A method for predicting landing loads on typical cross-sections of amphibious aircraft

Abstract

An embodiment of the present invention discloses a method for predicting the landing load of a typical cross-section of an amphibious aircraft. The typical landing cross-section is set as a binary wedge cross-section. The prediction method includes: Step 1: collecting the landing load of a preset number of discrete locations on the landing surface of the amphibious aircraft; Step 2: smoothing the data collected in Step 1 and selecting training samples and test samples from the processed data; wherein the test samples include training samples; Step 3: using a machine learning method to train the binary wedge cross-section using existing training samples to construct a load prediction model for predicting the pressure change trend of all surface points of the binary wedge cross-section. The technical solution provided by the embodiment of the present invention solves the problem that existing methods for predicting landing loads have difficulty in predicting landing loads for complex cross-sectional configurations and three-dimensional problems.

Description

Technical Field

[0001] The present application relates to, but is not limited to, the technical field of methods for predicting landing loads, and in particular to a method for predicting the landing loads of a typical cross-section of an amphibious aircraft. Background Art

[0002] Currently, the only existing method specifically used to predict the surface load of amphibious aircraft is Wagner's two-dimensional water impact theory.

[0003] The above existing method of predicting the landing load is only applicable to the load prediction of a simple amphibious aircraft with a typical cross section vertically entering the water. There is no mature landing load prediction technology for complex cross-sectional configurations and three-dimensional problems. Summary of the Invention

[0004] Purpose of the present invention: In order to solve the above technical problems, an embodiment of the present invention provides a method for predicting the landing surface load of a typical cross-section of an amphibious aircraft, so as to solve the problem that the existing methods for predicting the landing surface load are difficult to predict the landing surface load for complex cross-sectional configurations and three-dimensional problems.

[0005] Technical solution of the present invention: An embodiment of the present invention provides a method for predicting the landing surface load of a typical cross-section of an amphibious aircraft, wherein the typical landing cross-section of the amphibious aircraft is set as a binary wedge cross-section. The method for predicting the landing surface load includes:

[0006] Step 1: collecting the landing loads of a preset number of discrete locations on the landing surface of the amphibious aircraft;

[0007] Step 2: smoothing the data collected in step 1 and selecting training samples and test samples from the processed data; wherein the test samples include the training samples;

[0008] Step 3: Using a machine learning method, the existing training samples are used to train the model of the binary wedge section to construct a load prediction model, which is used to predict the pressure change trend of all surface points of the binary wedge section.

[0009] Optionally, in the above-mentioned method for predicting the landing surface load of a typical cross-section of an amphibious aircraft, step 3 includes:

[0010] A Gaussian process is used to establish a water impact load prediction model; combined with a multi-layer perception neural network model, the water impact load prediction model is fine-tuned, and then the pressure time series variation data at any point on the surface of a typical cross-section of an amphibious aircraft is obtained to complete the surface pressure variation trend of the typical cross-section of the entire amphibious aircraft when it enters the water.

[0011] Optionally, the above-mentioned method for predicting the landing load of a typical cross-section of an amphibious aircraft further includes:

[0012] Step A: parameterizing the physical model of a typical cross-section of an amphibious aircraft. The parameterization results include:

[0013] The transverse coordinates of each point in a typical cross section of an amphibious aircraft are:

[0014] X i =c0+(i-1)*c1;

[0015] Coordinates of any point on the typical cross-section surface of an amphibious aircraft (X i , Y i ) can be characterized by the ramp angle, that is, Y = tanα*X,

[0016] The ramp angle is the angle between the cross section of the binary wedge and the horizontal axis.

[0017] Optionally, in the above-mentioned method for predicting the landing surface load of a typical cross-section of an amphibious aircraft, step 1 includes:

[0018] Model tests or numerical simulations are used to obtain time series data of water pressure loads at finite points under different working conditions, with a sampling frequency of not less than 2500 Hz.

[0019] Optionally, in the above-mentioned method for predicting the landing surface load of a typical cross-section of an amphibious aircraft, step 3 includes:

[0020] Step 31, feature parameter extraction and selection: establishing a nonlinear mapping relationship between the ramp angle, water entry speed, horizontal axis coordinate of the measurement point, time and pressure value, using the ramp angle, water entry speed, horizontal axis coordinate of the measurement point, time and other water entry parameters as input parameters, and the pressure parameter as the output parameter;

[0021] Step 32, dividing the original problem into two regression fitting sub-problems;

[0022] Step 33: Establish a weak regression model of the Gaussian process, and use training samples to train the established weak regression model of the Gaussian process to obtain the corresponding relationship between the ramp angle, entry speed, lateral coordinate, time and pressure, and then obtain a load prediction model;

[0023] In step 34, a multi-layer perceptron (MLP) neural network is used based on a machine learning framework to fine-tune the load prediction model.

[0024] Optionally, in the above-mentioned method for predicting the landing load of a typical cross-section of an amphibious aircraft, the two regression fitting sub-problems divided in step 32 include:

[0025] Sub-problem 1: Regression fitting of water entry velocity, horizontal axis coordinate, time and pressure values ​​when the ramp angle is fixed;

[0026] Sub-problem 1: Regression fitting of ramp angle, horizontal axis coordinate, time and pressure values ​​when the entry speed is fixed.

[0027] Optionally, in the above-mentioned method for predicting the landing surface load of a typical cross-section of an amphibious aircraft, step 34 includes:

[0028] The load prediction model based on Gaussian process is used as a weak classifier to train a strong classifier, and a multi-layer perceptron (MLP) neural network is used to take the strong classifier as input to fine-tune the load prediction model and finally form a water load prediction model based on machine learning.

[0029] Optionally, in the above-mentioned method for predicting the landing load of a typical cross-section of an amphibious aircraft, in step 34, the process of constructing a landing load prediction model based on machine learning includes:

[0030] Three nonlinear weak learners, all of which are Gaussian regression models, and a three-layer neural network model are established, and the training samples are learned separately based on iterative algorithms. During the learning process, the parameters of the weak learners are first determined, and then the results of each base learner are predicted based on the base learner. The predicted results of the base learners are used as input, and the neural network is used to share weights.

[0031] Beneficial effects of the present invention: The method for predicting the landing surface load of a typical cross-section of an amphibious aircraft provided by an embodiment of the present invention can accurately predict the landing load at each point on the landing surface of the typical cross-section of the aircraft, providing more accurate load input for aircraft structural strength design. The landing load is a transient load with the characteristics of high peak value and short pulse width, which varies with spatial position and time. The prediction is difficult, and theoretical methods cannot meet the requirements of engineering applications. Experiments can only measure the load at local points, and the surface pressure predicted by simulation is insufficiently accurate. The prediction of surface load is one of the important factors currently limiting the development of high-performance amphibious aircraft. Specifically, it has the following beneficial effects:

[0032] First, a method for predicting the landing load of amphibious aircraft based on machine learning is proposed;

[0033] Second, the two weak classifiers, Gaussian process and neural network, are connected in series to achieve high-precision load prediction;

[0034] Third, the technical solution provided by the present invention is suitable for establishing a highly nonlinear water impact surface load prediction model, without the need to establish a complex combined kernel function;

[0035] Fourth, in traditional machine learning, kernel function selection is difficult and requires a lot of work. Serial machine learning avoids this complex kernel function selection process.

[0036] Fifth, the implementation method of the technical solution provided by the present invention is practical, feasible, and easy to implement, and the prediction results are reliable. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The accompanying drawings are used to provide a further understanding of the technical solution of the present invention and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the technical solution of the present invention and do not constitute a limitation on the technical solution of the present invention.

[0038] Figure 1 A flowchart of a method for predicting the landing load of a typical cross-section of an amphibious aircraft provided in an embodiment of the present invention;

[0039] Figure 2 A schematic diagram of a typical cross section of an amphibious aircraft;

[0040] Figure 3 Schematic diagram of original test data of water load in an embodiment of the present invention;

[0041] Figure 4 This is a schematic diagram showing the principle of converting the original problem into two sub-problems in an embodiment of the present invention;

[0042] Figure 5 Schematic diagram of the process of constructing a water load prediction model based on machine learning in an embodiment of the present invention;

[0043] Figure 6 It is a structural diagram of the neural network;

[0044] Figure 7 This is a comparison chart of the predicted landing load value and the actual value obtained by using the landing surface load prediction method provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0045] To make the purpose, technical solutions and advantages of the present invention more clearly understood, the embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be noted that, unless there is a conflict, the embodiments and features in the embodiments of the present application can be combined with each other in any manner.

[0046] As explained in the above background technology, the existing method for predicting landing loads is only applicable to the load prediction of a simple amphibious aircraft with a typical cross-section entering the water vertically. There is no mature landing load prediction technology for complex cross-sectional configurations and three-dimensional problems.

[0047] In response to the problems existing in the above-mentioned existing methods for predicting landing surface loads, an embodiment of the present invention provides a method for predicting the landing surface loads of a typical cross-section of an amphibious aircraft. This method can accurately predict the landing loads at various points on the landing surface of a typical cross-section of the aircraft, providing more accurate load input for the aircraft structural strength design. The landing load is a transient load with the characteristics of a high peak value and a short pulse width. It changes with spatial position and time, making it difficult to predict. Theoretical methods cannot meet the requirements of engineering applications, and experiments can only measure the loads at local points. The surface pressure predicted by simulation is insufficiently accurate. The prediction of surface loads is currently one of the important factors restricting the development of high-performance amphibious aircraft.

[0048] The present invention provides the following specific embodiments that can be combined with each other. The same or similar concepts or processes may not be described in detail in some embodiments.

[0049] The technical solution of the method for predicting landing surface load provided by the embodiment of the present invention mainly includes the following implementation schemes:

[0050] Step 1, using the collected data of the landing load of the amphibious aircraft hull at a certain number of discrete positions;

[0051] Step 2: Analyze the data patterns and smooth the data collected in step 1, and select training samples and test samples from the processed data; wherein the test samples include the training samples;

[0052] In step 3, a machine model was designed based on the machine learning framework to gradually understand the physical laws of the water surface flow field. The existing training samples were used to train the model of the binary wedge section to construct a load prediction model, which is used to predict the pressure change trend of all surface points of the binary wedge section.

[0053] In step 3, the load prediction model is built as follows: first, a Gaussian process (GP) is used to establish a water load prediction model, and then a multilayer perceptron (MLP) neural network model is combined to fine-tune the load prediction model. This allows the pressure time series variation data at any point on the surface of a typical cross-section of an amphibious aircraft to be obtained, thus completing the surface pressure variation trend of the typical cross-section of the entire amphibious aircraft when entering water.

[0054] The method for predicting the landing surface load of a typical cross-section of an amphibious aircraft provided in an embodiment of the present invention is specifically a technology for constructing and testing a model for predicting the landing surface load of a typical cross-section of an amphibious aircraft based on machine learning. It can be used to predict the landing surface load of models such as amphibious aircraft, ground effect aircraft, cross-medium aircraft, and forced landings on land-based aircraft.

[0055] Figure 1This is a flow chart of a method for predicting the landing load of a typical cross-section of an amphibious aircraft provided in an embodiment of the present invention. The specific implementation of the method includes the following steps:

[0056] (1) Parameterization of physical model:

[0057] Figure 2 This is a schematic diagram of a typical cross-section of an amphibious aircraft. A coordinate system XOY is defined, with the X-axis horizontally pointing rightward and the Y-axis vertically pointing upward. The coordinates of a point on the surface of the typical cross-section of an amphibious aircraft are calculated based on the distance between the point and the vertex along the hypotenuse of the bottom of the cross-section. In a specific implementation, the vertex at the bottom of the cross-section of the amphibious aircraft is set as the coordinate origin. Pressure measurement points are sequentially set along the surface of the cross-section of the amphibious aircraft toward the right. The distance between the first pressure measurement point and the vertex is c0, and the distance between the remaining points is c1. The formula for calculating the lateral coordinate of the i-th point is:

[0058] X i =c0+(i-1)*c1 (1)

[0059] Coordinates of any point on the typical cross-section surface of an amphibious aircraft (X i , Y i ) can be characterized by the ramp angle α, that is, Y = tanα*X, where the ramp angle α is defined as the angle between the wedge surface and the horizontal axis.

[0060] (2) Data collection:

[0061] The data collection adopts the principle of single parameter change, and the simulation working condition is designed. The changing parameters are the water entry speed and the hull configuration (the inclined angle of the binary wedge section). The time series data of the water pressure load at the finite point under different working conditions are obtained by model test or numerical simulation. The sampling frequency is not less than 2500 Hz. Figure 3 , which is a schematic diagram of the original test data of the water load in an embodiment of the present invention.

[0062] (3) Data processing:

[0063] The collected data is affected by noise and may fluctuate slightly. A smoothing method is used to process the water load data, such as removing individual points. Training samples and test samples are selected from the processed data to form a water load sample data set.

[0064] (IV) Building a load prediction model based on machine learning:

[0065] (1) Feature parameter extraction and selection

[0066] The input parameters of the impact load prediction model are key test conditions that influence the test results. Therefore, input parameters are crucial. Too few input parameters may not fully represent the load prediction model, while too many parameters may lead to overconstraint. Improper input parameter selection can significantly affect the model's precision and accuracy.

[0067] For the water entry impact test of a typical cross-section of an amphibious aircraft, since the entry of a typical cross-section of an amphibious aircraft into water is a time-series process, and the goal is to obtain the pressure change at any point on the surface of the typical cross-section of an amphibious aircraft, a nonlinear mapping relationship is established between the ramp angle, water entry speed, horizontal axis coordinate of the measurement point, time and pressure value. The water entry parameters such as the ramp angle, water entry speed, horizontal axis coordinate of the measurement point, and time are used as input parameters, and the pressure parameters are used as output parameters.

[0068] (2) Division of the original problem

[0069] According to the composition of the sample data set, the original problem is divided into two sub-problems: 1. The regression fitting sub-problem of the water entry speed, horizontal axis coordinate, time and pressure value when the ramp angle is fixed; 2. The regression fitting sub-problem of the water entry speed, horizontal axis coordinate, time and pressure value when the ramp angle is fixed. Figure 4 FIG. 1 is a schematic diagram showing the principle of converting an original problem into two sub-problems in an embodiment of the present invention.

[0070] (3) Establishing a weak regression model of Gaussian process

[0071] For the individual sub-problems obtained by the above division, we first use the characteristics of the Gaussian process model that is suitable for dealing with complex problems such as high dimensions, small samples and nonlinearity. The falling state of a typical cross-section of an amphibious aircraft and the corresponding pressure time series data when impacting the water are used as experimental samples for regularity learning, capturing the time series change pattern of the surface point pressure values ​​when the typical cross-section of an amphibious aircraft enters the water.

[0072] A Gaussian process weak regression model was established, with the Gaussian kernel function selected as the covariance function. Based on the maximum likelihood estimation of the hyperparameters, the gradient descent method was used to calculate the hyperparameters. These hyperparameters were then substituted into the Gaussian kernel function to obtain the kernel function's value. The established Gaussian process weak regression model was trained using training samples to obtain the corresponding relationships between the ramp angle, entry velocity, lateral coordinates, time, and pressure, thereby developing a load prediction model.

[0073] The above process of establishing a weak regression model of Gaussian process is:

[0074] Assume that the training sample set D = {(x i ,y i )|i=1,2,......,n} obeys Gaussian random process:

[0075] yi =f(x i )+ε i ; (2)

[0076] Among them, x i Input vector for the i-th sample point; y i is an output scalar used to represent the response value of the i-th sample; n represents the number of training sample data; f(x i ) is x i The distribution function of ε i is Gaussian noise, which is a Gaussian process with an independent and identical distribution and a mean function of 0, that is, is the variance of the Gaussian distribution, which means creating a local deviation from the global model with zero mean but non-zero variance.

[0077] f(x i ) is recorded as f, and from the properties of Gaussian process we know that:

[0078]

[0079] Where: cov(f,f) is an n×n order symmetric positive definite covariance matrix, representing the covariance function, denoted by K(X,X); X represents the feature matrix composed of all the water-falling state input vectors in the training sample data, X=[x1,x2……x n ] T ; y represents the column vector composed of all pressure response values ​​in the training sample data, y=[y1,y2,......,y n ] T ; I is the n-order identity matrix.

[0080] The prediction formula for the impact pressure response based on the prediction model is obtained as follows:

[0081]

[0082] Among them, x * represents a water-falling state input vector in the test sample data; f * Indicates the corresponding pressure prediction value; K(x * ,X)=K(x * ,X) T is the 1×n order covariance matrix between the test sample and the training set; K(x * ,x * ) is the covariance matrix between test samples.

[0083] According to formula (4), and through certain matrix operations, the predicted value f can be obtained * The conditional probability distribution of is:

[0084]

[0085] in,

[0086]

[0087] cov(f * )=K(x * ,x * )-K(x * ,X)×[K(X,X)+I] -1 K(X,x * ); (7)

[0088] Finally, formula (6) is used as the predicted value of the test sample, and formula (7) represents the variance of the prediction, which is used to measure the uncertainty of the prediction.

[0089] The Gaussian kernel function is as follows:

[0090]

[0091] Where x p 、x q is the input, σ f and λ are collectively referred to as the hyperparameters of the kernel function, denoted as θ, which determines the value of the kernel function. When the input X and output y are given, according to Bayes' theorem, we can get:

[0092]

[0093] The maximum likelihood function p(y|X,θ) of the hyperparameter is expressed as follows:

[0094]

[0095] After setting the initial values ​​of the hyperparameters, the conjugate gradient descent method, Newton's method and other optimization methods can be used to iterate the hyperparameters to obtain the optimal solution of the hyperparameters. The objective function is differentiated with respect to each parameter, and the function gradient value is continuously iterated so that it gradually decreases until it converges. Finally, the converged hyperparameters are solved and substituted into formulas (6) and (7), thereby establishing a weak regression model of the Gaussian process.

[0096] (4) Fine-tune the load prediction model using an MLP neural network based on a machine learning framework;

[0097] The load prediction model based on Gaussian process is used as a weak classifier, and the bagging method is used to train a strong classifier. Finally, a multi-layer perceptron (MLP) neural network is used to take the strong classifier as input to fine-tune the load prediction model and finally form a training model to improve the processing capability and accuracy of the nonlinear problem of the load prediction model.

[0098] like Figure 5 FIG. 1 is a schematic diagram of the process of constructing a water load prediction model based on machine learning in an embodiment of the present invention. The specific modeling process is as follows:

[0099] Three nonlinear weak learners, all based on Gaussian regression models, and a three-layer neural network model were established. The training samples were learned separately based on an iterative algorithm. The weak learner parameters were first determined, and then the results of each base learner were predicted based on the base learner. The base learner predictions were then used as input, and the neural network shared weights.

[0100] Collect the data generated during the test, form a training sample, and based on the training sample falling water state data X=[x1,x2……x n ] T , water pressure response data y=[y1,y2,......,y n ] T , use the bagging algorithm to repeatedly sample the training sample data m times to form a new training sample set X m =[x1, x2, ...x m ] T , 3 samples are taken each time, that is, x i =[x i1 , x i2 , x i3 ] T (i=1,2,3…,m), use the base learner to train the new training set and get 3 sub-models. Calculate the covariance function parameters θ=[λ,σ f , σ n ], determine the covariance function K(X, x * ), after the base model training is completed, the prediction results are compared with the mean absolute percentage error (MAPE) of the test data using the test data, and the parameters are updated to reduce the error value to less than 1×10 -6 .

[0101] like Figure 6 As shown in the figure, it is a structural diagram of a neural network. The neural network includes an input layer, a hidden layer, and an output layer. Figure 6 It can be seen that the number of nodes in the network input layer is n=3, the number of hidden layer nodes are H1=3 and H2=3 respectively, and the number of output layer nodes is 1.

[0102] The output of the first hidden layer is where z 1,j The input of (j=1,2,3…,H1) is X=[x1,x2……x n ] T

[0103]

[0104] In the formula, f1 and are the activation function and weight of the first hidden layer respectively.

[0105] The output of the second hidden layer is where z 2,j The input of (j=1,2,3…,H2) is the output of the first hidden layer node

[0106]

[0107] In the formula, f2 and are the activation function and weight of the first hidden layer respectively.

[0108] The network output is:

[0109]

[0110] Neural network parameter settings: the hidden layer activation function selects the rectified linear unit function (Rectified Linear Unit, ReLU), ReLU (x) = max (0, x); the training error mean square error is 0.001; the global initial rate Delay rate ρ = 0.88, learning rate 0.0001. i1 , x i2 , x i3 As input, initialize the weight w ij (1) =w ij (2) =w ij (3) = 0.1. The Gaussian regression model is integrated to obtain the input of the neural network.

[0111] In addition to the input nodes, the hidden layer nodes and output nodes contain activation functions. Usually, when a fully connected neural network is used for function approximation, f1 and f2 are nonlinear functions, and f3 is a linear function.

[0112] f1(x)=f2(x)=ReLU(x) (14)

[0113] f3(x)=x (15)

[0114] The multilayer perceptron (MLP) is the most typical artificial neural network model. This embodiment of the present invention uses the MLP and ReLU activation function to further fit the highly nonlinear mapping relationship in the original problem based on the temporal dependence of the Gaussian process model output, combined with the coordinate information and time information of the surface points of a typical cross-section of an amphibious aircraft, to obtain the final pressure change curve.

[0115] The method for predicting the landing load of a typical cross-section of an amphibious aircraft provided by an embodiment of the present invention has the following beneficial effects:

[0116] First, a method for predicting the landing load of amphibious aircraft based on machine learning is proposed;

[0117] Second, the two weak classifiers, Gaussian process and neural network, are connected in series to achieve high-precision load prediction;

[0118] Third, the technical solution provided by the present invention is suitable for establishing a highly nonlinear water impact surface load prediction model, without the need to establish a complex combined kernel function;

[0119] Fourth, in traditional machine learning, kernel function selection is difficult and requires a lot of work. Serial machine learning avoids this complex kernel function selection process.

[0120] Fifth, the implementation method of the technical solution provided by the present invention is practical, feasible, and easy to implement, and the prediction results are reliable.

[0121] In addition, the method for predicting the landing load of a typical cross-section of an amphibious aircraft based on machine learning provided by an embodiment of the present invention has been applied to the research on the surface load prediction and verification technology of XXX aircraft, proving the effectiveness of the method.

[0122] like Figure 7 , which is a comparison diagram of the predicted landing load value and the actual value obtained by using the landing surface load prediction method provided by an embodiment of the present invention.

[0123] Although the embodiments disclosed herein are as described above, the contents described herein are merely embodiments for facilitating understanding of the present invention and are not intended to limit the present invention. Any person skilled in the art may make any modifications and variations in the form and details of the embodiments without departing from the spirit and scope of the present invention. However, the scope of patent protection of the present invention shall remain subject to the scope defined by the appended claims.

Claims

1. A method for predicting the landing load of a typical cross-section of an amphibious aircraft, characterized in that: The typical landing section of the amphibious aircraft is set as a two-dimensional wedge section. The parameterized expression of the physical model of the typical section of the amphibious aircraft is: the transverse coordinates of each measuring point in the typical section of the amphibious aircraft are X i = c0 + (i-1) * c1, c0 is the distance between the first pressure measurement point set to the right of the bottom vertex of the typical cross-section of an amphibious aircraft and the vertex, and c1 is the distance between two adjacent measurement points; The coordinates of any point on the typical cross-section surface of the amphibious aircraft (X i , Y i ) is characterized by a ramp angle α as Y=tanα*X, where the ramp angle α is defined as the angle between the wedge surface and the horizontal axis; The landing surface load prediction method comprises: Step 1: collecting the landing loads of a preset number of discrete locations on the landing surface of the amphibious aircraft; Step 2: smoothing the data collected in step 1 and selecting training samples and test samples from the processed data; wherein the test samples include the training samples; Step 3: Using a machine learning approach, the existing training samples are used to train the model of the binary wedge section to construct a load prediction model, which is used to predict the pressure change trend of all surface points of the binary wedge section; In step 1, the impact load is collected by designing a simulated working condition based on the principle of a single parameter change. The changed parameter is the water entry speed or the ramp angle of the binary wedge cross section, and time series data of the impact pressure load at a finite point under different working conditions are collected; The step 3 comprises: Step 31, feature parameter extraction and selection: establishing a nonlinear mapping relationship between the ramp angle, water entry speed, horizontal axis coordinate of the measurement point, time and pressure value, using the ramp angle, water entry speed, horizontal axis coordinate of the measurement point, time as input parameters and the pressure parameter as output parameter; Step 32: Divide the original problem into two serial regression fitting sub-problems: Sub-problem 1: regression fitting of water entry velocity, horizontal axis coordinate, time, and pressure values ​​when the ramp angle is fixed; Sub-problem 2: regression fitting of water entry velocity, horizontal axis coordinate, time, and pressure values ​​when the ramp angle is fixed; Step 33: Establish a weak regression model of the Gaussian process, and use training samples to train the established weak regression model of the Gaussian process to obtain the corresponding relationship between the ramp angle, entry speed, lateral coordinate, time and pressure, and then obtain a load prediction model; In step 34, a multi-layer perceptron (MLP) neural network is used based on a machine learning framework to fine-tune the load prediction model.

2. The method for predicting the landing load of a typical cross-section of an amphibious aircraft according to claim 1, characterized in that: The step 3 comprises: A Gaussian process is used to establish a water impact load prediction model; combined with a multi-layer perception neural network model, the water impact load prediction model is fine-tuned, and then the pressure time series variation data at any point on the surface of a typical cross-section of an amphibious aircraft is obtained to complete the surface pressure variation trend of the typical cross-section of the entire amphibious aircraft when it enters the water.

3. The method for predicting the landing load of a typical cross-section of an amphibious aircraft according to claim 2, characterized in that: The step 1 comprises: Model tests or numerical simulations are used to obtain time series data of water pressure loads at finite points under different working conditions, with a sampling frequency of not less than 2500 Hz.

4. The method for predicting the landing load of a typical cross-section of an amphibious aircraft according to claim 3, characterized in that: The step 34 includes: The load prediction model based on Gaussian process is used as a weak classifier to train a strong classifier, and a multi-layer perceptron (MLP) neural network is used to take the strong classifier as input to fine-tune the load prediction model and finally form a water load prediction model based on machine learning.

5. The method for predicting the landing load of a typical cross-section of an amphibious aircraft according to claim 4, characterized in that: In step 34, the process of constructing the water load prediction model based on machine learning includes: Three nonlinear weak learners, all of which are Gaussian regression models, and a three-layer neural network model are established, and the training samples are learned separately based on iterative algorithms. During the learning process, the parameters of the weak learners are first determined, and then the results of each base learner are predicted based on the base learner. The predicted results of the base learners are used as input, and the neural network is used to share weights.

Images