction model and reducing training time. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is a method flow chart of a method for predicting the longitudinal available overload of an aircraft under fault conditions based on an integrated learning algorithm according to an embodiment of the present application.

[0054] Figure 2 It is the heat map result of correlation analysis of different motion parameters.

[0055] Figure 3 It is the heat map result of importance analysis of different motion parameters. DETAILED DESCRIPTION

[0056] The specific implementation of the present application is further described below in conjunction with the accompanying drawings.

[0057] This application discloses a method for predicting the longitudinal available overload of an aircraft under fault conditions based on an ensemble learning algorithm. Figure 1 The flowchart shown in the figure, the method for predicting the longitudinal overload available for the aircraft under the fault includes:

[0058] Step 1: Based on the kinematic model of the aircraft and the instantaneous equilibrium assumption, a theoretical model of the longitudinal available overload of the aircraft is constructed.

[0059] The obtained longitudinal available overload theoretical model represents the longitudinal available overload of the aircraft The relationship between the longitudinal available overload theoretical model of the aircraft is as follows:

[0060] First, the longitudinal overload model of the aircraft is constructed based on the kinematic model of the aircraft and the instantaneous balance hypothesis. The longitudinal overload model obtained in this step represents the longitudinal overload n of the aircraft. y Relationship with flight state parameters and rudder deflection angle. Including:

[0061] 1. First, fit the lift coefficient C Y The expression is:

[0062]

[0063] Among them, C Y0 is the lift coefficient when the angle of attack and elevator deflection are zero, C Yα is the stability derivative of the lift coefficient with respect to the angle of attack, is the stability derivative of the lift coefficient with respect to the rudder deflection angle term. α is the angle of attack. The aircraft of the present application includes four rudders arranged in an X-shaped structure. The rudders starting from the upper left corner in the clockwise direction from the tail of the aircraft along the flight direction are defined as rudder No. 1, rudder No. 2, rudder No. 3, and rudder No. 4, respectively. δ1 is the rudder deflection angle of rudder No. 1 of the aircraft, δ2 is the rudder deflection angle of rudder No. 2 of the aircraft, δ3 is the rudder deflection angle of rudder No. 3 of the aircraft, and δ4 is the rudder deflection angle of rudder No. 4 of the aircraft.

[0064] 2. The lift Y on an aircraft can be calculated by the following formula:

[0065]

[0066] Where Q is the dynamic pressure and ρ is the air density, V is the flight speed of the aircraft, and S is the reference area of ​​the rudder.

[0067] Substituting the above formula (1) into formula (2), we can get the expression of the lift Y on the aircraft:

[0068]

[0069] 3. The aircraft targeted by this application has no side slip during flight, and the engine thrust is in the direction of the flight speed. The lift Y on the aircraft is obtained by the resultant external force N on the aircraft except for gravity. y The expression is:

[0070] N y =Psinα+Y (4)

[0071] Where P is the thrust of the aircraft,

[0072] 4. Due to longitudinal overload n y The definition of is the total external force N on the aircraft except gravity. y The ratio of the longitudinal overload n is: y The expression is:

[0073]

[0074] Where m is the mass of the aircraft and g is the acceleration due to gravity.

[0075] When an aircraft is cruising, its angle of attack α is generally small, so within the allowable error range, the sinα of the aircraft during cruising can be equivalent to α. Thus, the longitudinal overload n can be obtained by rearranging equation (5): y The calculation formula is:

[0076]

[0077] 5. Based on the instantaneous equilibrium assumption of an aircraft, it can be determined that:

[0078]

[0079] in, is the pitching moment coefficient, is the pitch moment coefficient when the angle of attack, sideslip angle and multiple deviation terms are zero, is the stability derivative of the pitching moment with respect to the rudder angle, is the stability derivative of the pitching moment coefficient with respect to the angle of attack.

[0080] 6. The angle of attack α is obtained from equation (7) and substituted into the longitudinal overload n in equation (6): y The calculation formula of the longitudinal overload model is:

[0081]

[0082] 7. Taking the rudder deflection angle in the longitudinal overload model (8) to the extreme position, the longitudinal available overload theoretical model is obtained as:

[0083]

[0084] Among them, (δ1-δ2-δ3+δ4) limit It means that the rudder deflection angle δ1-δ2-δ3+δ4 reaches the extreme position.

[0085] Step 2, obtaining motion simulation data obtained by performing motion simulation of the aircraft under different flight states.

[0086] The flight state of an aircraft includes a fault-free state and a plurality of different control surface fault states. There is at least one difference between the faulty control surface and the fault degree of any two different control surface fault states of the aircraft. The faulty control surface indicates the faulty control surface and the type of fault that occurs. The fault type includes a defect fault, a stuck fault, and a loose fault. The fault degree corresponding to the defect fault indicates the degree of defect, and the fault degree corresponding to the stuck fault indicates the stuck angle. The control surface fault state referred to in the present application refers to the state in which a rudder has a single fault type. For example, in one example, the control surface fault state of each rudder under 20% defect fault, 40% defect fault, 60% defect fault, 5° stuck fault, 10° stuck fault, and loose fault is constructed respectively, and a total of 24 different control surface fault states are constructed. Combined with a fault-free state, the aircraft is subjected to motion simulation under the following 25 different flight states:

[0087]

[0088]

[0089] In addition, in order to enrich the amount of data, each flight state of the aircraft is simulated in different cruising states during the simulation. Each cruising state of the aircraft includes the corresponding mass m, cruising altitude and cruising flight speed V. For example, in one example, multiple masses m are obtained by traversing at 25kg as the traversal interval in the range of 1152kg to 1452kg, multiple cruising altitudes are obtained by traversing at 20m as the traversal interval in the range of 600m to 1000m, and multiple flight speeds V are obtained by traversing at 20m / s as the traversal interval in the range of 200m / s to 280m / s. The traversed mass m, cruising altitude and cruising flight speed V are fully arranged and combined to obtain 1365 different cruising states. Then, the aircraft is simulated in 1365 different cruising states in each flight state, and 1365 sets of motion simulation data of the aircraft in each flight state can be obtained.

[0090] Regardless of how the simulation is performed, each set of motion simulation data of the aircraft obtained through the motion simulation includes time series data of multiple motion parameters of the aircraft in the current flight state. There are quite a lot of motion parameters obtained, including the XYZ-axis displacement of the aircraft in the geodetic coordinate system (respectively denoted as XYZ), the xyz-axis flight speed in the body coordinate system (respectively denoted as Vb_x, Vb_y, Vb_z), the xyz-axis flight acceleration in the body coordinate system (respectively denoted as Accels_x, Accels_y, Accels_z), roll angle phi, yaw angle pasi, pitch angle thet, roll angular velocity p, yaw angular velocity q, pitch angular velocity r, roll angular acceleration dp, yaw angular acceleration dq, pitch angular acceleration dr, angle of attack α, sideslip angle beta, rudder deflection angle δ1 of rudder No. 1, rudder deflection angle δ2 of rudder No. 2, rudder deflection angle δ3 of rudder No. 3, rudder deflection angle δ4 of rudder No. 4, Mach number Ma, mass m, mass change rate dm and air density ρ.

[0091] Since the entire simulation duration is generally relatively long, in order to improve data processing efficiency, the time series data of each motion parameter in each group of motion simulation data in this step does not need to include the time series data of the entire simulation duration. When simulating the fault-free state of an aircraft, the time series data of each motion parameter within a predetermined time period is intercepted to obtain a set of motion simulation data. When simulating the fault state of the rudder of an aircraft, the time series data of each motion parameter within a period of time that is symmetrical before and after the fault injection moment and has a predetermined time period is intercepted to obtain a set of motion simulation data. The predetermined time period is defined according to actual needs, such as 0.2s. When simulating under the rudder fault state, the time series sequence of 0.1s before and after the fault injection moment is taken. In addition, it is necessary to eliminate the abnormal simulation data at the time of the fault occurrence and the cruise state is not controlled.

[0092] Step 3. It can be seen from the above that the motion parameters in the original acquired motion simulation data have a large number of parameters. In order to improve the data processing efficiency, the motion parameters are firstly subjected to feature screening to obtain feature parameters. Since the motion parameters that have a small impact on the longitudinal available overload are eliminated, the screened feature parameters at least include the flight state parameters required to be used in the longitudinal available overload theoretical model (9), including mass m, air density ρ, flight speed, etc.

[0093] The characteristic parameters obtained by screening include:

[0094] (1) Calculate the Spearman correlation coefficient between different motion parameters for correlation analysis, and remove one of the two motion parameters whose Spearman correlation coefficient reaches the correlation threshold. The correlation threshold can be customized, for example, set to 0.95.

[0095] The heat map obtained by correlation analysis of different motion parameters is as follows: Figure 2 As shown, according to Figure 2 The results of the correlation analysis shown in the figure eliminate five motion parameters, including: the X-axis displacement of the aircraft in the geodetic coordinate system, the Z-axis displacement of the aircraft in the geodetic coordinate system, the Y-axis flight acceleration Accels_y of the aircraft in the body coordinate system, the Z-axis flight acceleration Accels_z of the aircraft in the body coordinate system, and the yaw angle pasi.

[0096] (2) Performing importance analysis on the remaining motion parameters to obtain the importance result of each motion parameter, and eliminating the motion parameters whose importance results are lower than the importance threshold, and taking the remaining motion parameters as feature parameters. In one embodiment, the importance sub-results of the motion parameters are calculated using the gradient boosting decision tree and the random forest algorithm, respectively, and the weighted average of the two importance sub-results is performed to obtain the importance result of the motion parameters. The importance threshold can be customized. The results obtained by performing importance analysis on different motion parameters are as follows: Figure 3 As shown, according to Figure 3 The results of the importance analysis shown in the figure exclude two motion parameters whose importance results are 0, including: the y-axis flight speed Vb_y and the pitch angle thet of the aircraft in the body coordinate system.

[0097] Finally, after the above screening, 7 motion parameters were eliminated, and the characteristic parameters obtained by screening included: Y-axis displacement of the aircraft in the geodetic coordinate system, xz-axis flight speed in the body coordinate system, x-axis flight acceleration in the body coordinate system, roll angle, roll angular velocity, yaw angular velocity, pitch angular velocity, roll angular acceleration, yaw angular acceleration, pitch angular acceleration, angle of attack, sideslip angle, rudder deflection angle δ1 of rudder No. 1, rudder deflection angle δ2 of rudder No. 2, rudder deflection angle δ3 of rudder No. 3, rudder deflection angle δ4 of rudder No. 4, Mach number, mass m, mass change rate and air density ρ.

[0098] Step 4: Using the longitudinal available overload theoretical model, the longitudinal available overload in the current flight state can be calculated based on the characteristic parameter time series data in each flight state. Then, the difference between the longitudinal available overload in each flight state and the longitudinal available overload in the fault-free state is calculated to obtain the longitudinal available overload change in the current flight state.

[0099] Step 5: Taking the characteristic parameter time series data and flight state in each flight state as input and the longitudinal available overload change in the current flight state as output, a longitudinal available overload change prediction model is obtained based on ensemble learning algorithm training. It includes:

[0100] First, a model structure of a longitudinal available overload change prediction model is constructed based on an integrated learning algorithm, and then the longitudinal available overload change prediction model is trained based on the constructed model structure. The constructed model structure includes a base learner and a meta learner connected in sequence, and the base learner is obtained by model fusion of multiple models. In one embodiment, the base learner is obtained by model fusion of an extreme random tree model, a LightGBM model, and a CatBoost model.

[0101] Step 6: Then, the model can be deployed, using the longitudinal available overload theoretical model and the trained longitudinal available overload change prediction model, and based on the actual motion data of the aircraft in actual flight state, the longitudinal available overload prediction result of the aircraft in the control surface failure state is obtained. Including:

[0102] The actual motion data of the aircraft is obtained in the actual flight state. The data content of the actual motion data has the same meaning as the motion simulation data in the above step 2, and this embodiment will not be repeated.

[0103] By using the longitudinal available overload theoretical model and the actual motion data of the aircraft in a fault-free state, a relatively accurate baseline longitudinal available overload prediction value of the aircraft can be obtained. However, when the aircraft is detected to have a control surface fault state in the pre-sequence stage, the result directly calculated using the longitudinal available overload theoretical model often has a large error. At this time, the longitudinal available overload variation prediction model can be used to obtain a relatively accurate longitudinal available overload variation prediction value of the aircraft in the current control surface fault state based on the actual motion data and flight state of the aircraft in the control surface fault state. Further calculating the sum of the longitudinal available overload variation prediction value and the baseline longitudinal available overload prediction value can obtain the longitudinal available overload prediction result of the aircraft in the current control surface fault state.

[0104] The present application uses a longitudinal available overload theoretical model to provide a benchmark longitudinal available overload prediction value, and combines the longitudinal available overload change prediction model obtained by training based on an integrated learning algorithm to obtain the change. The longitudinal available overload prediction result is obtained by combining the two. The longitudinal available overload theoretical model provides a benchmark result, so the obtained longitudinal available overload prediction result will not deviate from this benchmark result and have a large error. Therefore, compared with directly using the model to predict the longitudinal available overload, the prediction accuracy is higher. In addition, feature screening is performed when training the longitudinal available overload change prediction model, which not only reduces the amount of data processing, but also reduces the interference caused by redundant data, which is conducive to improving the prediction accuracy of the longitudinal available overload change. The present application tests and compares the model obtained by directly training with all motion parameters and the model obtained by training with the screened feature parameters in the same test scenario. The number of test sets is 63678, and the test comparison results are as follows:

[0105]

[0106] As can be seen from the test comparison results in the above table, the model trained using the screened feature parameters has better prediction performance while reducing the amount of data processing. The relative error is basically within 5%, and the absolute error is basically within 0.025.

[0107] The above is only a preferred embodiment of the present application, and the present application is not limited to the above embodiments. It is understood that other improvements and changes directly derived or associated by those skilled in the art without departing from the spirit and concept of the present application should be considered to be included in the protection scope of the present application.

Claims

1. A method for predicting the longitudinal available overload of an aircraft under fault conditions based on an ensemble learning algorithm, characterized in that: The method for predicting the longitudinal available overload of an aircraft under fault conditions includes: Based on the kinematic model of the aircraft and the instantaneous balance assumption, a longitudinal available overload theoretical model of the aircraft is constructed, and the longitudinal available overload theoretical model represents the longitudinal available overload of the aircraft. Relationship with flight status parameters; Acquiring motion simulation data obtained by performing motion simulation on the aircraft in different flight states, wherein the flight state of the aircraft includes a fault-free state and a plurality of different control surface fault states, and each set of motion simulation data of the aircraft includes time series data of a plurality of motion parameters of the aircraft in the current flight state; Performing feature screening on the motion parameters to obtain feature parameters, the screened feature parameters including the flight state parameters; The longitudinal available overload theoretical model is used to calculate the longitudinal available overload under the flight state according to the characteristic parameter time series data under each flight state; and the difference between the longitudinal available overload under each flight state and the longitudinal available overload under a fault-free state is calculated to obtain the longitudinal available overload change under the flight state; Taking the characteristic parameter time series data and the flight state in each flight state as input and the longitudinal available overload change amount in the flight state as output, a longitudinal available overload change amount prediction model is obtained based on ensemble learning algorithm training; The longitudinal available overload theoretical model and the trained longitudinal available overload change prediction model are used to obtain the longitudinal available overload prediction result of the aircraft in a control surface failure state based on the actual motion data of the aircraft in an actual flight state.

2. The method for predicting the longitudinal available overload of an aircraft under fault conditions according to claim 1, characterized in that: The construction of the longitudinal available overload theoretical model of the aircraft comprises: Based on the kinematic model of the aircraft and the instantaneous balance assumption, a longitudinal overload model of the aircraft is constructed. The longitudinal overload model represents the longitudinal overload n of the aircraft. y Relationship with flight state parameters and rudder deflection angle; The rudder deflection angle in the longitudinal overload model is taken to an extreme position to obtain the longitudinal available overload theoretical model.

3. The method for predicting the longitudinal available overload of an aircraft under fault conditions according to claim 2, characterized in that: The obtained longitudinal available overload theoretical model is: Where Q is the dynamic pressure and ρ is the air density, V is the flight speed of the aircraft, S is the reference area of ​​the rudder, m is the mass of the aircraft, g is the acceleration of gravity, P is the thrust of the aircraft, C Y0 is the lift coefficient when the angle of attack and elevator deflection are zero, C Yα is the stability derivative of the lift coefficient with respect to the angle of attack, is the pitch moment coefficient when the angle of attack, sideslip angle and multiple deviation terms are zero, is the stability derivative of the pitching moment with respect to the rudder angle, is the stability derivative of the pitch moment coefficient with respect to the angle of attack; (δ1-δ2-δ3+δ4) limit It indicates that the rudder deflection angle δ1-δ2-δ3+δ4 is taken to the extreme position, δ1 is the rudder deflection angle of rudder No. 1 of the aircraft, δ2 is the rudder deflection angle of rudder No. 2 of the aircraft, δ3 is the rudder deflection angle of rudder No. 3 of the aircraft, δ4 is the rudder deflection angle of rudder No. 4 of the aircraft, and rudder No. 1, rudder No. 2, rudder No. 3, and rudder No. 4 are arranged in clockwise order from the upper left corner along the flight direction as viewed from the tail of the aircraft to form an X-shaped structure.

4. The method for predicting the longitudinal available overload of an aircraft under fault conditions according to claim 3, characterized in that: The constructing of the longitudinal overload model of the aircraft comprises: The longitudinal overload n is obtained according to the kinematic model of the aircraft. y The calculation formula is: Where α is the angle of attack, is the stability derivative of the lift coefficient with respect to the rudder deflection angle; Determined based on the instantaneous equilibrium assumption of the aircraft: in, is the pitching moment coefficient; Obtain the expression for the angle of attack α and substitute it into the longitudinal overload n y The calculation formula of the longitudinal overload model is:

5. The method for predicting the longitudinal available overload of an aircraft under fault conditions according to claim 4, characterized in that: The longitudinal overload n is obtained according to the kinematic model of the aircraft y The calculation formula includes: The lift coefficient C is obtained by fitting Y The expression is: According to the lift coefficient C Y The expression of the lift Y on the aircraft is obtained as follows: The resultant external force N other than gravity on the aircraft is obtained according to the lift Y on the aircraft. y The expression is: N y =Psinα+Y According to the resultant external force N y Get the longitudinal overload n y The expression is: Within the allowable error range, the sinα of the aircraft during cruising flight is equivalent to α, and the longitudinal overload n is obtained. y The calculation formula for .

6. The method for predicting the longitudinal available overload of an aircraft under fault conditions according to claim 1, characterized in that: The feature screening of the motion parameters to obtain the feature parameters comprises: Calculate the Spearman correlation coefficient between different motion parameters to perform correlation analysis, and eliminate one of the two motion parameters whose Spearman correlation coefficient reaches the correlation threshold; The importance of the remaining motion parameters is analyzed to obtain the importance result of each motion parameter, and the motion parameters whose importance results are lower than the importance threshold are eliminated, and the remaining motion parameters are regarded as feature parameters.

7. The method for predicting the longitudinal available overload of an aircraft under fault conditions according to claim 6, characterized in that: The importance of the remaining motion parameters is analyzed to obtain the importance of each motion parameter. For each of the remaining motion parameters: The importance sub-results of the motion parameters are calculated using a gradient boosting decision tree and a random forest algorithm respectively, and a weighted average is performed on the two importance sub-results to obtain the importance result of the motion parameters.

8. The method for predicting the longitudinal available overload of an aircraft under fault conditions according to claim 1, characterized in that: The longitudinal available overload variation prediction model obtained by training based on the ensemble learning algorithm includes: The model structure of the longitudinal available overload change prediction model is constructed based on an integrated learning algorithm, and the longitudinal available overload change prediction model is obtained by training based on the constructed model structure. The model structure includes a base learner and a meta learner connected in sequence, and the base learner is obtained by model fusion of multiple models.

9. The method for predicting the longitudinal available overload of an aircraft under fault conditions according to claim 8, characterized in that: The base learner in the model structure is obtained by fusing the extremely random tree model, the LightGBM model and the CatBoost model.

10. The method for predicting the longitudinal available overload of an aircraft under fault conditions according to claim 1, characterized in that: The obtaining of the longitudinal available overload prediction result of the aircraft when the aircraft is in a control surface failure state includes: Obtaining a reference longitudinal available overload prediction value of the aircraft according to actual motion data of the aircraft in a fault-free state using the longitudinal available overload theoretical model; Obtaining a predicted value of the longitudinal available overload change amount of the aircraft in the current control surface failure state by using the longitudinal available overload change amount prediction model according to actual motion data and flight state of the aircraft in the control surface failure state; The sum of the longitudinal available overload change prediction value and the reference longitudinal available overload prediction value is calculated to obtain the longitudinal available overload prediction result of the aircraft under the current control surface failure state.

Images