Sound explosion signal intelligent inversion method based on Transform network

Abstract

The invention provides a sound detonation signal intelligent inversion method based on a Transform network, and the method comprises the steps: rapidly generating a sound detonation near-far field signal sample data set through a forward propagation augmentation Burgers equation, training a Transform network through the sound detonation near-far field sample data, and carrying out the inversion of a near-field signal corresponding to a target far-field signal through a trained model. According to the method, the problem of dissipation existing in the numerical solution reverse augmentation Burgers equation is avoided, sound detonation inversion calculation can be carried out from any position, and the application range of sound detonation inversion is effectively widened. Meanwhile, a near-field signal can be designed based on a low-sound-explosion ground signal, and a design target is provided for low-sound-explosion optimization of the supersonic civil aircraft.

Description

Technical Field

[0001] The present invention relates to the field of aircraft design and acoustic technology, and in particular to a Transformer network-based sonic boom signal intelligent inversion method. Background Art

[0002] Supersonic civil aircraft are a revolutionary means of transportation that breaks through the cruising efficiency bottleneck of existing civil airliners. While the first generation of aircraft, exemplified by the Concorde in the 1970s, achieved Mach 2 cruise, they produced high sonic booms and suffered from high fuel consumption, ultimately withdrawing from commercial operation due to environmental and economical deficiencies. A new generation of aircraft, exemplified by NASA's QueSST program, prioritizes sonic boom reduction in supersonic civil aircraft design. Through low-sonic-boom design technology, they aim to reduce the sonic boom ground-perceived noise level to 70 PLdB, further expanding the market for supersonic civil aircraft.

[0003] The current sonic boom research system includes two major branches: forward propagation and inverse inversion: (1) Forward propagation technology uses low-precision modified linearization theory or high-precision computational fluid dynamics (CFD) to obtain near-field signals, and combines them with the augmented Burgers equation to perform far-field deduction; (2) Inverse inversion technology solves the inverse augmented Burgers equation to infer the near-field signal from the target far-field signal, and then calculates the equivalent area distribution of the aircraft, providing a reference for the design of low sonic boom shapes. However, the inverse augmented Burgers equation cannot well handle the dissipation problem during waveform propagation. The inverted near-field signal has the problem of smoothing the peak, so its application scenarios are limited to a certain extent. Summary of the Invention

[0004] In order to solve the problems existing in the prior art, the present invention proposes a sonic boom signal intelligent inversion method based on a Transformer network. By forward propagating the augmented Burgers equation, a sample data set of sonic boom near-field and far-field signals is quickly generated. The sonic boom near-field and far-field sample data are used to train the Transformer network. Then, the trained model is used to invert the near-field signal corresponding to the target far-field signal. This method avoids the dissipation problem existing in numerically solving the inverse augmented Burgers equation. The sonic boom inversion calculation can be performed from any position (including the ground), effectively broadening the application range of sonic boom inversion. At the same time, the near-field signal can be designed based on the low sonic boom ground signal, providing a design target for the low sonic boom optimization of supersonic civil aircraft.

[0005] The technical solution of the present invention is:

[0006] A Transformer network-based intelligent inversion method for sonic boom signals includes the following steps:

[0007] Step S1: Selecting a sonic boom reference near-field signal and calculating a sonic boom reference far-field signal;

[0008] Step S2: Designing a sonic boom target far-field signal based on the sonic boom reference far-field signal calculated in step S1;

[0009] Step S3: adding disturbance to the sonic boom reference near-field signal to obtain the sonic boom near-field signal after superimposing the disturbance, and using the sonic boom near-field signal after superimposing the disturbance to calculate the corresponding sonic boom far-field signal, thereby establishing a sonic boom near-field and far-field signal sample set;

[0010] Step S4: Using the sonic boom near-field and far-field signal sample sets established in step S3, train the Transformer network;

[0011] Step S5: Using the Transformer network to invert the sonic boom target far-field signal designed in step S2 to obtain the sonic boom target near-field signal, and then calculating the corresponding sonic boom target far-field verification signal based on the obtained sonic boom target near-field signal;

[0012] Step S6: Select the optimal sonic boom near-field signal from the sonic boom near-field and far-field signal sample sets and the inversion results as the new sonic boom reference near-field signal, and repeat steps S3 to S5 until the inversion result meets the requirements to obtain the final sonic boom target near-field signal.

[0013] Furthermore, the sonic boom reference near-field signal extraction position in step S1 is 3 to 4 times the fuselage length below the fuselage.

[0014] Furthermore, in step S1, the selected sonic boom reference near-field signal is used to calculate the sonic boom reference far-field signal by adopting the augmented Burgers equation method; in step S3, the sonic boom near-field signal after superimposing the disturbance is used to calculate the corresponding sonic boom far-field signal by adopting the augmented Burgers equation method.

[0015] Furthermore, in step S2, the process of designing the sonic boom far-field target signal is as follows: based on the sonic boom reference far-field signal calculated in step S1, the signal within the set range of the peak and trough is cut off to obtain the target far-field signal.

[0016] Furthermore, in step S3, the sonic boom near-field signal disturbance waveforms used are divided into three categories: left sawtooth wave, symmetrical triangle wave, and right sawtooth wave; the process of adding disturbance to the sonic boom reference near-field signal is: starting from the starting point of the sonic boom reference near-field signal, a disturbance waveform with a set amplitude and a set duration is added in sections.

[0017] Furthermore, in step S4, the samples are standardized before training the network: the sonic boom near-field signal in the sonic boom near-field and far-field signal sample set is subtracted from the sonic boom reference near-field signal to obtain a sonic boom near-field signal disturbance value; the sonic boom far-field signal in the sonic boom near-field and far-field signal sample set is subtracted from the sonic boom reference far-field signal to obtain a sonic boom far-field signal disturbance value; the sonic boom near-field signal disturbance value is used as a normalized true sample, and the sonic boom far-field signal disturbance value is used as a normalized label.

[0018] Furthermore, in step S5, the difference between the sonic boom target far-field signal and the sonic boom reference far-field signal is input to the Transformer network, and the Transformer network outputs the difference between the corresponding near-field signal and the sonic boom reference near-field signal. The output result is added to the sonic boom reference near-field signal to obtain the inverted sonic boom target near-field signal.

[0019]

[0020] Then the sonic boom target near-field signal Deduced to the far field, the corresponding sonic boom target far-field verification signal is obtained.

[0021] Furthermore, in step 6, the process of selecting the optimal sonic boom near-field signal is:

[0022] In the sonic boom near-field and far-field signal sample set, the relative average Euclidean distance between all sonic boom far-field signals and the sonic boom target far-field signal designed in step S2 is calculated; the relative average Euclidean distance between the sonic boom target far-field verification signal and the sonic boom target far-field signal designed in step S2 is calculated; and the sonic boom near-field signal corresponding to the sonic boom far-field signal with the smallest relative average Euclidean distance is taken as the new sonic boom reference near-field signal.

[0023] Furthermore, if the sonic boom reference near-field signal is still the original sonic boom reference near-field signal after step S6, the disturbance amplitude in step S3 is reduced, and steps S3 to S5 are repeated.

[0024] Furthermore, the process of reducing the disturbance amplitude is:

[0025] A(i)=A ∞ +(A0-A ∞ )e -αi

[0026] Where i is the number of times the disturbance amplitude is reduced, A0 is the initial disturbance amplitude, A ∞ is the final disturbance amplitude set, A(i) is the current disturbance amplitude, α is the reduction factor, and e is the natural logarithm.

[0027] Beneficial effects

[0028] This paper proposes a Transformer-based intelligent sonic boom signal inversion method. By constructing and training a Transformer network, it intelligently predicts the near-field signal corresponding to a given target's far-field signal. This method avoids the dissipation problem inherent in numerically solving the inverse augmented Burgers equation and can perform sonic boom inversion calculations from any location (including the ground), effectively broadening the application range of sonic boom inversion. Furthermore, near-field signals can be designed based on low-sonic boom ground signals, providing a design target for low-sonic boom optimization of supersonic civil aircraft.

[0029] Additional aspects and advantages of the present invention will be set forth in part in the description which follows and, in part, will be obvious from the description which follows, or may be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments with reference to the accompanying drawings, in which:

[0031] Figure 1 This is a flow chart of the intelligent inversion method for sonic boom signals based on a Transformer network proposed in an embodiment of the present invention;

[0032] Figure 2 is the sonic boom reference near-field signal selected in the embodiment of the present invention;

[0033] Figure 3 The sonic boom reference far-field signal calculated by an embodiment of the present invention;

[0034] Figure 4 The far-field signal of the sonic boom target designed in the embodiment of the present invention;

[0035] Figure 5 The sonic boom near-field signal disturbance used in the embodiment of the present invention;

[0036] Figure 6 This is a portion of the sonic boom near-field and far-field signal sample data set used in an embodiment of the present invention;

[0037] Figure 7 A schematic diagram of the Transformer network structure used in an embodiment of the present invention;

[0038] Figure 8 The iterative process of the far-field signal obtained by inversion using the Transformer network in an embodiment of the present invention;

[0039] Figure 9 The optimal sonic boom near-field signal obtained by using the Transformer network inversion in an embodiment of the present invention;

[0040] Figure 10 This is the sonic boom far-field signal corresponding to the optimal sonic boom near-field signal obtained by inversion using the Transformer network in an embodiment of the present invention. DETAILED DESCRIPTION

[0041] In order to make the technical problems, technical solutions and beneficial effects solved by the present invention more clearly understood and to enable those skilled in the art to better understand the solutions of the present invention, the present invention is further described and fully explained below in conjunction with the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only intended to illustrate the present invention and are not intended to limit the present invention.

[0042] This embodiment takes the LM1021 aircraft as an example and adopts the intelligent inversion method of sonic boom signals based on the Transformer network proposed in the present invention to invert its sonic boom signals to obtain a near-field signal that meets the far-field low sonic boom characteristics.

[0043] like Figure 1 As shown, an embodiment of the present invention proposes an intelligent inversion method for sonic boom signals based on a Transformer network, comprising the following steps:

[0044] Step S1: Selecting a sonic boom reference near-field signal and calculating a sonic boom reference far-field signal;

[0045] In the embodiment of the present invention, the flight conditions of LM1021 are shown in Table 1, the sonic boom reference near-field signal extraction position is 3 to 4 times the fuselage length below the LM1021 aircraft fuselage, and the sonic boom near-field signal extraction position is shown in Table 2.

[0046] Table 1 LM1021 flight conditions

[0047]

[0048] Table 2 LM1021 sonic boom near-field signal extraction position

[0049]

[0050] Figure 2 is the sonic boom reference near-field signal selected in the embodiment of the present invention, Figure 3 This is the sonic boom reference far-field signal calculated by an embodiment of the present invention.

[0051] Using the selected sonic boom reference near-field signal, the augmented Burgers equation method is used to calculate the sonic boom reference far-field signal, as follows:

[0052] The augmented Burgers equation describes the propagation of sound waves in dissipative media such as the atmosphere. Its derivation process is as follows:

[0053] The Navier-Stokes equations are approximated to obtain the Westervelt equations:

[0054]

[0055] Where p′ is atmospheric pressure; c0 is the speed of sound; t is time; ρ0 is atmospheric density; b is the classical absorption coefficient; and β is the nonlinear coefficient.

[0056] The one-dimensional Westervelt equation is:

[0057]

[0058] Where x is the position and t is the propagation time starting from the moment the sonic boom is generated. Introducing the delay time t' starting from the moment it propagates to x, the following conversion relationship is obtained:

[0059]

[0060] By performing coordinate transformation on the one-dimensional Westervelt equation and ignoring high-order small quantities, we can obtain the classic Burgers equation:

[0061]

[0062] Where p′ is atmospheric pressure; x is distance; ρ0 is atmospheric density; c0 is the speed of sound; t is time; β is the nonlinear coefficient; and b is the classical absorption coefficient.

[0063] The classic Burgers equation is dimensionless. Introducing dimensionless pressure P = p' / p0 and dimensionless position Reference length The dimensionless time τ = ω0t′ and the dimensionless gas dissipation parameter Γ = (bω0) / (2βp0) give the dimensionless classic Burgers equation:

[0064]

[0065] On the basis of the classic Burgers equation, considering atmospheric stratification and geometric diffusion, the generalized Burgers equation is obtained:

[0066]

[0067] Where P is the dimensionless pressure; σ is the dimensionless position; τ is the dimensionless time; Γ is the dimensionless gas dissipation parameter; ρ0 is the free stream atmospheric density; c0 is the speed of sound; and S is the area of the sound tube.

[0068] By adding the molecular relaxation effect term to the generalized Burgers equation, we get the augmented Burgers equation:

[0069]

[0070] Where P is the dimensionless pressure; σ is the dimensionless position; τ is the dimensionless time; Γ is the dimensionless gas dissipation parameter; ρ0 is the free flow atmospheric density; c0 is the speed of sound; S is the area of the sound tube; ν is a certain gas molecule; θ v =ω0τ v is the dimensionless molecular relaxation time; is the dimensionless molecular relaxation coefficient.

[0071] In the actual solution process, the operator splitting method is often used to solve the augmented Burgers equation. The equation is split into the following five small equations:

[0072]

[0073] The above equations correspond to nonlinear effects, classical dissipation, inhomogeneous media, geometric diffusion, and molecular relaxation effects, respectively. They are solved using a numerical discretization scheme:

[0074]

[0075] Where, the superscript is the time discrete number; the subscript is the space discrete number.

[0076] Step S2: Designing a sonic boom target far-field signal based on the sonic boom reference far-field signal calculated in step S1;

[0077] Based on the sonic boom reference far-field signal calculated in step S1, the signal within the set range of peaks and troughs is truncated to obtain the target far-field signal. Although the truncated signal has a sudden change in non-physical characteristics, the target signal only represents the design trend of low sonic boom and is not a physical signal that must be achieved. Table 3 shows the peak and trough processing positions of the target far-field waveform in an embodiment of the present invention. Figure 4 This is the far-field signal of the sonic boom target designed in the embodiment of the present invention.

[0078] Table 3 Peak and trough processing positions of target far-field waveform

[0079]

[0080] Step S3: Add disturbance to the sonic boom baseline near-field signal to obtain the sonic boom near-field signal after superimposing the disturbance. Use the sonic boom near-field signal after superimposing the disturbance to calculate the corresponding sonic boom far-field signal using the augmented Burgers equation method to establish a sonic boom near-field and far-field signal sample set.

[0081] The sonic boom near-field signal disturbance waveforms used in the embodiments of the present invention are divided into three categories: left sawtooth wave, symmetrical triangle wave, and right sawtooth wave. In the embodiments of the present invention, starting from the starting point of the sonic boom reference near-field signal, disturbance waveforms with set amplitude and set duration are added in sections. Table 4 shows the amplitude and duration parameters of the sonic boom near-field signal disturbance waveforms used in the embodiments of the present invention. Figure 5 This is a diagram of the sonic boom near-field signal disturbance used in an embodiment of the present invention.

[0082] Table 4. Waveform parameters of sonic boom near-field signal disturbance

[0083]

[0084] The augmented Burgers equation is used to calculate the far-field signal of the sonic boom after the superimposed disturbance. The specific calculation method is the same as step S1. The final sonic boom near-field and far-field signal sample set consists of 5000 pairs of near-field and far-field signals. Figure 6 Some samples in the sonic boom near-field and far-field signal sample dataset used in the embodiments of the present invention are given.

[0085] Step S4: Use the sonic boom near-field and far-field signal sample sets established in step S3 to train the Transformer network.

[0086] The Transformer is a deep learning architecture that relies entirely on the self-attention mechanism. It abandons the serial time-series processing of traditional recurrent neural networks (RNNs) and convolutional neural networks (CNNs). By computing global sequence relationships in parallel, it significantly improves training efficiency and the ability to capture long-range dependencies. The model employs a stacked encoder-decoder structure: the encoder, consisting of a multi-head attention layer and a feedforward neural network, dynamically captures the global semantic connections of the input sequence; the decoder uses masked self-attention to prevent future information leakage and combines the encoder output to generate the target sequence.

[0087] Before training the network, the samples are standardized: the sonic boom near-field signal in the sonic boom near-field and far-field signal sample sets is subtracted from the sonic boom benchmark near-field signal to obtain the sonic boom near-field signal disturbance value; the sonic boom far-field signal in the sonic boom near-field and far-field signal sample sets is subtracted from the sonic boom benchmark far-field signal to obtain the sonic boom far-field signal disturbance value; the sonic boom near-field signal disturbance value is used as the normalized true sample, and the sonic boom far-field signal disturbance value is used as the normalized label:

[0088]

[0089] Where Δdpp near Indicates the sonic boom near-field signal disturbance value, dpp near represents the sonic boom near-field signal, Represents the sonic boom reference near-field signal; Δdp farIndicates the far-field disturbance value of the sonic boom signal, dp far represents the far-field signal of the sonic boom, Represents the sonic boom reference far-field signal.

[0090] The Transformer network architecture consists of four parts: input, encoder, decoder, and output:

[0091] The input layer consists of an embedding layer that converts the input into a fixed-dimensional vector representation, and a positional encoding layer that adds position information to the preprocessed vector, enabling the Transformer model to process sequence order.

[0092] The encoder consists of N stacked encoders, each of which includes a multi-head self-attention layer and a feedforward network layer. The multi-head self-attention mechanism independently focuses on the different characteristics of the input signal and establishes global correlations based on these characteristics. Furthermore, compared to the serial signal processing of recurrent neural networks (RNNs), the multi-head self-attention mechanism can process signal sequences in parallel, significantly improving efficiency.

[0093] The decoder consists of N stacked decoders, each of which includes a masked multi-head self-attention layer, an encoder-decoder multi-head attention layer, and a feedforward network layer. The masked multi-head self-attention layer masks future position information using a triangular matrix to construct temporal dependencies of the target sequence. The encoder-decoder multi-head attention layer establishes cross-modal alignment between the source and target sequences, enabling conditional generation.

[0094] The output part consists of a linear layer and a softmax layer. The linear layer converts the output vector into the required dimension; the softmax layer converts the output of the linear layer into a probability distribution for the final prediction.

[0095] In an embodiment of the present invention, a sonic boom far-field signal is input to a Transformer network, and the network outputs a sonic boom near-field signal. Figure 7 Schematic diagram of the Transformer network structure used in the embodiment of the present invention. Table 5 shows the Transformer network structure parameters used in the embodiment of the present invention.

[0096] Table 5 Transformer network structure parameters

[0097]

[0098] The network was trained using the SGD optimizer. SGD stands for stochastic gradient descent, and its core concept is to minimize the loss function by iteratively adjusting model parameters. Although SGD has a slower convergence rate, it is more suitable for highly volatile data such as sonic boom signals and reduces the risk of overfitting.

[0099] The accuracy of the Transformer network built in step S4 is evaluated using the following indicators: mean absolute error (MAE), mean absolute percentage error (MAPE), correlation coefficient (R 2 ), root mean square error (RMSE).

[0100] The closer the mean absolute error (MAE) is to 0, the better the prediction performance of the model. The calculation method of mean absolute error (MAE) is:

[0101]

[0102] Where N is the number of data points; y i is the true value; is the predicted value.

[0103] The closer the mean absolute percentage error (MAPE) is to 0, the better the prediction performance of the model. The calculation method of mean absolute percentage error (MAPE) is:

[0104]

[0105] Correlation coefficient (R 2 ) is closer to 1, the better the prediction performance of the model. 2 ) is calculated as follows:

[0106]

[0107] Where SSR is the regression sum of squares; SST is the total sum of squares.

[0108] The closer the root mean square error (RMSE) is to 0, the better the prediction performance of the model. The root mean square error (RMSE) is calculated as:

[0109]

[0110] Step S5: Use the Transformer network to invert the sonic boom target far-field signal designed in step S2 to obtain the sonic boom target near-field signal, and then calculate the corresponding sonic boom target far-field verification signal based on the obtained sonic boom target near-field signal.

[0111] The specific process is:

[0112] The difference between the sonic boom target far-field signal and the sonic boom reference far-field signal is input to the Transformer network, and the difference between the corresponding near-field signal and the sonic boom reference near-field signal is output by the Transformer network. The output result is added to the sonic boom reference near-field signal to obtain the inverted sonic boom target near-field signal.

[0113]

[0114] Then, the augmented Burgers equation is used to transform the near-field signal of the sonic boom target into Deduced to the far field, the corresponding sonic boom target far-field verification signal is obtained.

[0115] Step S6: Select the optimal sonic boom near-field signal from the sonic boom near-field and far-field signal sample sets and the inversion results as the new sonic boom reference near-field signal, and repeat steps S3 to S5 until the inversion result meets the requirements to obtain the final sonic boom target near-field signal.

[0116] The process of selecting the optimal sonic boom near-field signal is as follows:

[0117] In the sonic boom near-field and far-field signal sample sets, the relative average Euclidean distance between all sonic boom far-field signals and the sonic boom target far-field signal designed in step S2 is calculated; the relative average Euclidean distance between the sonic boom target far-field verification signal and the sonic boom target far-field signal designed in step S2 is calculated;

[0118] The relative average Euclidean distance is defined as follows:

[0119]

[0120] Where dp is the true value, is the predicted value, and n is the number of signal sampling points.

[0121] The sonic boom near-field signal corresponding to the sonic boom far-field signal with the smallest relative average Euclidean distance is taken as the new sonic boom reference near-field signal.

[0122] There are two criteria for judging whether the inversion result meets the requirements: first, the relative average Euclidean distance between the sonic boom far-field signal corresponding to the optimal sonic boom near-field signal and the sonic boom target far-field signal designed in step S2 is less than a set threshold; second, the number of iteration steps meets the given requirements.

[0123] Furthermore, after step S6, the sonic boom reference near-field signal is updated, and this signal is used as the sonic boom reference near-field signal for a new round of iteration, and steps S3 to S5 are repeated.

[0124] If the sonic boom reference near-field signal remains the same as the original sonic boom reference near-field signal after step S6, the disturbance amplitude in step S3 is reduced, and steps S3 to S5 are repeated. The method for reducing the disturbance amplitude is as follows:

[0125] A(i)=A ∞ +(A0-A ∞ )e -αi

[0126] Where i is the number of times the disturbance amplitude is reduced, A0 is the initial disturbance amplitude, A ∞ is the final (minimum) disturbance amplitude set, A(i) is the current disturbance amplitude, α is the reduction factor, and e is the natural logarithm.

[0127] The embodiment of the present invention iterates for a total of 5 rounds. In the last 4 rounds of iteration, 1500 pairs of near-field and far-field signals are calculated in each round and added to the sonic boom near-field and far-field signal sample dataset.

[0128] Figure 8 The iterative process of the far-field signal obtained by inversion using the Transformer network in an embodiment of the present invention is given. Figure 9 The optimal sonic boom near-field signal obtained by inversion according to the embodiment of the present invention is given. Figure 10 The sonic boom far-field signal corresponding to the optimal sonic boom near-field signal is given.

[0129] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention without departing from the principles and purpose of the present invention.

Claims

1. A Transformer network-based intelligent inversion method for sonic boom signals, characterized by: The following steps are involved: Step S1: Selecting a sonic boom reference near-field signal and calculating a sonic boom reference far-field signal; Step S2: Designing a sonic boom target far-field signal based on the sonic boom reference far-field signal calculated in step S1; Step S3: adding disturbance to the sonic boom reference near-field signal to obtain the sonic boom near-field signal after superimposing the disturbance, and using the sonic boom near-field signal after superimposing the disturbance to calculate the corresponding sonic boom far-field signal, thereby establishing a sonic boom near-field and far-field signal sample set; Step S4: Using the sonic boom near-field and far-field signal sample sets established in step S3, train the Transformer network; Step S5: Using the Transformer network to invert the sonic boom target far-field signal designed in step S2 to obtain the sonic boom target near-field signal, and then calculating the corresponding sonic boom target far-field verification signal based on the obtained sonic boom target near-field signal; Step S6: Select the optimal sonic boom near-field signal from the sonic boom near-field and far-field signal sample sets and the inversion results as the new sonic boom reference near-field signal, and repeat steps S3 to S5 until the inversion result meets the requirements to obtain the final sonic boom target near-field signal.

2. The intelligent inversion method for sonic boom signals based on a Transformer network according to claim 1, characterized in that: In step S1, the sonic boom reference near-field signal extraction position is 3 to 4 times the fuselage length below the fuselage.

3. The intelligent inversion method for sonic boom signals based on a Transformer network according to claim 1, characterized in that: In step S1, the selected sonic boom reference near-field signal is used to calculate the sonic boom reference far-field signal by adopting the augmented Burgers equation method; in step S3, the sonic boom near-field signal after superimposing the disturbance is used to calculate the corresponding sonic boom far-field signal by adopting the augmented Burgers equation method.

4. The intelligent inversion method for sonic boom signals based on a Transformer network according to claim 1, characterized in that: In step S2, the process of designing the sonic boom far-field target signal is as follows: based on the sonic boom reference far-field signal calculated in step S1, the signal within the set range of the peak and trough is cut off to obtain the target far-field signal.

5. The intelligent inversion method for sonic boom signals based on a Transformer network according to claim 1, characterized in that: In step S3, the sonic boom near-field signal disturbance waveforms used are divided into three categories: left sawtooth wave, symmetrical triangle wave, and right sawtooth wave. The process of adding disturbance to the sonic boom reference near-field signal is as follows: starting from the starting point of the sonic boom reference near-field signal, a disturbance waveform with a set amplitude and a set duration is added in sections.

6. The intelligent inversion method for sonic boom signals based on a Transformer network according to claim 1, characterized in that: In the step S4, the samples are first standardized before the network is trained: the sonic boom near-field signal in the sonic boom near-field and far-field signal sample set is subtracted from the sonic boom reference near-field signal to obtain a sonic boom near-field signal disturbance value; Subtracting the sonic boom far-field signal in the sonic boom near-field and far-field signal sample sets from the sonic boom reference far-field signal to obtain a sonic boom far-field signal disturbance value; The perturbation value of the sonic boom near-field signal is used as the normalized true sample, and the perturbation value of the sonic boom far-field signal is used as the normalized label.

7. The intelligent inversion method for sonic boom signals based on a Transformer network according to claim 1, characterized in that: In step S5, the difference between the sonic boom target far-field signal and the sonic boom reference far-field signal is input to the Transformer network, and the Transformer network outputs the difference between the corresponding near-field signal and the sonic boom reference near-field signal. The output result is added to the sonic boom reference near-field signal to obtain the inverted sonic boom target near-field signal. Then the sonic boom target near-field signal Deduced to the far field, the corresponding sonic boom target far-field verification signal is obtained.

8. The intelligent inversion method for sonic boom signals based on a Transformer network according to claim 1, characterized in that: In step 6, the process of selecting the optimal sonic boom near-field signal is as follows: In the sonic boom near-field and far-field signal sample sets, the relative average Euclidean distance between all sonic boom far-field signals and the sonic boom target far-field signal designed in step S2 is calculated; Calculate the relative average Euclidean distance between the sonic boom target far-field verification signal and the sonic boom target far-field signal designed in step S2; and take the sonic boom near-field signal corresponding to the sonic boom far-field signal with the smallest relative average Euclidean distance as the new sonic boom reference near-field signal.

9. The intelligent inversion method for sonic boom signals based on a Transformer network according to claim 8, characterized in that: If the sonic boom reference near-field signal remains the original sonic boom reference near-field signal after step S6, the disturbance amplitude in step S3 is reduced, and steps S3 to S5 are repeated.

10. The intelligent inversion method for sonic boom signals based on a Transformer network according to claim 9, characterized in that: The process of reducing the disturbance amplitude is: A(i)=A ∞ +(A0-A ∞ )And -αi Where i is the number of times the disturbance amplitude is reduced, A0 is the initial disturbance amplitude, A ∞ is the final disturbance amplitude set, A(i) is the current disturbance amplitude, α is the reduction factor, and e is the natural logarithm.

Images