FADS system atmospheric parameter resolving method based on geometric feature fitting neural network

Abstract

The invention discloses an FADS system atmospheric parameter resolving method based on a geometric feature fitting neural network. The method comprises the following steps: step 1, acquiring a surface pressure matrix ptrain and a reference atmospheric data vector required for training the neural network by using a wind tunnel and aircraft equal-proportion scaling model; 2, constructing a neural network training input matrix gamma train; step 3, calculating a reference pressure coefficient matrix Cp, reference; 4, constructing a geometric feature fitting neural network; 5, completing geometric feature fitting neural network training; step 6, acquiring an actual surface pressure vector # imgabs0 # at the current moment by utilizing a pressure measuring hole in the front end of an actual aircraft; step 7, constructing an actual network input vector # imgabs1 # at the current moment; step 8, acquiring a fitting attack angle # imgabs2 # sideslip angle # imgabs3 # and a fitting pressure coefficient vector # imgabs4 # at the current moment; step 9, calculating a dynamic pressure # imgabs5 # and a static pressure # imgabs6 # at the current moment; and step 11, calculating the current flight Mach number. According to the method, relatively low atmospheric data fitting errors can be maintained.

Description

Technical Field

[0001] The invention belongs to the technical field of atmospheric parameter calculation, and in particular relates to an atmospheric parameter calculation method of a FADS system based on a geometric feature fitting neural network. Background Art

[0002] The atmospheric parameter calculation method of the flush air data sensing (FADS) system is to use the aircraft surface pressure information and surface pressure distribution model to measure and sense flight atmospheric parameters such as angle of attack, sideslip angle, total static pressure and Mach number. This method uses pressure measuring holes embedded in the fuselage to replace the invasive pitot tube to realize the embedded atmospheric data system with good stealth performance, strong environmental adaptability and relatively high accuracy. Therefore, it has been widely used in various advanced aircraft such as stealth aircraft, hypersonic aircraft, and space shuttle aircraft.

[0003] The core of the FADS system atmospheric parameter solution method is the surface pressure distribution model, which has a strong correlation with the geometric characteristics of the area where the pressure measuring holes are arranged. Early FADS systems were mostly used in aircraft with a nose cone front end. The surface pressure distribution model of the nose cone with pressure measuring holes can be established based on potential flow theory, and based on this, an analytical FADS system atmospheric parameter solution method was constructed.

[0004] With the development of aerospace technology and the needs of application scenarios, modern advanced aircraft mostly adopt complex configurations such as wing-body fusion design and pointed wedge forebody, and its surface pressure distribution model is difficult to establish. On the other hand, the installation of various airborne sensor equipment also imposes more restrictions and constraints on the layout of the pressure measuring holes of the FADS system. Therefore, the traditional analytical FADS system atmospheric parameter solution method is no longer applicable.

[0005] Artificial neural networks have been widely used and studied in the atmospheric parameter solution method of FADS system because of their strong nonlinear mapping ability. Rholff et al. from the University of California first cooperated with NASA in 1998 to build two neural networks based on the measurement data of F-18SRA, which were used to fit the dynamic pressure and static pressure respectively, ignoring the influence of angle of attack and sideslip angle. With the development of sensor technology and neural network technology, various types of neural networks are used to design different method architectures to solve specific problems in different scenarios. In 2022, X.Na et al. used neural networks to construct the mapping relationship between the measurement information of flexible sensor array and angle of attack and airspeed; in 2021, Q.Jia et al. compared a variety of FADS solution algorithms including neural network methods, and pointed out that although the FADS system algorithm based on neural network can obtain relatively accurate results, it requires a large amount of data for training, and the network training is difficult to converge. With the deepening of research, some scholars have found that establishing multiple neural networks to fit different types of atmospheric parameters can effectively improve the fitting accuracy of atmospheric data. In 2022, L. Yang and others from Northwestern Polytechnical University proposed a FADS perception method based on a dimensionless input-output neural network to improve the accuracy of atmospheric data fitting under the conditions of fewer training samples.

[0006] As a data-driven algorithm, the performance and fitting accuracy of neural networks depend largely on whether the training set can support the knowledge learned by the network to cover all possible situations in actual application scenarios. Therefore, using measured data to construct a training data set is an important means to ensure the consistency between the knowledge learned by the network and the actual application scenarios. In the field of aerospace, wind tunnel tests and flight tests are the main means of obtaining this type of data. However, due to cost and safety considerations, the data obtained from wind tunnel tests and flight tests are relatively limited, and the vast majority of existing FADS system atmospheric parameter solution methods based on neural networks may be difficult to meet the needs of their network training. On the other hand, limited by the black box characteristics of neural networks, the interpretability of neural network-based methods has always been an important factor restricting the application of such methods in practice. Summary of the invention

[0007] The object of the present invention is to provide a method for calculating atmospheric parameters of a FADS system based on a geometric feature fitting neural network.

[0008] In order to solve the above technical problems, the technical solution of the present invention is:

[0009] A method for calculating atmospheric parameters of a FADS system based on a geometric feature fitting neural network comprises the following steps:

[0010] Step 1: Use the wind tunnel and aircraft scale model to obtain the surface pressure matrix p required for training the neural network train and a reference atmospheric data vector, the reference atmospheric data vector including a reference angle of attack vector Reference sideslip angle vector Reference static pressure vector and the reference dynamic pressure vector

[0011] Step 2: Based on the surface pressure matrix p train Construct neural network training input matrix Γ train ;

[0012] Step 3: Based on the surface pressure matrix p train , reference static pressure vector and the reference dynamic pressure vector Calculate the reference pressure coefficient matrix C p,reference ;

[0013] Step 4: Construct a geometric feature fitting neural network;

[0014] Step 5: Train the neural network with the input matrix Γ train As the input of the geometric feature fitting neural network training, the reference angle of attack vector Reference sideslip angle vector and the reference pressure coefficient matrix C p,reference As the output of the geometric feature fitting neural network training, the geometric feature fitting neural network training is completed;

[0015] Step 6: Use the pressure measuring hole at the front end of the actual aircraft to obtain the actual surface pressure vector at the current moment

[0016] Step 7: Construct the actual network input vector at the current moment

[0017] Step 8: Use the actual network input vector at the current moment The geometric feature fitting neural network trained in step 5 is used to obtain the current fitting angle of attack Side slip angle and the fitted pressure coefficient vector

[0018] Step 9: Based on the actual surface pressure vector at the current moment And the current moment fitting pressure coefficient vector output by the neural network Calculate the current dynamic pressure and static pressure

[0019] Step 10: Based on the calculated dynamic pressure at the current moment and static pressure Determine the current flight status;

[0020] Step 11: Calculate the current flight Mach number based on the determined flight status.

[0021] Compared with the prior art, the technical effects achieved by the present invention are as follows:

[0022] 1. Theoretical analysis of the FADS system atmospheric parameter solution method based on the geometric feature fitting neural network proposed in the present invention shows that the geometric feature fitting neural network adopted by the present invention can suppress the influence of other atmospheric data such as Mach number and total static pressure parameters to the greatest extent when fitting the airflow angle, so that it is only related to the geometric characteristics of the aircraft under certain conditions, thereby enhancing the extrapolation ability of the neural network.

[0023] 2. The geometric feature fitting neural network proposed in the present invention classifies the hidden layer neurons according to the actual physical characteristics of different network outputs, and adopts a non-fully connected architecture to form two relatively independent sub-networks, thereby reducing the complexity of the fitting objectives of each sub-network and being able to still meet the expected training objectives under limited training sample conditions.

[0024] 3. By decoupling the airflow angle parameters and total static pressure parameters in the surface pressure distribution model, the influence of the black box effect of the neural network on the atmospheric parameter solution algorithm is reduced. On the other hand, based on the analysis of the physical characteristics of the geometric feature fitting neural network input, the network fitting target is linked to the actual physical object, which improves the interpretability of the neural network and thus enhances the overall interpretability of the method.

[0025] 4. The verification results of wind tunnel test data show that compared with the traditional FADS system atmospheric parameter solution method based on neural network, the method proposed in the present invention has higher atmospheric parameter solution accuracy and better stability under the condition of limited training data, and shows better solution accuracy and algorithm stability under unfavorable conditions such as sparse feature points and network extrapolation. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 Schematic diagram of the neural network architecture for geometric feature fitting.

[0027] Figure 2 Schematic diagram of the distribution of pressure measuring holes in the proportional model of wind tunnel test.

[0028] Figure 3 This is the angle of attack fitting error distribution diagram in normal state.

[0029] Figure 4 This is the sideslip angle fitting error distribution diagram in normal state.

[0030] Figure 5This is the error distribution diagram of dynamic pressure solution in normal state.

[0031] Figure 6 This is the static pressure solution error distribution diagram under normal conditions.

[0032] Figure 7 This is the error distribution diagram of the Mach number solution in normal state.

[0033] Figure 8 This is the angle of attack fitting error distribution diagram in the feature point sparse state.

[0034] Fig. 9 It is the distribution diagram of the sideslip angle fitting error in the state of sparse feature points.

[0035] Fig.10 This is the error distribution diagram of dynamic pressure solution in the sparse state of feature points.

[0036] Fig.11 This is the error distribution diagram of sparse static pressure solution of feature points.

[0037] Fig.12 This is the error distribution diagram of Mach number solution in the sparse state of feature points.

[0038] Fig.13 This is the angle of attack fitting error distribution diagram of the network extrapolation state.

[0039] Fig.14 It is the distribution diagram of the sideslip angle fitting error in the network extrapolation state.

[0040] Fig.15 This is the error distribution diagram of dynamic pressure solution in the network extrapolation state.

[0041] Fig.16 This is the error distribution diagram of static pressure solution in network extrapolation state.

[0042] Fig.17 This is the error distribution diagram of the network extrapolated state Mach number solution. DETAILED DESCRIPTION

[0043] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0044] A method for calculating atmospheric parameters of a FADS system based on a geometric feature fitting neural network comprises the following steps:

[0045] Step 1: Use the wind tunnel and aircraft scale model to obtain the surface pressure matrix p required for training the neural network train and a reference atmospheric data vector, the reference atmospheric data vector including a reference angle of attack vector Reference sideslip angle vector Reference static pressure vector and the reference dynamic pressure vector

[0046] Step 2: Based on the surface pressure matrix p train Construct neural network training input matrix Γ train ;

[0047] Step 3: Based on the surface pressure matrix p train , reference static pressure vector and the reference dynamic pressure vector Calculate the reference pressure coefficient matrix C p,reference ;

[0048] Step 4: Construct a geometric feature fitting neural network;

[0049] Step 5: Train the neural network with the input matrix Γ train As the input of the geometric feature fitting neural network training, the reference angle of attack vector Reference sideslip angle vector and the reference pressure coefficient matrix C p,reference As the output of the geometric feature fitting neural network training, the geometric feature fitting neural network training is completed;

[0050] Step 6: Use the pressure measuring hole at the front end of the actual aircraft to obtain the actual surface pressure vector at the current moment

[0051] Step 7: Construct the actual network input vector at the current moment

[0052] Step 8: Use the actual network input vector at the current moment The geometric feature fitting neural network trained in step 5 is used to obtain the current fitting angle of attack Side slip angle and the fitted pressure coefficient vector

[0053] Step 9: Based on the actual surface pressure vector at the current moment And the current moment fitting pressure coefficient vector output by the neural network Calculate the current dynamic pressure and static pressure

[0054] Step 10: Based on the calculated dynamic pressure at the current moment and static pressure Determine the current flight status;

[0055] Step 11: Calculate the current flight Mach number based on the determined flight status.

[0056] Specifically, in step 1, the surface pressure matrix p trainis an m×n dimensional matrix, m is the number of wind tunnel test condition combinations (wind tunnel test conditions include Mach number, angle of attack and sideslip angle. Different combinations of these three conditions constitute different wind tunnel test condition combinations. For example, when the wind tunnel test condition is only "Mach number 0.3, angle of attack -9°, sideslip angle -6°", m is recorded as 1. When the wind tunnel test conditions include "Mach number 0.3, angle of attack -9°, sideslip angle -6°" and "Mach number 0.3, angle of attack -6°, sideslip angle -6°", m can be recorded as 2. By analogy, m is the number of different condition combinations in the wind tunnel test), n is the number of FADS pressure gauges, each row is the surface pressure information measured at the same time, and different rows represent the surface pressure information under different wind tunnel test condition combinations. The reference angle of attack vector Reference sideslip angle matrix Reference Static Pressure Matrix and the reference dynamic pressure matrix are all m-dimensional vectors, and each dimension element is related to p train The atmospheric data corresponding to the surface pressure information of the same row number in the graph, wherein the atmospheric data includes the angle of attack α, the sideslip angle β, the static pressure P0, the dynamic pressure q c , the elements in the reference atmospheric data vector are directly obtained through wind tunnel tests.

[0057] Specifically, in step 2, Γ train It is an m×(n+1)-dimensional matrix, with each row vector The element Γ mi Only use p train Corresponding row vector Calculated, The first n-dimensional elements Γ mi The calculation method is shown in formula (1):

[0058]

[0059] Among them, p mi is a row vector The surface pressure value measured by the i-th pressure measuring hole, p m,max is a row vector The maximum surface pressure value, p m,min Row vector The minimum surface pressure value in The last dimension element Γ m,end The calculation method is shown in formula (2):

[0060]

[0061] Specifically, in step 3, the reference pressure coefficient matrix C p,reference is an m×n dimensional matrix, C p,reference Each row vector is the pressure coefficient vector composed of the pressure coefficients of each pressure measuring hole under wind tunnel test conditions, Elements The calculation method of is shown in formula (3):

[0062]

[0063] in, is the reference static pressure matrix Center and surface pressure row vector The corresponding static pressure, is the reference dynamic pressure matrix Center and surface pressure row vector The corresponding dynamic pressure.

[0064] Specifically, the geometric feature fitting neural network constructed in step 4 is a three-layer non-fully connected neural network, in which the hidden layer neurons are divided into two sub-networks according to different functions: sub-network 1 and sub-network 2. Sub-network 1 is a single hidden layer structure with a total of 30 nodes, and sub-network 2 is a double hidden layer structure with 35 nodes in each layer. The input layer neuron activation function uses the Sigmoid function, and its function form is:

[0065]

[0066] Among them, x is the input of the activation function, and σ(x) is the output of the activation function;

[0067] Number of input layer nodes and neural network input matrix Γ train The number of columns is the same as n+1, the number of output layer nodes is n+2, the output layer nodes corresponding to the angle of attack α and the sideslip angle β are only connected to the nodes of subnetwork 1, and the output pressure coefficient vector The nodes of are only connected to the nodes of sub-network 2. The output layer activation function uses the linear identity function purelin function. The constructed network architecture is as follows: Figure 1 shown.

[0068] Specifically, the neural network training method in step 5 adopts the gradient descent method, and the learning rate is 1×10 -3 , the training target is set to 1×10 -6 .

[0069] Specifically, in step 7, the actual input vector of the network at the current moment is is an n+1-dimensional vector, and its calculation method is the same as Γ in step 2 train Each row vector The calculation method is the same, expressed as:

[0070]

[0071] Among them, p i is the surface pressure vector at the current moment The surface pressure corresponding to the i-th pressure measuring hole in is p max yes The maximum surface pressure value, p min yes The minimum surface pressure value, Γ i for In the first n dimensions, i The corresponding i-th element, Γ en d is The last element in .

[0072] Specifically, in step 9, the dynamic pressure and static pressure The calculation formula is shown in formula (7):

[0073]

[0074] in:

[0075] The fitted pressure coefficient vector output by the geometric feature fitting neural network;

[0076] and Unit vector of the same dimension.

[0077] Specifically, in step 10, the current flight status determination method is:

[0078]

[0079] When Ma≤1, it is judged as subsonic flight, and when Ma>1, it is judged as supersonic flight.

[0080] Specifically, in step 11, according to the determined flight state, the specific Mach number is solved by referring to equation (9):

[0081]

[0082] In order to verify the effectiveness and advancement of the FADS system atmospheric parameter solution method proposed in the present invention, a wind tunnel test was first conducted using a scale model of a certain type of aircraft head to obtain a total of 168 sets of wind tunnel test data under different conditions, which were used to construct a neural network training set and a validation set. The training set is used to train the neural network used in the present invention, and the validation set is used to test the effect of the present invention in solving atmospheric parameters. The layout of the pressure measuring holes on the scale model is as follows: Figure 2 shown.

[0083] In order to simulate the three different states of normal, sparse feature points and network extrapolation that may appear in the actual application scenarios of neural networks, three groups of "training-testing" data sets in different states were constructed, and the error mean and root mean square error verified by 10 repeated experiments were used to analyze the accuracy and stability of solving atmospheric parameters. First, the atmospheric data is solved according to the steps described in claim 1 under normal conditions, and the error mean and root mean square error of the atmospheric data solved 10 times are statistically recorded after 10 verifications, and then the verification and result recording in the sparse feature point state and network extrapolation state are completed in turn. The comparison effect is shown in the attached table and the attached figure.

[0084] Under normal conditions, the atmospheric data solution results obtained by the method proposed in the present invention are compared with the true values ​​in the wind tunnel test data in 10 repeated tests. Figures 3 to 7 The accompanying drawings describe the changes in the mean value of the fitting error and the root mean square error of the angle of attack, sideslip angle, static pressure, dynamic pressure and Mach number of the method proposed by the present invention under this state. The triangle symbol represents the mean value of the fitting error, and the length of the error bar represents the root mean square error.

[0085] The statistical results of the average fitting error of 10 repeated tests are shown in Table 1.

[0086] Table 1 Mean error of atmospheric parameter calculation in normal state repeated verification test

[0087] α error (°) β error (°) <![CDATA[q c Error (%)]]> <![CDATA[P ∞ Error (%)]]> Mach number error (%) MEAN 0.063 0.007 0.010 0.065 0.015 RMSE 0.248 0.116 2.026 0.344 1.074

[0088] The verification results show that the errors of the atmospheric parameters calculated by the method proposed in the present invention are maintained at a relatively low level. Among them, the average errors of the angle of attack and sideslip angle are less than 0.1°, the average errors of the dynamic pressure, static pressure and Mach number are less than 0.1%, and the average error of the static pressure equivalent height is less than 20m.

[0089] The atmospheric data solution results obtained by the method proposed in this invention under the condition of sparse feature points are compared with the true values ​​in the wind tunnel test data in 10 repeated tests. Figures 8 to 12 The accompanying drawings describe the changes in the mean value of the fitting error and the root mean square error of the angle of attack, sideslip angle, static pressure, dynamic pressure and Mach number of the method proposed by the present invention under this state.

[0090] The statistical results of the average fitting error of 10 repeated tests are shown in Table 2.

[0091] Table 2 Mean error of atmospheric parameter solution in repeated verification test of feature point sparse state

[0092] α error (°) β error (°) <![CDATA[q c Error (%)]]> <![CDATA[P ∞ Error (%)]]> Mach number error (%) MEAN 0.026 0.051 0.245 0.121 0.152 RMSE 0.355 0.261 2.469 1.034 1.449

[0093] The atmospheric parameter errors calculated under the sparse feature point state have increased to a certain extent compared with the normal state, but the average errors of the angle of attack and sideslip angle are still less than 0.1°, the average errors of the dynamic pressure and static pressure are less than 0.3%, and the average error of the Mach number is less than 2%.

[0094] The atmospheric data solution results obtained by the method proposed in the present invention under the network extrapolation state are compared with the true values ​​in the wind tunnel test data in 10 repeated tests. Figures 13 to 17 The accompanying drawings describe the changes in the mean value of the fitting error and the root mean square error of the angle of attack, sideslip angle, static pressure, dynamic pressure and Mach number of the method proposed by the present invention under this state.

[0095] The statistical results of the average fitting error of 10 repeated tests are shown in Table 3.

[0096] The verification results under the network extrapolation state show that the average error of the atmospheric parameter solution obtained by the method proposed in the present invention can still be maintained at a relatively low level, but the average root mean square error is significantly increased compared with the previous two states. Among them, the average error of the angle of attack and sideslip angle solution is still less than 0.1°, the average error of the dynamic pressure and static pressure solution is less than 0.7%, and the average error of the Mach number solution is less than 5%.

[0097] Table 3 Mean error of atmospheric parameter solution in repeated verification test of network extrapolation state

[0098] α error (°) β error (°) <![CDATA[q c Error (%)]]> <![CDATA[P ∞ Error (%)]]> Mach number error (%) MEAN 0.070 0.018 0.618 0.394 0.431 RMSE 0.385 0.191 1.261 1.539 2.074

[0099] Verification results show that the method proposed in the present invention can maintain a relatively low atmospheric data fitting error under various conditions. Among them, the fitting errors of angle of attack and sideslip angle are relatively stable, and the fitting errors of dynamic pressure, static pressure and Mach number will increase with the decrease of the overlap between the verification set and the training set, but are generally maintained at a relatively low level.

[0100] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A method for calculating atmospheric parameters of a FADS system based on a geometric feature fitting neural network, characterized in that: The following steps are involved: Step 1: Use the wind tunnel and aircraft scale model to obtain the surface pressure matrix p required for training the neural network train and a reference atmospheric data vector, the reference atmospheric data vector including a reference angle of attack vector Reference sideslip angle vector Reference static pressure vector and the reference dynamic pressure vector Step 2: Based on the surface pressure matrix p train Construct neural network training input matrix Γ train ; Step 3: Based on the surface pressure matrix p train , reference static pressure vector and the reference dynamic pressure vector Calculate the reference pressure coefficient matrix C p,reference ; Step 4: Construct a geometric feature fitting neural network; Step 5: Train the neural network with the input matrix Γ train As the input of the geometric feature fitting neural network training, the reference angle of attack vector Reference sideslip angle vector and the reference pressure coefficient matrix C p,reference As the output of the geometric feature fitting neural network training, the geometric feature fitting neural network training is completed; Step 6: Use the pressure measuring hole at the front end of the actual aircraft to obtain the actual surface pressure vector at the current moment Step 7: Construct the actual network input vector at the current moment Step 8: Use the actual network input vector at the current moment The geometric feature fitting neural network trained in step 5 is used to obtain the current fitting angle of attack Side slip angle and the fitted pressure coefficient vector Step 9: Based on the actual surface pressure vector at the current moment And the current moment fitting pressure coefficient vector output by the neural network Calculate the current dynamic pressure and static pressure Step 10: Based on the calculated dynamic pressure at the current moment and static pressure Determine the current flight status; Step 11: Calculate the current flight Mach number based on the determined flight status.

2. The method for calculating atmospheric parameters of a FADS system based on a geometric feature fitting neural network according to claim 1, characterized in that: In step 1, the surface pressure matrix p train is an m×n dimensional matrix, where m is the number of wind tunnel test condition combinations, n is the number of FADS pressure gauges, and each row represents the surface pressure information measured at the same time. Different rows represent the surface pressure information under different wind tunnel test condition combinations. The reference angle of attack vector Reference sideslip angle matrix Reference Static Pressure Matrix and the reference dynamic pressure matrix are all m-dimensional vectors, and each dimension element is related to p train The atmospheric data corresponding to the surface pressure information of the same row number in the graph, wherein the atmospheric data includes the angle of attack α, the sideslip angle β, the static pressure P0, the dynamic pressure q c , the elements in the reference atmospheric data vector are directly obtained through wind tunnel tests.

3. The method for calculating atmospheric parameters of a FADS system based on a geometric feature fitting neural network according to claim 2, characterized in that: In step 2, Γ train It is an m×(n+1)-dimensional matrix, with each row vector The element Γ mi Only use p train Corresponding row vector Calculated, The first n-dimensional elements Γ mi The calculation method is shown in formula (1): Among them, p mi is a row vector The surface pressure value measured by the i-th pressure measuring hole, p m,max is a row vector The maximum surface pressure value, p m,min Row vector The minimum surface pressure value in The last dimension element Γ m,end The calculation method is shown in formula (2):

4. The method for calculating atmospheric parameters of a FADS system based on a geometric feature fitting neural network according to claim 3 is characterized in that: In step 3, the reference pressure coefficient matrix C p,reference is an m×n dimensional matrix, C p,reference Each row vector is the pressure coefficient vector composed of the pressure coefficients of each pressure measuring hole under wind tunnel test conditions, Elements The calculation method of is shown in formula (3): in, is the reference static pressure matrix Center and surface pressure row vector The corresponding static pressure, is the reference dynamic pressure matrix Center and surface pressure row vector The corresponding dynamic pressure.

5. The method for calculating atmospheric parameters of a FADS system based on a geometric feature fitting neural network according to claim 4, characterized in that: The geometric feature fitting neural network constructed in step 4 is a three-layer non-fully connected neural network, in which the hidden layer neurons are divided into two sub-networks according to different functions: sub-network 1 and sub-network 2. Sub-network 1 is a single hidden layer structure with a total of 30 nodes, and sub-network 2 is a double hidden layer structure with 35 nodes in each layer. The input layer neuron activation function uses the Sigmoid function, and its function form is: Among them, x is the input of the activation function, and σ(x) is the output of the activation function; Number of input layer nodes and neural network input matrix Γ train The number of columns is the same as n+1, the number of output layer nodes is n+2, the output layer nodes corresponding to the angle of attack α and the sideslip angle β are only connected to the nodes of subnetwork 1, and the output pressure coefficient vector The nodes are only connected to the nodes of sub-network 2, and the output layer activation function uses the linear identity function purelin function.

6. The method for calculating atmospheric parameters of a FADS system based on a geometric feature fitting neural network according to claim 5, characterized in that: The neural network training method in step 5 adopts the gradient descent method, and the learning rate is 1×10 -3 , the training target is set to 1×10 -6 .

7. The method for calculating atmospheric parameters of a FADS system based on a geometric feature fitting neural network according to claim 6, characterized in that: In step 7, the actual input vector of the network at the current moment is is an n+1-dimensional vector, and its calculation method is the same as Γ in step 2 train Each row vector The calculation method is the same, expressed as: Among them, p i is the surface pressure vector at the current moment The surface pressure corresponding to the i-th pressure measuring hole in is p max yes The maximum surface pressure value, p min yes The minimum surface pressure value, Γ i for In the first n dimensions, i The corresponding i-th element, Γ end yes The last element in .

8. The method for calculating atmospheric parameters of a FADS system based on a geometric feature fitting neural network according to claim 7, characterized in that: In step 9, the dynamic pressure and static pressure The calculation formula is shown in formula (7): in: The fitted pressure coefficient vector output by the geometric feature fitting neural network; and Unit vector of the same dimension.

9. The method for calculating atmospheric parameters of a FADS system based on a geometric feature fitting neural network according to claim 8, characterized in that: In step 10, the current flight status determination method is: When Ma≤1, it is judged as subsonic flight, and when Ma>1, it is judged as supersonic flight.

10. The method for calculating atmospheric parameters of a FADS system based on a geometric feature fitting neural network according to claim 9, characterized in that: In step 11, according to the determined flight state, the specific Mach number is solved by referring to equation (9):

Images