Fixed-wing aircraft pneumatic data fusion method and system based on artificial intelligence

Abstract

The invention relates to the technical field of data fusion, in particular to a fixed-wing aircraft pneumatic data fusion method and system based on artificial intelligence. The method comprises the following steps: acquiring low-precision pneumatic data, and constructing a low-precision pneumatic data set; constructing a low-precision aerodynamic model based on the low-precision aerodynamic data set; obtaining high-precision aerodynamic data, and constructing a multi-layer feedforward neural network model; performing normalization processing on input parameters and output parameters of the multi-layer feedforward neural network model; and training the constructed multi-layer feedforward neural network model based on the normalized input parameter and output parameter data to finally obtain an aerodynamic data fusion network model, and the aerodynamic data fusion network model is used for predicting aerodynamic characteristics of the fixed-wing aircraft in a full envelope range. According to the method, low-precision pneumatic data and high-precision test flight data are fused, the technical threshold is reduced, and the flexibility and accuracy of pneumatic data fusion are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data fusion, and more specifically, to an artificial intelligence-based fixed-wing aircraft aerodynamic data fusion method and system. Background Art

[0002] One of the core tasks of fixed-wing aircraft aerodynamic design is to provide an accurate and complete aerodynamic characteristics database. Wind tunnel tests, numerical calculations and flight tests are three means of obtaining aircraft aerodynamic data, each of which has its own characteristics and limitations:

[0003] Wind tunnel tests can simulate various flight environments, but due to differences in support interference, tunnel wall interference, and Reynolds number, there are certain differences between wind tunnel test data and actual flight data.

[0004] Numerical calculations are low-cost and can simulate various flight environments. However, because the turbulence model cannot fully simulate the actual flow state, the calculation results are less accurate.

[0005] The aerodynamic data obtained from flight tests are highly accurate, but the flight tests are costly, the test cycle is long, and the flight tests cannot obtain data within the complete envelope.

[0006] The above three methods have their own defects, so there is an urgent need to develop low-cost methods to obtain high-precision aerodynamic databases.

[0007] In order to make full use of the advantages of wind tunnel tests, fluid numerical calculations and flight tests for fixed-wing aircraft, avoid their respective shortcomings, and obtain accurate and complete aerodynamic data at low cost, domestic and foreign experts and scholars have proposed a variety of aerodynamic modeling methods based on variable credibility data fusion. The methods currently used mainly include correction models based on scaling functions and Kriging models.

[0008] In 1991, Haftka proposed a correction method for the multiplication scaling function, introducing a scaling coefficient based on the low-credibility model, which locally approximated the results of the high-credibility analysis; in 2000, Alexandrov proposed a first-order addition and multiplication scaling function method, combining the scaling function with the trust region method, and applied it to the optimization design of airfoils and wings, significantly reducing the computational cost. In 2012, Han Zhonghua proposed a method for implementing the Co-Kriging model, which can avoid complex cross-covariance calculations; in 2016, Han Zhonghua proposed a hierarchical Kriging (HK) model, which is more versatile and simpler in modeling. The basic idea of ​​the model is to establish a global trend model with a large amount of low-credibility data, and to correct it with a small amount of high-credibility samples, and finally to establish a high-precision prediction model.

[0009] Assume that the high and low precision data sets for aerodynamic modeling are

[0010]

[0011] In the formula, x is the model independent variable. The subscript "1" indicates high credibility, and "2" indicates low credibility; n 1 and n 2 is the number of high and low confidence sample points.

[0012] The corresponding response value is

[0013]

[0014] Where y is the model output. The subscript "1" indicates high confidence and "2" indicates low confidence; n 1 and n 2 is the number of high and low confidence sample points.

[0015] The core of the aerodynamic data fusion problem is to efficiently integrate aerodynamic data from different sources to obtain an accurate and complete aerodynamic database.

[0016] The variable credibility fusion model based on scaling function correction is an approximate model based on the low credibility model, which introduces the scaling function by addition, multiplication or a mixture of the two to build a high credibility model. The Kriging proxy model builds a proxy model based on low credibility sample data. Then thought Global trend model, build the required proxy model based on high-confidence sample data set

[0017] Although scaling function models and Kriging surrogate models have made significant progress in data fusion, they still have the following limitations:

[0018] First, the technical threshold is high. Scaling function models and Kriging proxy models usually need to be combined with aerodynamic laws or statistical analysis methods to determine the mathematical form of the model. This places high demands on the user's professional background and model building ability, increasing the complexity of engineering applications.

[0019] Secondly, the modeling flexibility is insufficient. The scaling function model and Kriging proxy model mostly adopt a fixed polynomial form, which makes it difficult to flexibly adapt to complex flow conditions and cannot perfectly fit the aerodynamic characteristics within the full envelope, which ultimately leads to certain errors in the fusion results.

[0020] In summary, traditional methods cannot fully utilize the advantages of various data sources in the fusion of different data sources, resulting in deviations in the final prediction results. Currently, there is an urgent need for a low-cost, high-precision aerodynamic data modeling method. Summary of the invention

[0021] The purpose of the present invention is to provide a method and system for aerodynamic data fusion of fixed-wing aircraft based on artificial intelligence, so as to solve the problems of high technical threshold and low precision of aerodynamic data fusion of fixed-wing aircraft in the prior art.

[0022] In order to achieve the above object, the present invention provides a fixed-wing aircraft aerodynamic data fusion method based on artificial intelligence, comprising the following steps:

[0023] Acquire low-precision aerodynamic data and construct a low-precision aerodynamic dataset;

[0024] Based on the low-precision aerodynamic data set, a low-precision aerodynamic model is constructed, wherein the low-precision aerodynamic model is used to output a low-precision aerodynamic coefficient prediction value;

[0025] Acquire high-precision aerodynamic data and construct a multi-layer feedforward neural network model, wherein the multi-layer feedforward neural network model is used to fuse low-precision aerodynamic coefficient prediction values ​​with high-precision aerodynamic data and map them through a neural network to obtain high-precision aerodynamic coefficient prediction values;

[0026] Normalizing the input parameters and output parameters of the multi-layer feedforward neural network model, wherein the input parameters include low-precision aerodynamic coefficient prediction values ​​and high-precision aerodynamic data, and the output parameters include high-precision aerodynamic coefficient prediction values;

[0027] The constructed multi-layer feedforward neural network model is trained based on the normalized input parameters and output parameter data, and finally an aerodynamic data fusion network model is obtained. The aerodynamic data fusion network model is used to predict the aerodynamic characteristics within the full envelope of the fixed-wing aircraft.

[0028] In some embodiments, the low-precision aerodynamic data is obtained through wind tunnel testing or CFD simulation calculation of a fixed-wing aircraft;

[0029] The high-precision aerodynamic data is test flight data.

[0030] In some embodiments, the low-precision aerodynamic model is an interpolation table model:

[0031] The interpolation table model includes a plurality of multi-dimensional interpolation tables, which are used to characterize the relationship between the aerodynamic coefficients of the aircraft and the flight motion state and the deflection angle of the control surface.

[0032] In some embodiments, the aircraft aerodynamic coefficients include the aircraft total lift coefficient:

[0033] The total lift coefficient of the aircraft is composed of the basic lift coefficient, the elevator lift coefficient increment and the dynamic reciprocal lift coefficient increment;

[0034] The basic lift coefficient is a two-dimensional interpolation table related to the flight Mach number and angle of attack of the aircraft;

[0035] The elevator lift coefficient increment is a three-dimensional interpolation table related to the flight Mach number, angle of attack and elevator deflection angle of the aircraft.

[0036] In some embodiments, the multi-layer feedforward neural network model includes a linear part sub-network and a nonlinear part sub-network:

[0037] The linear part sub-network adopts a single-layer neural network structure, each input node is directly connected to a neuron, outputs a single neuron result, and does not involve an activation function;

[0038] The nonlinear partial sub-network adopts a multi-layer neural network structure, each layer of the neural network contains multiple neurons, and processes nonlinear mapping through an activation function to output a single neuron result;

[0039] The output results of the linear part sub-network and the nonlinear part sub-network are weighted and summed to form a final fusion result.

[0040] In some embodiments, the aerodynamic data includes flight altitude, flight Mach number, angle of attack, rate of change of angle of attack, and control surface deflection angle;

[0041] The aerodynamic coefficients include lift coefficient, drag coefficient and pitch moment coefficient.

[0042] In some embodiments, the step of normalizing the input parameters and output parameters of the multi-layer feedforward neural network model further includes:

[0043] The min-max normalization method is used for normalization.

[0044] In some embodiments, the step of training the constructed multi-layer feedforward neural network model based on the normalized input parameter and output parameter data further includes:

[0045] The Adam optimization algorithm is used, and the average least squares loss function is used as the activation function to train the multi-layer feedforward neural network model, and the hyperparameters are adjusted during the training process.

[0046] In order to achieve the above object, the present invention provides an artificial intelligence-based fixed-wing aircraft aerodynamic data fusion system, including a memory and a processor:

[0047] The memory is used to store instructions executable by the processor;

[0048] The processor is used to execute the instructions to implement the above method.

[0049] In order to achieve the above object, the present invention provides a computer storage medium on which computer instructions are stored, wherein when the computer instructions are executed by a processor, the above method is executed.

[0050] The present invention provides an artificial intelligence-based fixed-wing aircraft aerodynamic data fusion method and system, which adopts a deep learning MFNN aerodynamic data fusion model to further lower the technical threshold and improve the flexibility and accuracy of aerodynamic data fusion by fusing low-precision aerodynamic data with high-precision test flight data, thereby meeting the demand for a full-envelope aerodynamic database for fixed-wing aircraft aerodynamic design and promoting the practical application of aerodynamic data fusion in engineering. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The above and other features, properties and advantages of the present invention will become more apparent through the following description in conjunction with the accompanying drawings and embodiments, in which the same reference numerals always represent the same features, wherein:

[0052] Figure 1 A step diagram of a method for aerodynamic data fusion of a fixed-wing aircraft based on artificial intelligence according to an embodiment of the present invention is disclosed;

[0053] Figure 2 A schematic diagram of an MFNN model structure according to an embodiment of the present invention is disclosed. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the invention and are not used to limit the invention.

[0055] In view of the shortcomings of traditional aerodynamic data fusion technology, the present invention proposes a fixed-wing aircraft aerodynamic data fusion method and system based on artificial intelligence. This method effectively reduces the technical threshold of aerodynamic data fusion by introducing artificial intelligence technology for aerodynamic data fusion, significantly improves the accuracy of aerodynamic data fusion results, and thus strongly promotes the widespread application of aerodynamic data fusion methods in the engineering field.

[0056] Figure 1 The following is a step diagram of a method for aerodynamic data fusion of a fixed-wing aircraft based on artificial intelligence according to an embodiment of the present invention. Figure 1 As shown, the present invention proposes an artificial intelligence-based fixed-wing aircraft aerodynamic data fusion method, comprising the following steps:

[0057] Step S1, acquiring low-precision aerodynamic data and constructing a low-precision aerodynamic data set;

[0058] Step S2, constructing a low-precision aerodynamic model based on the low-precision aerodynamic data set, wherein the low-precision aerodynamic model is used to output a low-precision aerodynamic coefficient prediction value;

[0059] Step S3, obtaining high-precision aerodynamic data and constructing a multi-layer feedforward neural network model, wherein the multi-layer feedforward neural network model is used to fuse the low-precision aerodynamic coefficient prediction value with the high-precision aerodynamic data and map them through a neural network to obtain the high-precision aerodynamic coefficient prediction value;

[0060] Step S4, normalizing the input parameters and output parameters of the multi-layer feedforward neural network model, wherein the input parameters include low-precision aerodynamic coefficient prediction values ​​and high-precision aerodynamic data, and the output parameters include high-precision aerodynamic coefficient prediction values;

[0061] Step S5, training the constructed multi-layer feedforward neural network model based on the normalized input parameters and output parameter data, and finally obtaining an aerodynamic data fusion network model, wherein the aerodynamic data fusion network model is used to predict the aerodynamic characteristics within the full envelope of the fixed-wing aircraft.

[0062] The present invention proposes a method for aerodynamic data fusion of fixed-wing aircraft based on artificial intelligence. It adopts deep learning neural network technology to establish a low-precision aerodynamic model with low credibility based on low-precision samples, and then uses the predicted value of the low-precision aerodynamic model and high-precision aerodynamic data as input of a high-credibility multi-layer feedforward neural network model for iterative training. The use of deep learning method for fixed-wing aircraft aerodynamic data fusion only requires adjusting the neural network hyperparameters, avoiding complex aerodynamic laws, and lowering the technical threshold of aerodynamic data fusion; at the same time, the neural network has the characteristic of fitting any nonlinear function, which can perfectly fit the aerodynamic characteristics of fixed-wing aircraft under different working conditions within the full envelope, and has high aerodynamic data fusion accuracy.

[0063] These steps will be described in detail below. It should be understood that within the scope of the present invention, the above-mentioned technical features of the present invention and the technical features specifically described below (such as embodiments) can be combined and related to each other to form a preferred technical solution.

[0064] Step S1, acquiring low-precision aerodynamic data and constructing a low-precision aerodynamic data set;

[0065] Obtain low-precision aerodynamic data of fixed-wing aircraft through wind tunnel tests or CFD calculations of fixed-wing aircraft.

[0066] These data describe the relationship between the aircraft's aerodynamic coefficients and flight conditions and control surface deflection angles.

[0067] Common aerodynamic coefficients such as flight altitude, flight Mach number, angle of attack, rate of change of angle of attack, and rudder angle usually involve multi-dimensional independent variables, such as Mach number, angle of attack, etc.

[0068] Step S2, constructing a low-precision aerodynamic model based on the low-precision aerodynamic data set, wherein the low-precision aerodynamic model is used to output a low-precision aerodynamic coefficient prediction value;

[0069] The acquired low-precision aerodynamic data is represented in the form of an interpolation table to construct an interpolation table model. The interpolation table model is used as a low-precision aerodynamic model to represent the relationship between the aircraft's aerodynamic coefficients and the flight motion state and the rudder surface deflection angle.

[0070] More specifically, the interpolation table model includes several multi-dimensional interpolation tables, which are divided into one-dimensional, two-dimensional and multi-dimensional interpolation tables, depending on the number of independent variables.

[0071] The aircraft aerodynamic coefficients include lift coefficient, drag coefficient, pitch moment coefficient, etc., and usually involve multi-dimensional independent variables, such as Mach number, angle of attack, etc.

[0072] In this embodiment, the total lift coefficient CL of the aircraft is the basic lift coefficient CL basic , elevator lift coefficient increment ΔCL ele , dynamic reciprocal lift coefficient increment ΔCL dynamics The corresponding expression is:

[0073] CL=CL basic +ΔCL ele +ΔCL dynamics +…;

[0074] Among them, the lift coefficient of the aircraft wing-body assembly is also called the basic lift coefficient CL basic , which is related to the aircraft's flight Mach number mach and angle of attack α;

[0075] More specifically, the basic lift coefficient CL basβc The interpolation table is a two-dimensional interpolation table with mach and α as independent variables, and the corresponding function expression is as follows:

[0076] CL basic =f(mach,α);

[0077] The lift coefficient increment ΔCL generated by the elevator deflection is ele , which is related to the aircraft's flight Mach number mach, angle of attack α and elevator deflection angle ele;

[0078] More specifically, the elevator lift coefficient increment ΔCL eleThe interpolation table is a three-dimensional interpolation table with mach, α and ele as independent variables. The corresponding function expression is as follows:

[0079] ΔCL ele =f(mach,α,ele).

[0080] Step S3, obtaining high-precision aerodynamic data and constructing a multi-layer feedforward neural network model, wherein the multi-layer feedforward neural network model is used to fuse the low-precision aerodynamic coefficient prediction value with the high-precision aerodynamic data and map them through a neural network to obtain the high-precision aerodynamic coefficient prediction value;

[0081] The multi-layer feedforward neural network (MFNN) model is used for aerodynamic data fusion. The multi-layer feedforward neural network (MFNN) model combines low-precision aerodynamic data with high-precision test flight data, maps them through neural networks, and obtains high-precision aerodynamic coefficient prediction values.

[0082] Assume that the relationship between high and low precision data can be expressed as:

[0083] y 1 =f(x,y 2 );

[0084] In the formula, f represents the mapping relationship of data from the low-precision layer to the high-precision layer, x is the input parameter, and y 2 Output results in low precision.

[0085] It is generally believed that there are linear and nonlinear correlations in both high and low precision data layers, so the mapping function f can be divided into two parts:

[0086] f=f l +f nl ;

[0087] Among them, f l is a linear mapping function, f n1 is a nonlinear mapping function;

[0088] In order to describe the degree of linear and nonlinear correlation between high-precision and low-precision data, a hyperparameter scaling factor p is introduced, so the high-precision aerodynamic coefficient estimation expression of the MFNN model is:

[0089]

[0090] Where p∈[0,1], x is the input parameter, It is an estimate of the low-precision output result.

[0091] The scaling factor p controls the contribution ratio of the linear part and the nonlinear part to the final fusion result. By adjusting p, the relationship between high-precision and low-precision data can be flexibly adapted.

[0092] Figure 2 A schematic diagram of the MFNN model structure according to an embodiment of the present invention is disclosed. Figure 2 As shown in the figure, x1, x2, x3, ..., xn on the left side represent n input parameters of high-precision test flight data. These input parameters include flight altitude, flight Mach number, angle of attack, rate of change of angle of attack, and rudder angle, etc. p is the hyperparameter scaling factor.

[0093] y1 is the high-precision aerodynamic coefficient calculated and output by the MFNN model, and y2 represents the low-precision aerodynamic coefficient obtained by the low-precision aerodynamic model with x1, x2, x3, ..., xn as input parameters.

[0094] like Figure 2 As shown, the MFNN model includes a linear partial sub-network and a nonlinear partial sub-network:

[0095] The linear part of the MFNN model is used to describe the linear correlation between high-precision and low-precision data. The linear part represents the linear mapping function of the model. A single-layer neural network structure is used, with n neurons, each input node is connected to a neuron, and finally a single neuron result is output without an activation function. The output result is weighted by a scaling factor p.

[0096] The nonlinear part of the MFNN model is used to describe the nonlinear correlation between high-precision and low-precision data. The nonlinear part represents the nonlinear mapping function of the model. A two-layer neural network is used, each layer of which contains multiple neurons, and an activation function (such as softpuls) is used to process complex nonlinear mappings, and finally a single neuron result is output. The final output result is weighted by a scaling factor of 1-p.

[0097] The output results of the linear part sub-network and the nonlinear part sub-network are weighted and summed to form the final fusion result y1.

[0098] Step S4, normalizing the input parameters and output parameters of the multi-layer feedforward neural network model, wherein the input parameters include low-precision aerodynamic coefficient prediction values ​​and high-precision aerodynamic data, and the output parameters include high-precision aerodynamic coefficient prediction values;

[0099] The input parameters of the MFNN model are high-precision aerodynamic data such as flight altitude, flight Mach number, angle of attack, rate of change of angle of attack from the test flight data, and low-precision aerodynamic coefficients such as lift coefficient, drag coefficient, and pitch moment coefficient output by the low-precision aerodynamic model;

[0100] The output parameters of the MFNN model are high-precision aerodynamic coefficients such as lift coefficient, drag coefficient, and pitch moment coefficient.

[0101] The purpose of normalization is to eliminate the influence of magnitude differences between different feature data on model convergence;

[0102] In this embodiment, the min-max normalization method is used to convert the data of different features into the [0,1] interval;

[0103] The formula for min-max normalization is as follows:

[0104]

[0105] In the formula, is the normalized data of each column feature, x is the original data of each column feature, and x min is the minimum value of each column of feature data, x max is the maximum value of each column of feature data.

[0106] After completing the normalization of the input and output parameters of the MFNN model, the dataset is divided into a training set and a validation set.

[0107] Step S5, training the constructed multi-layer feedforward neural network model based on the normalized input parameters and output parameter data, and finally obtaining an aerodynamic data fusion network model, wherein the aerodynamic data fusion network model is used to predict the aerodynamic characteristics within the full envelope of the fixed-wing aircraft.

[0108] The MFNN model is trained using normalized test flight data and low-precision aerodynamic model output data.

[0109] The Adam optimization algorithm is used and the mean least squares (MSE) loss function is adopted as the activation function to train the multi-layer feedforward neural network model. The hyperparameters (such as the scaling factor p) are adjusted during the training process.

[0110] The Adam optimization algorithm is an optimization method with adaptive learning rate, which can effectively deal with the gradient vanishing and gradient exploding problems. The MSE loss function is used for regression tasks to measure the gap between the predicted value and the actual value, which is defined as the average of the squares of the difference between the predicted value and the actual value.

[0111] The maximum number of training iterations is 500.

[0112] After training, an aerodynamic data fusion network model capable of fusing high- and low-precision data is obtained.

[0113] The aerodynamic data fusion network model is used to predict the aerodynamic characteristics of fixed-wing aircraft within the full envelope. Through the deep learning ability of the neural network, it can avoid complex aerodynamic calculations and directly drive modeling through data, thereby lowering the technical threshold. The deep learning model can fit any nonlinear function and better reflect the aerodynamic characteristics of fixed-wing aircraft under different working conditions.

[0114] The present invention also provides a fixed-wing aircraft pneumatic data fusion system based on artificial intelligence. The fixed-wing aircraft pneumatic data fusion system based on artificial intelligence includes a memory and a processor, the memory can store a program, and the processor can execute the program. When the processor executes the program, the fixed-wing aircraft pneumatic data fusion method based on artificial intelligence provided by the present invention is implemented.

[0115] The present invention also provides a computer-readable storage medium, which is a non-volatile storage medium or a non-transient storage medium, on which computer instructions are stored. When the computer instructions are executed, the steps corresponding to any of the above methods are executed, which will not be repeated here.

[0116] When the implementation process file of the fixed-wing aircraft aerodynamic data fusion method based on artificial intelligence is a computer program, it can also be stored in a computer-readable storage medium as a product. For example, a computer-readable storage medium may include, but is not limited to, a magnetic storage device (e.g., a hard disk, a floppy disk, a magnetic strip), an optical disk (e.g., a compact disk (CD), a digital versatile disk (DVD)), a smart card, and a flash memory device (e.g., an electrically erasable programmable read-only memory (EPROM), a card, a stick, a key drive). In addition, the various storage media described herein can represent one or more devices and / or other machine-readable media for storing information. The term "machine-readable medium" may include, but is not limited to, wireless channels and various other media (and / or storage media) that can store, contain and / or carry code and / or instructions and / or data.

[0117] The present invention proposes a method and system for aerodynamic data fusion of fixed-wing aircraft based on artificial intelligence, which specifically has the following beneficial effects:

[0118] 1) The MFNN model using deep learning technology is used for the fusion of fixed-wing aircraft aerodynamic data. It only needs to adjust the hyperparameters of the neural network, avoiding complex aerodynamic principles, thereby reducing the technical difficulty of aerodynamic data fusion;

[0119] 2) The neural network of the MFNN model can fit any nonlinear function and perfectly fit the aerodynamic characteristics of fixed-wing aircraft under different working conditions within the full envelope, ensuring a high aerodynamic data fusion accuracy;

[0120] 3) By combining the advantages of three data sources: wind tunnel tests, numerical calculations, and flight tests, high-precision data is used to correct low-precision data, achieving more accurate predictions of aerodynamic characteristics and significantly improving the accuracy of the model.

[0121] Although the above methods are illustrated and described as a series of actions for simplicity of explanation, it should be understood and appreciated that these methods are not limited by the order of the actions, because according to one or more embodiments, some actions may occur in a different order and / or concurrently with other actions from those illustrated and described herein or not illustrated and described herein but understandable to those skilled in the art.

[0122] As shown in this application and claims, unless the context clearly indicates an exception, the words "a", "an", "an" and / or "the" do not refer to the singular and may also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0123] Those skilled in the art will appreciate that information, signals, and data may be represented using any of a variety of different technologies and techniques. For example, data, instructions, commands, information, signals, bits, symbols, and chips cited throughout the above description may be represented by voltage, current, electromagnetic waves, magnetic fields or magnetic particles, optical fields or optical particles, or any combination thereof.

[0124] Those skilled in the art will further appreciate that the various illustrative logic blocks, modules, circuits, and algorithm steps described in conjunction with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or a combination of the two. To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, circuits, and steps are generally described above in terms of their functionality. Whether such functionality is implemented as hardware or software depends on the specific application and the design constraints imposed on the overall system. The technician may implement the described functionality in different ways for each specific application, but such implementation decisions should not be interpreted as resulting in a departure from the scope of the present invention.

[0125] The various illustrative logic modules and circuits described in conjunction with the embodiments disclosed herein may be implemented or executed with a general purpose processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. A general purpose processor may be a microprocessor, but in the alternative, the processor may be any conventional processor, controller, microcontroller, or state machine. The processor may also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in cooperation with a DSP core, or any other such configuration.

[0126] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. The software module may reside in a RAM memory, a flash memory, a ROM memory, an EPROM memory, an EEPROM memory, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to a processor so that the processor can read and write information from / to the storage medium. In an alternative, a storage medium may be integrated into a processor. The processor and the storage medium may reside in an ASIC. The ASIC may reside in a user terminal. In an alternative, the processor and the storage medium may reside in a user terminal as discrete components.

[0127] The above embodiments are provided for persons familiar with the art to implement or use the present invention. Personnel familiar with the art can make various modifications or changes to the above embodiments without departing from the inventive concept of the present invention. Therefore, the protection scope of the present invention is not limited to the above embodiments, but should be the maximum scope of the innovative features mentioned in the claims.

Claims

1. A fixed-wing aircraft aerodynamic data fusion method based on artificial intelligence, characterized in that: The following steps are involved: Acquire low-precision aerodynamic data and construct a low-precision aerodynamic dataset; Based on the low-precision aerodynamic data set, a low-precision aerodynamic model is constructed, wherein the low-precision aerodynamic model is used to output a low-precision aerodynamic coefficient prediction value; Acquire high-precision aerodynamic data and construct a multi-layer feedforward neural network model, wherein the multi-layer feedforward neural network model is used to fuse low-precision aerodynamic coefficient prediction values ​​with high-precision aerodynamic data and map them through a neural network to obtain high-precision aerodynamic coefficient prediction values; Normalizing the input parameters and output parameters of the multi-layer feedforward neural network model, wherein the input parameters include low-precision aerodynamic coefficient prediction values ​​and high-precision aerodynamic data, and the output parameters include high-precision aerodynamic coefficient prediction values; The constructed multi-layer feedforward neural network model is trained based on the normalized input parameters and output parameter data, and finally an aerodynamic data fusion network model is obtained. The aerodynamic data fusion network model is used to predict the aerodynamic characteristics within the full envelope of the fixed-wing aircraft.

2. The fixed-wing aircraft aerodynamic data fusion method based on artificial intelligence according to claim 1 is characterized in that: The low-precision aerodynamic data is obtained through wind tunnel tests or CFD simulation calculations of fixed-wing aircraft; The high-precision aerodynamic data is test flight data.

3. The fixed-wing aircraft aerodynamic data fusion method based on artificial intelligence according to claim 1, characterized in that: The low-precision aerodynamic model is an interpolation table model: The interpolation table model includes a plurality of multi-dimensional interpolation tables, which are used to characterize the relationship between the aerodynamic coefficients of the aircraft and the flight motion state and the deflection angle of the control surface.

4. The fixed-wing aircraft aerodynamic data fusion method based on artificial intelligence according to claim 3 is characterized in that: The aircraft aerodynamic coefficients include the aircraft total lift coefficient: The total lift coefficient of the aircraft is composed of the basic lift coefficient, the elevator lift coefficient increment and the dynamic reciprocal lift coefficient increment; The basic lift coefficient is a two-dimensional interpolation table related to the flight Mach number and angle of attack of the aircraft; The elevator lift coefficient increment is a three-dimensional interpolation table related to the flight Mach number, angle of attack and elevator deflection angle of the aircraft.

5. The fixed-wing aircraft aerodynamic data fusion method based on artificial intelligence according to claim 1, characterized in that: The multi-layer feedforward neural network model includes a linear part sub-network and a nonlinear part sub-network: The linear part sub-network adopts a single-layer neural network structure, each input node is directly connected to a neuron, outputs a single neuron result, and does not involve an activation function; The nonlinear partial sub-network adopts a multi-layer neural network structure, each layer of the neural network contains multiple neurons, and processes nonlinear mapping through an activation function to output a single neuron result; The output results of the linear part sub-network and the nonlinear part sub-network are weighted and summed to form a final fusion result.

6. The fixed-wing aircraft aerodynamic data fusion method based on artificial intelligence according to claim 1, characterized in that: The aerodynamic data include flight altitude, flight Mach number, angle of attack, rate of change of angle of attack and control surface deflection angle; The aerodynamic coefficients include lift coefficient, drag coefficient and pitch moment coefficient.

7. The fixed-wing aircraft aerodynamic data fusion method based on artificial intelligence according to claim 1, characterized in that: The step of normalizing the input parameters and output parameters of the multi-layer feedforward neural network model further includes: The min-max normalization method is used for normalization.

8. The fixed-wing aircraft aerodynamic data fusion method based on artificial intelligence according to claim 1, characterized in that: The step of training the constructed multi-layer feedforward neural network model based on the normalized input parameter and output parameter data further includes: The Adam optimization algorithm is used, and the average least squares loss function is adopted as the activation function to train the multi-layer feedforward neural network model, and the hyperparameters are adjusted during the training process.

9. An artificial intelligence-based fixed-wing aircraft aerodynamic data fusion system, characterized in that: Including memory and processor: The memory is used to store instructions executable by the processor; The processor is used to execute the instructions to implement the method according to any one of claims 1-8.

10. A computer storage medium having computer instructions stored thereon, wherein when the computer instructions are executed by a processor, the method according to any one of claims 1 to 8 is executed.

Images