Artificial vibration time history intelligent generation method

Abstract

The invention discloses an intelligent generation method for an artificial vibration time history. The method comprises the following steps: establishing a diffusion model of the artificial vibration time history; obtaining a plurality of artificial vibration time-history curves capable of matching the target response spectrum and the target power spectrum, and performing forward diffusion on the artificial vibration time-history curves to obtain a group of corresponding training data; training the diffusion model by using the training data; and inputting a new standard Gaussian noise curve into the trained diffusion model, and outputting an artificial vibration time history curve matched with the response spectrum and the power spectrum by the diffusion model. According to the method, the diffusion model of the artificial vibration time history is established and trained, so that the trained diffusion model can output an artificial vibration time history curve matched with a response spectrum and a power spectrum when a standard Gaussian noise curve is input, and the method has the advantages of high matching degree and precision, and is suitable for popularization and application. As the diffusion process does not involve complex iterative calculation, the calculation efficiency of the manual vibration time history can be greatly improved.

Description

Technical Field

[0001] The invention relates to a structural seismic design and analysis method, and in particular to an artificial earthquake motion time history intelligent generation method. Background Art

[0002] When performing random seismic response analysis on civil engineering structures, it is necessary to generate a large number of artificial seismic time histories that match the target spectrum as the excitation of the structure for time history analysis and to understand its dynamic response law. For structures with higher safety requirements, such as nuclear power structures, it is required that the artificial seismic time histories of the input structure not only match the target response spectrum, but also the target power spectrum. When using traditional numerical algorithms to generate seismic time histories that match both the target response spectrum and the target power spectrum, it is necessary to iterate the calculation repeatedly, resulting in a long time-consuming process to obtain an artificial seismic time histories that meet the requirements.

[0003] Generative models can solve the time-consuming problem of traditional numerical methods because they can establish a joint probability density distribution model of the acceleration time history curve and the corresponding response spectrum without the iterative calculation and adjustment of the acceleration curve in traditional numerical methods. However, in the existing research on using generative models to generate artificial ground motion time histories, the generated artificial ground motion time histories have not achieved high-precision matching, and the generation accuracy of generative models still needs to be further improved. Summary of the invention

[0004] In view of the problems of low matching accuracy and low efficiency of artificial seismic time histories generated in the prior art, the present invention provides a method for intelligently generating artificial seismic time histories, which uses a diffusion model to generate artificial seismic time histories that match target response spectra and target power spectra, thereby improving the generation efficiency and matching accuracy of artificial seismic time histories.

[0005] In order to solve the above technical problems, the present invention includes the following technical solutions:

[0006] A method for intelligently generating an artificial earthquake time history comprises the following steps:

[0007] S1. Establishing a diffusion model of artificial earthquake time history, wherein the diffusion model adopts a neural network model based on machine learning;

[0008] S2. Obtain several artificial earthquake time history curves that can match the target response spectrum and the target power spectrum, and use Formula 1 to perform forward diffusion on each artificial earthquake time history curve to obtain a corresponding set of training data x 0 、x 1 , …, x T-1 、x T ; Among them, formula 1 is:

[0009]

[0010] In the formula, q(x t |x t-1 ) is a forward diffusion process, N(·) is a normal distribution, β t is a parameter related to the time step; t=1,2,…,T,x t 、x t-1 are the seismic time history curves of the tth and t-1th steps respectively, x 0 To match the ground motion time history curve of the target response spectrum and power spectrum at the same time, x T is the standard Gaussian noise curve, T is the total number of sampling steps, and I is the unit diagonal matrix;

[0011] S3. Use the training data obtained in step S2 to train the diffusion model, input x t , output x t-1 ; The trained diffusion model can complete the reverse diffusion process in formula 2; formula 2 is:

[0012] p θ (x t-1 |x t )=N(x t-1 ;μ θ (x t ,t),∑ θ (x t ,t));

[0013] In the formula, p θ (x t-1 |x t ) is the reverse diffusion process of the diffusion model, μ represents the mean of the normal distribution, Σ is the covariance, θ is the parameter of the neural network, and x t is the artificial ground motion time history curve obtained from the t-step diffusion process;

[0014] S4. A new standard Gaussian noise curve x T 'Input into the trained diffusion model, the diffusion model outputs the artificial ground motion time history curve x that matches the response spectrum and power spectrum 0 '.

[0015] Furthermore, a loss function is set in the diffusion model, which is denoted by L and satisfies:

[0016]

[0017] in, γ s =1-β s ,

[0018] Where ε is the real noise at time t, ε θ is the noise predicted by the neural network, is the weight factor accumulated according to the time step, γ s is the weight adjustment factor, β s is a parameter related to the time step.

[0019] Further, in step S4, the diffusion model is used to T 'Perform T-step reverse denoising. The T-step reverse denoising process is expressed as:

[0020]

[0021] In the formula, is the weight factor accumulated according to the time step, σ t is the covariance of the noise z, where z is the noise of N(0,I).

[0022] Further, in step S2, obtaining an artificial ground motion time history curve that can match the target response spectrum and the target power spectrum specifically includes:

[0023] S21. Matching the artificial earthquake time history curve with the target response spectrum;

[0024] S22. Matching the artificial earthquake time history curve that has matched the target response spectrum in step S21 to the target power spectrum;

[0025] S23. Determine whether the response spectrum corresponding to the artificial earthquake time history curve that has matched the target power spectrum in step S22 is greater than or equal to the target response spectrum, if so, proceed to step S24, if not, loop through steps S21 to S23 for the artificial earthquake time history curve that has matched the target power spectrum;

[0026] S24. The artificial earthquake time history curve that has matched the target power spectrum can match the target response spectrum and the target power spectrum.

[0027] Further, in step S21, the artificial earthquake time history curve is matched with the target response spectrum, which specifically includes:

[0028] S211. For the initial random ground motion time history, the Duhamel integral is used to calculate the response of the single degree of freedom system:

[0029]

[0030] in,

[0031] In the formula, a a (0) (t) is the response of the single-degree-of-freedom system, a g (0) (t) is the initial random earthquake time history, which can be randomly selected from the earthquake database, h Tn,ξ (t) is the unit impulse response function, ωd is the damped circular frequency, ω d =ω n / (1-ξ 2 ) 1 / 2 ;

[0032] S212. Calculate the response spectrum corresponding to the initial earthquake motion time history:

[0033]

[0034] In the formula, S a (0) (T n ,ξ) is the natural oscillation period T n and the earthquake response spectrum under the damping ratio ξ, |a a (0) (t m )| is the maximum response of the single-degree-of-freedom system, t m It is the moment of greatest response;

[0035] S213. Calculate the difference between the earthquake time history response spectrum and the target response spectrum:

[0036]

[0037] Where sgn(·) is the sign function;

[0038] S214. Construct the narrow-band incremental time history curve of the single-degree-of-freedom system response based on the difference between the response spectra:

[0039] Δa a (t) = ΔS(T n ,ξ)s n (t),

[0040] in,

[0041] In the formula, Δa a (t) corresponds to ΔS(T n ,ξ) of the narrowband increment time history curve, s n (t) is a narrowband amplitude modulation function;

[0042] S215. According to the narrow-band incremental time history curve of the single-degree-of-freedom system response, the narrow-band incremental time history curve of the single-degree-of-freedom response is converted into a narrow-band incremental time history curve of the excitation based on the frequency response transfer function:

[0043]

[0044] In the formula, H a (ω) is the frequency response function of the single degree of freedom system, Δa g(t) is the narrowband incremental time history curve of the excitation, F(·) and F -1 (·) is Fourier transform and inverse Fourier transform;

[0045] S216. Superimpose the excitation narrowband incremental time history curve on the original vibration time history so that the modified time history curve is within the natural vibration period T n is greater than or equal to the target response spectrum, where

[0046]

[0047] In the formula, a g (1) (t) is the artificial ground motion time history obtained in the first iteration;

[0048] S217. Correct different natural oscillation periods T through steps S213 to S216 n The response spectrum corresponding to the ground motion time history is obtained to obtain the artificial ground motion time history a that matches the target response spectrum g (l) (t).

[0049] Further, in step S22, the artificial earthquake time history curve that has been matched with the target response spectrum in step S21 is matched with the target power spectrum, which specifically includes:

[0050] S221. Calculate the artificial earthquake time history a g (l) The power spectral density G of (t) g (l) (ω), calculate the power spectrum density G of the artificial ground motion time history g (l) (ω) and the target power spectrum density G T The difference between (ω)

[0051] Where α is the proportional coefficient of the interval [0,1], i = 1, 2, ..., N-1, and N is the number of sampling points;

[0052] S222. The difference power spectrum ΔG g (l) (ω) is transformed into a Fourier amplitude spectrum, and then the Fourier amplitude spectrum is inversely transformed into an incremental acceleration time history curve:

[0053]

[0054] In the formula, |ΔA g (l) (ω i )| is the Fourier amplitude spectrum corresponding to the difference power spectrum, Δt is the sampling time interval, Δa g(l) (t) is the incremental acceleration time history curve;

[0055] S223. The incremental acceleration time history curve is shaped and then superimposed on the artificial ground motion time history. When the power spectrum corresponding to the artificial ground motion time history is greater than or equal to αG T (ω), determines the completion of power spectrum matching; where,

[0056]

[0057] Where Ψ(t) is the shaping envelope function, Δa′ g (l) (t) is the incremental acceleration time history curve after envelope, a′ g (l) (t) is the time course curve of matching the target power spectrum.

[0058] Due to the adoption of the above technical scheme, the present invention has the following advantages and positive effects compared with the prior art: the intelligent generation method of artificial seismic time history provided by the present invention establishes a diffusion model of artificial seismic time history and trains the diffusion model, so that the trained diffusion model can output an artificial seismic time history curve that matches the response spectrum and the power spectrum when a standard Gaussian noise curve is input, which has the advantage of high matching accuracy. Moreover, since the diffusion process does not involve complex iterative calculations, the calculation efficiency of the artificial seismic time history can be greatly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 is a flow chart of a method for intelligently generating artificial earthquake time history in one embodiment of the present invention;

[0060] Figure 2 is the target response spectrum and its matching target limit;

[0061] Figure 3 is the target power spectrum and its matching target limit;

[0062] Figures 4 to 6 They are respectively the time history curve, target response spectrum and target power spectrum of the artificial ground motion time history before matching;

[0063] Figures 7 to 9 They are respectively the time history curve, target response spectrum and target power spectrum of the artificial ground motion time history after matching;

[0064] Fig.10 Schematic diagram of the diffusion process.

[0065] Fig.11 This is a schematic diagram of the network architecture of the diffusion model;

[0066] Figure 12 to Figure 14The time history curve, target response spectrum and target power spectrum of the artificial ground motion generated by the diffusion model respectively. DETAILED DESCRIPTION

[0067] The following is a further detailed description of an artificial seismic time history intelligent generation method provided by the present invention in conjunction with the accompanying drawings and specific embodiments. The advantages and features of the present invention will become clearer in conjunction with the following description. It should be noted that the accompanying drawings are all in a very simplified form and are not in precise proportions, and are only used to conveniently and clearly assist in explaining the purpose of the embodiments of the present invention.

[0068] like Figure 1 As shown, the method for intelligently generating an artificial earthquake time history provided in this embodiment includes the following steps:

[0069] S1. Establishing a diffusion model of the artificial earthquake time history, wherein the diffusion model adopts a neural network model based on machine learning, such as a U-Net neural network model.

[0070] S2. Obtain several artificial earthquake time history curves that can match the target response spectrum and the target power spectrum, and use Formula 1 to perform forward diffusion on each artificial earthquake time history curve to obtain a corresponding set of training data x 0 、x 1 , …, x T-1 、x T ; Among them, formula 1 is:

[0071]

[0072] In the formula, q(x t |x t-1 ) is a forward diffusion process, N(·) is a normal distribution, β t is a parameter related to the time step; t=1,2,…,T,x t 、x t-1 are the seismic time history curves of the tth and t-1th steps respectively, x 0 To match the ground motion time history curve of the target response spectrum and power spectrum at the same time, x T is the standard Gaussian noise curve, T is the total number of sampling steps, and I is the unit diagonal matrix.

[0073] In the forward diffusion in formula 1, each step adds a mean value of The variance is β t The noise is shown in the artificial ground motion time history curve x 0 Based on this, we can get x 1 、x 2 , …, x T-1 、x T . A linear increase method can be used, that is, βt =β min +(β max -β min )t / T, where β min It can take a small positive value, such as 0.0001 or smaller, β max You can take a larger positive value, such as 0.1. 0 、x 1 、x 2 , …, x T-1 、x T As a set of training data, several sets of training data can be obtained through step S2 to form a training data set for subsequent training of the diffusion model.

[0074] S3. Use the training data obtained in step S2 to train the diffusion model, input x t , output x t-1 ; The trained diffusion model can complete the reverse diffusion process in formula 2, and the diffusion model can predict the noise mean μ of the current step θ (x t ,t) and variance ∑ θ (x t ,t), complete the reverse denoising process; Formula 2 is:

[0075] p θ (x t-1 |x t )=N(x t-1 ;μ θ (x t ,t),∑ θ (x t ,t));

[0076] In the formula, p θ (x t-1 |x t ) is the reverse diffusion process of the diffusion model, μ represents the mean of the normal distribution, Σ is the covariance, θ is the parameter of the neural network, and x t is the artificial ground motion time history curve obtained in the t-th step diffusion process.

[0077] S4. A new standard Gaussian noise curve x T 'Input into the trained diffusion model, and output the artificial ground motion time history curve x that matches the target response spectrum and target power spectrum after T steps of reverse denoising process 0 '.

[0078] The T-step reverse denoising process can be expressed by Formula 3, which is:

[0079]

[0080] Where z is the noise of N(0,I).

[0081] The T-step reverse denoising process is as follows: a new standard Gaussian noise curve x T 'Input into the trained diffusion model, the diffusion model outputs the first step of denoising curve x T-1 ', then x T-1 'Input into the trained neural network, the diffusion model outputs the second step denoising curve x T-2 ', and so on, we finally get the artificial ground motion time history curve x that is completely denoised and has matching response spectrum and power spectrum with similar characteristics to the training data. 0 The above process is repeated to obtain a series of new artificial ground motion time history curves that match the response spectrum and power spectrum.

[0082] The intelligent generation method of artificial seismic time history provided in this embodiment establishes a diffusion model of the artificial seismic time history and trains the diffusion model so that the trained diffusion model can output an artificial seismic time history curve that matches the response spectrum and the power spectrum when a standard Gaussian noise curve is input. This method has the advantage of high matching accuracy. Moreover, since the diffusion process does not involve complex iterative calculations, the calculation efficiency of the artificial seismic time history can be greatly improved.

[0083] In a specific embodiment, a loss function is set in the diffusion model, and the loss function represents minimizing the difference between the noise predicted by the diffusion model and the actual noise at the current sampling step. The diffusion model is trained, and the parameters of the neural network are optimized using a gradient optimization method to minimize the loss function. The loss function is denoted as L, and satisfies:

[0084]

[0085] in, γ s =1-β s ,

[0086] Where L is the loss function of the neural network, ε is the real noise at time t, and ε θ is the noise predicted by the neural network, γ s is the weight adjustment factor, β s is a parameter related to the time step, s is the seismic time history curve of the sth step, s=1,2,…,t-1,t.

[0087] It should be noted that the artificial earthquake time history curve that can match the target response spectrum and the target power spectrum in step S2 can be matched using existing technologies. This embodiment provides a method for quickly matching the artificial earthquake time history curve with the target response spectrum and the target power spectrum, specifically comprising:

[0088] S21. Matching the artificial earthquake time history curve with the target response spectrum;

[0089] S22. Matching the artificial earthquake time history curve that has matched the target response spectrum in step S21 to the target power spectrum;

[0090] S23. Determine whether the response spectrum corresponding to the artificial earthquake time history curve that has matched the target power spectrum in step S22 is greater than or equal to the target response spectrum, if so, proceed to step S24, if not, loop through steps S21 to S23 for the artificial earthquake time history curve that has matched the target power spectrum;

[0091] S24. The artificial earthquake time history curve that has matched the target power spectrum can match the target response spectrum and the target power spectrum.

[0092] The artificial earthquake time history curve that matches the target response spectrum in step S21 does not necessarily match the target power spectrum in step S22. Similarly, the artificial earthquake time history curve that matches the target power spectrum in S22 does not necessarily match the target response spectrum in step S21. Therefore, the processes of step S21 and step S22 are performed alternately until the obtained artificial earthquake time history curve matches both the target power spectrum and the target response spectrum. Steps S21 and S22 are repeated to obtain a number of artificial earthquake time history curves that match both the target power spectrum and the target response spectrum, which are used for subsequent training of the diffusion model.

[0093] Furthermore, the artificial earthquake time history is matched with the target response spectrum, and the existing technology can be used for matching. This embodiment also provides a method for matching the artificial earthquake time history curve with the target response spectrum by a numerical iteration method, which specifically includes:

[0094] S211. For the initial random ground motion time history, the Duhamel integral is used to calculate the response of the single-degree-of-freedom system:

[0095]

[0096] in,

[0097] In the formula, a a (0) (t) is the response of the single-degree-of-freedom system, a g (0) (t) is the initial random earthquake time history, which can be randomly selected from the earthquake database, h Tn,ξ (t) is the unit impulse response function, ω d is the damped circular frequency, ω d =ω n / (1-ξ 2 ) 1 / 2 .

[0098] S212. Calculate the response spectrum corresponding to the initial earthquake motion time history:

[0099]

[0100] In the formula, S a (0) (T n ,ξ) is the natural oscillation period T n and the earthquake response spectrum under the damping ratio ξ, |a a (0) (t m )| is the maximum response of the single-degree-of-freedom system, t m It is the moment of greatest response.

[0101] S213. Calculate the difference between the earthquake time history response spectrum and the target response spectrum:

[0102]

[0103] Where sgn(·) is the sign function.

[0104] S214. Construct the narrow-band incremental time history curve of the single-degree-of-freedom system response based on the difference between the response spectra:

[0105] Δa a (t) = ΔS(T n ,ξ)s n (t),

[0106] in,

[0107] In the formula, Δa a (t) corresponds to ΔS(T n ,ξ) of the narrowband increment time history curve, s n (t) is a narrowband amplitude modulation function.

[0108] S215. According to the narrow-band incremental time history curve of the single-degree-of-freedom system response, the narrow-band incremental time history curve of the single-degree-of-freedom response is converted into a narrow-band incremental time history curve of the excitation based on the frequency response transfer function:

[0109]

[0110] In the formula, H a (ω) is the frequency response function of the single degree of freedom system, Δa g (t) is the narrowband incremental time history curve of the excitation, F(·) and F -1 (·) is the Fourier transform and inverse Fourier transform.

[0111] S216. Superimpose the excitation narrowband incremental time history curve on the original vibration time history so that the modified time history curve is within the natural vibration period T n The following is greater than or equal to the target response spectrum:

[0112]

[0113] In the formula, a g (1) (t) is the artificial ground motion time history obtained by the first iteration.

[0114] S217. Correct different natural oscillation periods T through steps S213 to S216 n The response spectrum corresponding to the ground motion time history is obtained to obtain the artificial ground motion time history a that matches the target response spectrum g (l) (t). When the response spectrum corresponding to the artificial earthquake time history is greater than or equal to the target response spectrum, the response spectrum matching is considered to be completed. Since the correction of a point on the response spectrum may affect other points on the response spectrum, causing other points not to match the target response spectrum, the correction process needs to be repeated many times until all points on the response spectrum match the target response spectrum, and the artificial earthquake time history a matching the target response spectrum is obtained. g (l) (t).

[0115] Furthermore, the artificial earthquake time history is matched with the target power spectrum, and the existing technology can be used for matching. This embodiment also provides a method for quickly matching the artificial earthquake time history curve with the target power spectrum, which specifically includes:

[0116] S221. Calculate the artificial earthquake time history a g (l) The power spectral density G of (t) g (l) (ω), calculate the power spectrum density G of the artificial ground motion time history g (l) (ω) and the target power spectrum density G T The difference between (ω)

[0117] Where α is the proportional coefficient of the interval [0,1], i = 1, 2, …, N-1, and N is the number of sampling points.

[0118] S222. The difference power spectrum ΔG g (l) (ω) is transformed into a Fourier amplitude spectrum, and then the Fourier amplitude spectrum is inversely transformed into an incremental acceleration time history curve:

[0119]

[0120] In the formula, |ΔA g (l) (ω i )| is the Fourier amplitude spectrum corresponding to the difference power spectrum, Δt is the sampling time interval, Δa g (l) (t) is the incremental acceleration time history curve.

[0121] S223. In order to prevent the incremental acceleration time history curve from affecting the shape of the artificial ground motion time history, the incremental acceleration time history curve is shaped and then superimposed on the artificial ground motion time history, and the power spectrum corresponding to the artificial ground motion time history is greater than or equal to αG T (ω), it is considered that the matching of the power spectrum is completed.

[0122]

[0123] Where Ψ(t) is the shaping envelope function, Δa′ g (l) (t) is the incremental acceleration time history curve after envelope, a′ g (l) (t) is the time course curve of matching the target power spectrum.

[0124] As an example, Figure 2 A target response spectrum and its corresponding matching target limit are given. Figure 3 The target power spectrum and its corresponding matching target limit are given, and the generated artificial ground motion time history needs to be within the matching target limit range. Figure 2 , CP, ULB, and LLB represent the control point, upper boundary, and lower boundary, respectively. Figure 3 In the above, only CP and LLB are involved, namely control point and lower limit boundary. Figure 4 is the initial artificial seismic time history curve, that is, the artificial seismic time history curve that does not match the target response spectrum and target power spectrum. Figure 5 Schematic diagram showing the response spectrum of the initial artificial ground motion time history curve and the target response spectrum. Figure 6 Schematic diagram showing the power spectrum of the initial artificial ground motion time history curve and the target power spectrum. Figure 7 To match the artificial ground motion time history curve after the target response spectrum and target power spectrum, Figure 8 A schematic diagram showing the response spectrum of the matched artificial ground motion time history curve and the target response spectrum. Fig. 9 Schematic diagram showing the power spectrum of the matched artificial ground motion time history curve and the target power spectrum. Figure 5 and Figure 8 RS and TD RS in the equation represent the response spectrum and target response spectrum, respectively. Figure 6 and Fig. 9 The PSD and TPSD in it represent the power spectrum and the target power spectrum respectively.

[0125] As an example, Fig.10 A diffusion process of a time history curve is given. From left to right is the forward diffusion process of the diffusion model. Through gradual diffusion, the noise in the time history curve is gradually eliminated, and an artificial seismic time history that meets the requirements is obtained; from right to left is the reverse diffusion process of the diffusion model. Through the reverse diffusion process, the noise is gradually increased in the seismic time history until a white noise time history curve is obtained.

[0126] As an example, Fig.11 A U-Net neural network architecture model is given. This network architecture model compresses the tensor feature extraction process at the same time, and then upsamples the compressed feature tensor to regenerate a tensor with new features, which is suitable for various generation tasks. Fig.10 In the figure, Conv, Res, DS, Att and US represent convolutional layer, residual block, downsampling layer, self-attention layer and upsampling layer respectively. The network consists of upsampling block (US block), middle block (Middleblock), downsampling block (DS block), and a skip connection (Skipconnection) is established between upsampling block and downsampling block. The input of the network is the seismic time history curve of a certain diffusion step, and the output is the seismic time history curve of the next diffusion step. It should be pointed out that Fig.11 The network model architecture given is not the only one. Here we only give one feasible form of the network model architecture.

[0127] As an example, Fig.12 A time history curve of artificial ground motion generated by the diffusion model is given, which completely matches the target response spectrum and target power spectrum. Fig.13 A schematic diagram showing the response spectrum of the artificial ground motion time history curve and the target response spectrum. Fig.14 A schematic diagram showing the power spectrum of the artificial ground motion time history curve and the target power spectrum. Fig.12 The time history curve of artificial ground motion generated by the diffusion model is Figure 3 However, when the artificial seismic time history is generated by using the trained diffusion model, the diffusion process of the diffusion model does not involve complex calculations and iterations but only involves a sampling process. Therefore, the speed of generating artificial seismic time history is much higher than that of the numerical algorithm.

[0128] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0129] The above-mentioned embodiments only express several implementation modes of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the invention. It should be pointed out that, for a person of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the attached claims.

Claims

1. A method for intelligently generating artificial earthquake time history, characterized in that: The steps include: S1. Establishing a diffusion model of artificial earthquake time history, wherein the diffusion model adopts a neural network model based on machine learning; S2. Obtain several artificial earthquake time history curves that can match the target response spectrum and the target power spectrum, and use Formula 1 to perform forward diffusion on each artificial earthquake time history curve to obtain a corresponding set of training data x0, x1, ..., x T-1 、x T ; Among them, formula 1 is: In the formula, q(x t |x t-1 ) is a forward diffusion process, N(·) is a normal distribution, β t is a parameter related to the time step; t=1,2,…,T,x t 、x t-1 are the seismic time history curves of the tth and t-1th steps respectively, x0 is the seismic time history curve that matches the target response spectrum and power spectrum at the same time, x T is the standard Gaussian noise curve, T is the total number of sampling steps, and I is the unit diagonal matrix; S3. Use the training data obtained in step S2 to train the diffusion model, input x t , output x t-1 ; The trained diffusion model can complete the reverse diffusion process in formula 2; formula 2 is: p θ (x t-1 |x t )=N(x t-1 ;μ θ (x t ,t),∑ θ (x t ,t)); In the formula, p θ (x t-1 |x t ) is the reverse diffusion process of the diffusion model, μ represents the mean of the normal distribution, Σ is the covariance, θ is the parameter of the neural network, and x t is the artificial ground motion time history curve obtained from the t-step diffusion process; S4. A new standard Gaussian noise curve x T 'Input into the trained diffusion model, the diffusion model outputs the artificial ground motion time history curve x0' that matches the response spectrum and power spectrum.

2. The method for intelligently generating artificial earthquake time history according to claim 1, characterized in that: A loss function is set in the diffusion model, which is denoted by L and satisfies: in, Where ε is the real noise at time t, ε θ is the noise predicted by the neural network, is the weight factor accumulated according to the time step, γ s is the weight adjustment factor, β s is a parameter related to the time step.

3. The method for intelligently generating artificial earthquake time history according to claim 1, characterized in that: In step S4, the diffusion model is used to T 'Perform T-step reverse denoising. The T-step reverse denoising process is expressed as: In the formula, is the weight factor accumulated according to the time step, σ t is the covariance of the noise z, where z is the noise of N(0,I).

4. The method for intelligently generating artificial earthquake time history according to any one of claims 1 to 3, characterized in that: In step S2, obtaining an artificial ground motion time history curve that can match the target response spectrum and the target power spectrum specifically includes: S21. Matching the artificial earthquake time history curve with the target response spectrum; S22. Matching the artificial earthquake time history curve that has matched the target response spectrum in step S21 to the target power spectrum; S23. Determine whether the response spectrum corresponding to the artificial earthquake time history curve that has matched the target power spectrum in step S22 is greater than or equal to the target response spectrum, if so, proceed to step S24, if not, loop through steps S21 to S23 for the artificial earthquake time history curve that has matched the target power spectrum; S24. The artificial earthquake time history curve that has matched the target power spectrum can match the target response spectrum and the target power spectrum.

5. The method for intelligently generating artificial earthquake time history according to claim 4, characterized in that: In step S21, the artificial earthquake time history curve is matched with the target response spectrum, which specifically includes: S211. For the initial random ground motion time history, the Duhamel integral is used to calculate the response of the single-degree-of-freedom system: in, In the formula, a a (0) (t) is the response of the single-degree-of-freedom system, a g (0) (t) is the initial random earthquake time history, which can be randomly selected from the earthquake database, h Tn,ξ (t) is the unit impulse response function, ω d is the damped circular frequency, ω d =ω n / (1-ξ 2 ) 1 / 2 ; S212. Calculate the response spectrum corresponding to the initial earthquake motion time history: In the formula, S a (0) (T n ,ξ) is the natural oscillation period T n and the earthquake response spectrum under the damping ratio ξ, |a a (0) (t m )| is the maximum response of the single degree of freedom system, t m It is the moment of greatest response; S213. Calculate the difference between the earthquake time history response spectrum and the target response spectrum: Where sgn(·) is the sign function; S214. Construct the narrow-band incremental time history curve of the single-degree-of-freedom system response based on the difference between the response spectra: Δa a (t)=ΔS(T n ,ξ)s n (t), in, In the formula, Δa a (t) corresponds to ΔS(T n ,ξ) of the narrowband increment time history curve, s n (t) is a narrowband amplitude modulation function; S215. According to the narrow-band incremental time history curve of the single-degree-of-freedom system response, the narrow-band incremental time history curve of the single-degree-of-freedom response is converted into a narrow-band incremental time history curve of the excitation based on the frequency response transfer function: In the formula, H a (ω) is the frequency response function of the single degree of freedom system, Δa g (t) is the narrowband incremental time history curve of the excitation, F(·) and F -1 (·) is Fourier transform and inverse Fourier transform; S216. Superimpose the excitation narrowband incremental time history curve on the original vibration time history so that the modified time history curve is within the natural vibration period T n is greater than or equal to the target response spectrum, where In the formula, a g (1) (t) is the artificial ground motion time history obtained in the first iteration; S217. Correct different natural oscillation periods T through steps S213 to S216 n The response spectrum corresponding to the ground motion time history is obtained to obtain the artificial ground motion time history a that matches the target response spectrum g (l) (t).

6. The method for intelligently generating artificial earthquake time history according to claim 5, characterized in that: In step S22, the artificial ground motion time history curve that has been matched with the target response spectrum in step S21 is matched with the target power spectrum, which specifically includes: S221. Calculate the artificial earthquake time history a g (l) The power spectral density G of (t) g (l) (ω), calculate the power spectrum density G of the artificial ground motion time history g (l) (ω) and the target power spectrum density G T The difference between (ω) Where α is the proportional coefficient of the interval [0,1], i = 1, 2, ..., N-1, and N is the number of sampling points; S222. The difference power spectrum ΔG g (l) (ω) is transformed into a Fourier amplitude spectrum, and then the Fourier amplitude spectrum is inversely transformed into an incremental acceleration time history curve: In the formula, |ΔA g (l) (ω i )| is the Fourier amplitude spectrum corresponding to the difference power spectrum, Δt is the sampling time interval, Δa g (l) (t) is the incremental acceleration time history curve; S223. The incremental acceleration time history curve is shaped and then superimposed on the artificial ground motion time history. When the power spectrum corresponding to the artificial ground motion time history is greater than or equal to αG T (ω), determines the completion of power spectrum matching; where, Where Ψ(t) is the shaping envelope function, Δa′ g (l) (t) is the incremental acceleration time history curve after envelope, a′ g (l) (t) is the time course curve of matching the target power spectrum.

Images