A downhole seismic motion time history inversion method and system based on closed-loop reconstruction constraints

Abstract

The application relates to the technical field of earthquake engineering, and discloses a downhole ground motion time history inversion method and system based on closed-loop reconstruction constraints, which comprises the following steps: acquiring measured ground motion acceleration time history, downhole ground motion acceleration time history and site parameters; performing pretreatment to obtain a data set; inputting the data set into an inversion branch of a time sequence neural network model to obtain predicted downhole ground motion acceleration time history; inputting the predicted downhole ground motion acceleration time history into a forward reconstruction branch of the time sequence neural network model to obtain reconstructed ground motion acceleration time history; constructing a joint loss function; training the time sequence neural network model by using the joint loss function to obtain a joint prediction model; and inputting the ground motion acceleration time history and the site parameters of a target site into the joint prediction model to obtain a predicted downhole ground motion acceleration time history result. By using the application, the stability, physical consistency and engineering applicability of downhole ground motion inversion are improved.

Description

Technical Field

[0001] This invention relates to the field of earthquake engineering technology, specifically to a method and system for downhole ground motion time history inversion based on closed-loop reconstruction constraints. Background Technology

[0002] Seismic waves propagating from bedrock to the surface are affected by site conditions, resulting in significant amplification, filtering, and nonlinear effects, thereby altering the spectral structure and temporal characteristics of surface ground motions. Existing research shows significant differences between downhole and surface ground motion records; under strong earthquake conditions, surface ground motions also exhibit nonlinear characteristics such as attenuation of high-frequency components and prolongation of dominant periods. Traditional site response analysis methods primarily employ numerical methods, but these methods are sensitive to the selected soil parameters and dynamic constitutive models of the soil mass, and are prone to misjudgment of seismic response when parameter uncertainties are large. To address these issues, existing research has utilized a large number of seismic records and employed artificial intelligence algorithms to establish surface ground motion time-history prediction models, mainly focusing on the one-way prediction problem from downhole to surface. However, research on inverting downhole ground motions when only surface stations are deployed, compensating for missing downhole records, and improving inversion stability through closed-loop reconstruction has not yet been addressed. Therefore, we have invented a downhole ground motion time-history inversion method and system based on closed-loop reconstruction constraints, solving the above-mentioned technical problems. Summary of the Invention

[0003] This invention provides a method and system for downhole ground motion time history inversion based on closed-loop reconstruction constraints. By introducing surface-downhole-surface bidirectional reconstruction constraints and frequency domain physical constraints, the stability, physical consistency and engineering applicability of downhole ground motion inversion are improved.

[0004] Therefore, the present invention provides the following technical solution: A method for downhole ground motion time history inversion based on closed-loop reconstruction constraints, the method comprising: Step 1: Obtain the measured time history of surface ground motion acceleration, the measured time history of downhole ground motion acceleration, and the measured site parameters. Step 2: Preprocess the measured surface ground motion acceleration time history, the measured downhole ground motion acceleration time history, and the measured site parameters to obtain a dataset; Step 3: Input the dataset into the inversion branch of the time-series neural network model to obtain the predicted downhole ground motion acceleration time history; Step 4: Input the predicted downhole ground motion acceleration time history into the forward reconstruction branch of the time-series neural network model to obtain the reconstructed surface ground motion acceleration time history; Step 5: Based on the predicted downhole ground motion acceleration time history, the reconstructed surface ground motion acceleration time history, and the Fourier spectrum ratio of the surface and downhole ground motion acceleration time histories, construct a joint loss function. Step 6: Train the temporal neural network model using the joint loss function to obtain the joint prediction model; Step 7: Input the ground motion acceleration time history of the target site and the site parameters of the target site into the joint prediction model to obtain the predicted downhole ground motion acceleration time history results.

[0005] Optionally, in step 1, the measured site parameters include the average shear wave velocity V at a depth of 30 meters near the ground surface. S30 and shear wave velocity profile gradient parameter B 30 The shear wave velocity profile gradient parameter B 30 The shear wave velocity logarithm and depth logarithm within a 30-meter depth range near the surface of the site are obtained by linear regression and used to characterize the gradient characteristics of soil stiffness with depth. The measured ground ground acceleration time history and the measured underground ground ground acceleration time history both include east-west and north-south components. The measured site parameters are extended to have the same time series length as the measured ground ground acceleration time history, and the measured site parameters and the measured ground ground acceleration time history together form the input tensor of the time series neural network model.

[0006] Optionally, in step 2, the preprocessing includes one or more of filtering, truncation, zero padding, interpolation, and normalization.

[0007] Optionally, in step 3, the temporal neural network model includes a shared temporal encoder, a downhole inversion decoder, and a surface forward reconstruction decoder; the shared temporal encoder employs a Long Short-Term Memory (LSTM) network, a Bidirectional Long Short-Term Memory (BiLSTM) network, a Transformer, or a combination thereof; and the temporal neural network model is trained using the Adam optimizer.

[0008] Optionally, in step 5, the joint loss function includes a downhole ground motion acceleration time history prediction loss term, a surface ground motion acceleration time history reconstruction loss term, and a frequency domain physical constraint loss term.

[0009] Optionally, in step 5, the downhole ground motion acceleration time history prediction loss term is determined by comparing the time-by-time difference between the measured downhole ground motion acceleration time history and the predicted downhole ground motion acceleration time history; the time-by-time difference is quantified using mean square error and calculated on the normalized time series.

[0010] Optionally, in step 5, the surface ground acceleration time history reconstruction loss term is determined by comparing the time-by-time difference between the measured surface ground acceleration time history and the reconstructed surface ground acceleration time history; the time-by-time difference is quantified using mean square error and calculated on the normalized time history sequence.

[0011] Optionally, in step 5, the reference Fourier spectrum ratio of the measured surface ground motion acceleration time history and the measured downhole ground motion acceleration time history under strong and weak earthquake conditions is calculated. and Calculate the benchmark ratio of the site's nonlinear to linear response. The dynamic Fourier spectrum ratio of the measured surface ground motion acceleration time history and the predicted downhole ground motion acceleration time history under strong and weak earthquake conditions. and Calculate the dynamic ratio of the site's nonlinear to linear response. The frequency domain physical constraint loss term is obtained by comparing the benchmark ratio during training. The dynamic ratio The differences are determined and quantified using mean square error.

[0012] A downhole ground motion time history inversion system based on closed-loop reconstruction constraints, the system comprising: The data acquisition module acquires the measured time histories of surface ground motion acceleration, the measured time histories of downhole ground motion acceleration, and the measured site parameters. The preprocessing module preprocesses the measured surface ground motion acceleration time history, the measured downhole ground motion acceleration time history, and the measured site parameters to obtain a dataset. The downhole data prediction module inputs the dataset into the inversion branch of the time-series neural network model to obtain the predicted downhole ground motion acceleration time history; The surface data reconstruction module inputs the predicted downhole ground motion acceleration time history into the forward reconstruction branch of the time-series neural network model to obtain the reconstructed surface ground motion acceleration time history. The loss function construction unit constructs a joint loss function based on the predicted downhole ground motion acceleration time history, the reconstructed surface ground motion acceleration time history, and the Fourier spectrum ratio of the surface and downhole ground motion acceleration time histories. The joint prediction model construction unit uses the joint loss function to train the temporal neural network model to obtain the joint prediction model; The result output unit inputs the ground motion acceleration time history of the target site and the site parameters of the target site into the joint prediction model to obtain the predicted downhole ground motion acceleration time history results.

[0013] A computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor to perform the steps of the downhole ground motion time history inversion method based on closed-loop reconstruction constraints.

[0014] This invention provides a method and system for downhole ground motion time history inversion based on closed-loop reconstruction constraints. Addressing the shortcomings of existing ground motion time history prediction methods, which primarily focus on unidirectional prediction from downhole to surface, are difficult to apply to scenarios with missing downhole records, suffer from insufficient stability in solving inverse problems, and lack adequate constraints on site nonlinear characteristics, this invention jointly inputs surface ground motion acceleration time histories and site parameters into a time-series neural network model. While achieving surface-to-downhole ground motion inversion, it establishes closed-loop constraints through forward reconstruction from downhole to surface, and combines this with frequency domain physical constraints to improve the stability, physical consistency, and engineering applicability of downhole ground motion acceleration time history prediction results. Compared to traditional unidirectional prediction methods, this invention improves the stability, physical consistency, and engineering applicability of downhole ground motion inversion by introducing surface-downhole-surface bidirectional reconstruction constraints and frequency domain physical constraints. It can be used for downhole record missing compensation, site nonlinear response analysis, and earthquake early warning auxiliary decision-making. Compared with existing technologies, this invention has at least the following beneficial effects: (1) This invention extends the traditional one-way prediction to inversion prediction from the surface to the well and forward reconstruction from the well to the surface. It can verify whether the prediction results meet the seismic wave propagation logic while inverting the ground motion in the well, effectively reducing the multiple solutions and instability in the inverse problem.

[0015] (2) The present invention will use V S30 and B 30 Using the combined parameters as site parameter inputs, it can not only characterize the average stiffness characteristics of the site surface, but also characterize the variation law of soil stiffness along the depth direction, thereby improving the model's ability to express the characteristics of ground motion propagation under different site conditions.

[0016] (3) This invention introduces a method based on and The frequency domain physical constraint loss term enables the prediction results to not only approximate the target results in the time domain, but also conform to the site seismic response mechanism in the frequency domain and nonlinear characteristics, thereby improving the physical interpretability and engineering credibility of the model.

[0017] (4) This invention is applicable to downhole ground motion inversion, site nonlinear response analysis and earthquake early warning auxiliary decision-making under the condition of missing downhole records, and has strong application and promotion value. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly described below. Obviously, the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the downhole ground motion time history inversion method based on closed-loop reconstruction constraints as described in this embodiment of the invention. Figure 2 This is a schematic diagram of the temporal neural network structure of the present invention; Figure 3 This is a schematic diagram illustrating the joint loss function of the present invention; Figure 4 This is a schematic diagram of the surface-underground-surface reconstruction closed loop in this invention; Figure 5 This is a schematic diagram comparing the predicted downhole ground motion acceleration time history with the measured downhole ground motion acceleration time history in an embodiment of the present invention (unit: cm / s). 2 ); Figure 6 This is a schematic diagram of the structure of a downhole ground motion time history inversion system based on closed-loop reconstruction constraints in a specific embodiment of the present invention. Detailed Implementation

[0020] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0021] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0022] like Figure 1 As shown, Figure 1 This is a flowchart of a downhole ground motion time history inversion method based on closed-loop reconstruction constraints in a specific embodiment of the present invention. The downhole ground motion time history inversion method based on closed-loop reconstruction constraints includes: S1. Obtain the measured time histories of ground motion acceleration on the ground surface and underground, as well as the corresponding site parameters.

[0023] The site parameters include at least the average shear wave velocity V at a depth of 30 meters from the surface of the site. S30and shear wave velocity profile gradient parameter B 30 The shear wave velocity profile gradient parameter B 30 The shear wave velocity logarithm and depth logarithm within a 30-meter depth range near the surface of the site are obtained through linear regression and used to characterize the gradient characteristics of soil stiffness with depth. The time histories of ground motion acceleration at the surface and underground include at least east-west and north-south components, and the measured site parameters are extended to have the same time series length as the measured ground motion acceleration time histories. The measured site parameters and the measured ground motion acceleration time histories together form the input tensor of the time-series neural network model.

[0024] S2. Preprocess the measured ground motion acceleration time histories and corresponding site parameters to obtain a dataset.

[0025] The preprocessing includes one or more of the following: filtering, truncation, zero padding, interpolation, and normalization.

[0026] S3. Input the dataset into the inversion branch of the time-series neural network model and output the predicted downhole ground motion acceleration time history.

[0027] like Figure 2 As shown, the inversion branch of the time-series neural network model includes an input layer, a long short-term memory network layer, a flattened layer, a fully connected layer, and an output layer. The input layer includes the measured ground ground acceleration time history and site parameters, and the output layer is the predicted downhole ground acceleration time history.

[0028] The temporal neural network model includes a shared temporal encoder, a downhole inversion decoder, and a surface forward reconstruction decoder. The shared temporal encoder employs a Long Short-Term Memory (LSTM) network, a Bidirectional Long Short-Term Memory (BiLSTM) network, a Transformer, or a combination thereof.

[0029] The Adam optimizer is used when training the temporal neural network model.

[0030] S4. Input the predicted downhole ground motion acceleration time history into the forward reconstruction branch of the time-series neural network model, and output the reconstructed surface ground motion acceleration time history.

[0031] S5. Construct a joint loss function based on the predicted downhole ground motion acceleration time history, the reconstructed surface ground motion acceleration time history, and the reference spectral ratio characteristics; the joint loss function includes a downhole ground motion acceleration time history prediction loss term, a surface ground motion acceleration time history reconstruction loss term, and a frequency domain physical constraint loss term; such as Figure 3 As shown, the total loss function is calculated by weighting the downhole ground motion acceleration time history prediction loss term, the surface ground motion time history reconstruction loss term, and the frequency domain physical constraint loss term.

[0032] The frequency domain physical constraint loss term is constructed based on the Fourier amplitude spectrum ratio (BSR) of the surface ground acceleration time history and the downhole ground acceleration time history. The frequency domain physical constraint loss term is further constructed based on the ratio between the strong earthquake spectral ratio and the weak earthquake reference spectral ratio to construct the nonlinear and linear response ratio (RSR). NL The system was constructed to constrain the nonlinear characteristics of the site. The Fourier amplitude spectrum was smoothed using Konno-Ohmachi before calculation.

[0033] The loss term for the prediction of underground ground motion acceleration time history is determined by comparing the time-by-time difference between the measured underground ground motion acceleration time history and the predicted underground ground motion acceleration time history; the time-by-time difference is quantified using mean square error and can be calculated on the normalized time series.

[0034] The surface ground motion time history reconstruction loss term is determined by comparing the time-by-time differences between the measured surface ground motion acceleration time history and the reconstructed surface ground motion acceleration time history; the time-by-time differences are quantified using mean square error and can be calculated on the normalized time series.

[0035] like Figure 4 As shown, the reconstructed surface ground motion acceleration time history is obtained from the predicted downhole ground motion acceleration time history through the forward reconstruction branch of the time-series neural network, and the predicted downhole ground motion acceleration time history is obtained from the measured surface ground motion acceleration time history through the inversion branch of the time-series neural network.

[0036] S6. Train the temporal neural network model using the joint loss function to obtain the joint prediction model; like Figure 3 As shown, during the training process, the weighted sum of each loss term is used, and the optimizer is employed to minimize the total loss, iteratively updating the parameters of the temporal neural network.

[0037] S7. Input the ground motion acceleration time history of the target site and the corresponding site parameters into the joint prediction model, and output the predicted downhole ground motion acceleration time history results.

[0038] like Figure 5 As shown, in different target sites, the predicted downhole ground motion acceleration time history results are consistent with the measured downhole ground motion acceleration time history waveforms, and the maximum error is less than 8%.

[0039] The method is used for downhole ground motion time history reconstruction, site nonlinear response analysis, or earthquake early warning auxiliary decision-making under conditions of missing downhole records.

[0040] Accordingly, embodiments of the present invention also provide a downhole ground motion time history inversion system based on closed-loop reconstruction constraints, such as... Figure 6The image shows a schematic diagram of the system. This downhole ground motion time history inversion system based on closed-loop reconstruction constraints includes the following modules: The data acquisition module 601 acquires the measured time history of ground ground acceleration, the measured time history of ground ground acceleration, and the measured site parameters. The preprocessing module 602 preprocesses the measured surface ground motion acceleration time history, the measured downhole ground motion acceleration time history, and the measured site parameters to obtain a dataset. The downhole data prediction module 603 inputs the dataset into the inversion branch of the time-series neural network model to obtain the predicted downhole ground motion acceleration time history; The surface data reconstruction module 604 inputs the predicted downhole ground motion acceleration time history into the forward reconstruction branch of the time-series neural network model to obtain the reconstructed surface ground motion acceleration time history. The loss function construction unit 605 constructs a joint loss function based on the predicted downhole ground motion acceleration time history, the reconstructed surface ground motion acceleration time history, and the reference spectral ratio characteristics. The joint prediction model construction unit 606 uses the joint loss function to train the temporal neural network model to obtain the joint prediction model; The result output unit 607 inputs the ground motion acceleration time history of the target site and the site parameters of the target site into the joint prediction model to obtain the predicted downhole ground motion acceleration time history results.

[0041] In a specific embodiment of the present invention, the downhole ground motion time history inversion based on closed-loop reconstruction constraints includes the following steps: Step 101: Obtain the measured time histories of ground motion accelerations at the surface and underground, along with corresponding site parameters. The site parameters include at least the average shear wave velocity V30m near the surface. S30 and shear wave velocity profile gradient parameter B 30 .

[0042] V S30 It can be determined by the following formula: ; in, For the first Soil layer thickness, For the first Layer shear wave velocity, This refers to the total number of soil layers within 30m of the ground surface.

[0043] B 30 It can be determined by the following formula: ; in, Within 30m of the ground surface of the site and The slope obtained from linear regression, To correspond to the regression standard deviation, Let be the shear wave velocity, for the j Soil layer by layer , Let be the vertical distance from the site surface to a certain underground location. If we calculate the distance from the site surface to the [missing information] underground location... j The vertical distance between the surfaces of the soil layers is the front. j- The total thickness of one soil layer.

[0044] In a preferred embodiment, the time histories of ground motion accelerations at the surface and underground include east-west and north-south components. Because V S30 and B 30 As a scalar, it is extended to the same length as the time sequence, and then combined with the ground motion acceleration time history to form the model input tensor.

[0045] Step 102: Perform frequency domain transformation on the time histories of ground motion acceleration at the surface and underground to construct frequency domain physical constraints.

[0046] Earthquake acceleration time history a ( t The Fourier amplitude spectrum of ) can be written as: ; in, a ( t ( ) represents the measured time history of ground motion acceleration at the surface, in the well, or as predicted. Its Fourier spectrum, t For time, f Here, i represents the frequency, i is the imaginary unit, and π is the mathematical constant pi.

[0047] The reference Fourier spectrum ratio of measured surface ground motion acceleration time history and measured downhole ground motion acceleration time history under strong and weak earthquake conditions. and It can be written as: ; ; in, F (·) represents the smoothed Fourier amplitude spectrum of ground motion acceleration; S represents the surface, B represents the downhole; EW and NS represent the east-west and north-south components, respectively; Mea represents the measured data; NL indicates that the site is nonlinear under strong earthquake conditions, and L indicates that the site is linear under weak earthquake conditions. Konno-Ohmachi smoothing is preferred for the Fourier amplitude spectrum.

[0048] Correspondingly, under strong and weak earthquake conditions, the dynamic Fourier spectrum ratio of the measured surface ground motion acceleration time history and the predicted downhole ground motion acceleration time history is... and It can be written as: ; ; Here, Pre represents the predicted data.

[0049] Site nonlinearity and linear response ratio RSR NL Used to measure the nonlinear variation characteristics of a site under strong earthquakes relative to a weak earthquake benchmark, the benchmark Fourier spectral ratio is the ratio of the site's nonlinear to its linear response to the benchmark. It can be written as: ; in, N iw ≥5 represents the number of minor earthquakes.

[0050] For the dynamic Fourier spectral ratio, it is the dynamic ratio of the site nonlinearity to the linear response. It can be written as: ; Step 103: Construct a temporal neural network containing an inversion branch and a forward reconstruction branch. Specifically: The inversion branch is used to receive the measured time histories of ground motion acceleration and V. S30 B 30 Output the predicted downhole ground motion acceleration time history; The forward reconstruction branch is used to receive the predicted downhole ground motion acceleration time history and output the reconstructed surface ground motion acceleration time history; The shared encoder is used to perform joint feature extraction of ground motion acceleration time history and site parameters.

[0051] In a preferred embodiment, both the inversion branch and the forward reconstruction branch are implemented using LSTM.

[0052] Step 104: Calculation of the downhole ground motion acceleration time history prediction loss term. To constrain the difference between the predicted downhole ground motion time history and the measured downhole ground motion time history, a downhole ground motion acceleration time history prediction loss term is constructed. : ; in, These are the predicted time histories of east-west and north-south ground motion accelerations in the well, respectively. These are the measured time histories of east-west and north-south ground motion accelerations in the well, respectively. N For the sample size, TThe number of sampling points. λ EW and λ NS The directional weights are used to predict the loss.

[0053] Step 105: Calculation of the surface ground motion acceleration time history reconstruction loss term. To establish a closed-loop consistency between the "surface-downhole-surface" system, the difference between the measured surface ground motion acceleration time history and the reconstructed surface ground motion acceleration time history obtained by remapping the predicted downhole ground motion acceleration time history is compared, and the surface ground motion acceleration time history reconstruction loss term is constructed. : ; in, These are the reconstructed time histories of east-west and north-south oriented ground motion accelerations, respectively. These are the measured time histories of east-west and north-south ground motion accelerations, respectively. η EW and η NS The directional weights are used to reconstruct the loss.

[0054] Preferably, the loss can be calculated on the normalized time sequence to reduce the impact of different peak samples on training.

[0055] Step 106: Calculate the frequency domain physical constraint loss term, using the benchmark ratio of site nonlinearity to linear response. The dynamic ratio of site nonlinearity to linear response Forming frequency domain physical constraint loss : ; in, M The number of discrete frequency points.

[0056] Step 107: Combine loss function with model training.

[0057] like Figure 3 As shown, the total loss function is constructed by weighted summation of the above loss terms. : ; in, These are non-negative weighting coefficients.

[0058] In a preferred embodiment, the total loss function is minimized by the Adam optimizer, and the model parameters are iteratively updated until the loss converges, thus obtaining the trained surface-downhole joint prediction model.

[0059] After training, the time history of ground motion acceleration and site parameters V of the target site will be recorded.S30 B 30 By inputting a time-series neural network model, the system can output the time history prediction results of underground ground motion at the target site. These results can be used for underground ground motion time history compensation in scenarios where downhole sensors are unavailable, bedrock ground motion recovery, site nonlinear response analysis, seismic input ground motion structures for structural seismic resistance, and earthquake early warning and decision support.

[0060] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0061] The present invention also provides a storage medium, which is a computer-readable storage medium storing a computer program thereon, the computer program being executable when it runs. Figure 1 The method shown may include some or all of the steps. The storage medium may include read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk, etc. The storage medium may also include non-volatile memory or non-transitory memory, etc.

[0062] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data provider to another website, computer, server, or data provider via wired or wireless means.

[0063] The embodiments of the present invention have been described in detail above. Specific implementation methods have been used to illustrate the present invention. The descriptions of the embodiments above are only for the purpose of helping to understand the methods and systems of the present invention, and are merely some, not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention, and the content of this specification should not be construed as a limitation of the present invention. Therefore, any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for downhole ground motion time history inversion based on closed-loop reconstruction constraints, characterized in that, The method includes: Step 1: Obtain the measured time history of surface ground motion acceleration, the measured time history of downhole ground motion acceleration, and the measured site parameters. Step 2: Preprocess the measured surface ground motion acceleration time history, the measured downhole ground motion acceleration time history, and the measured site parameters to obtain a dataset; Step 3: Input the dataset into the inversion branch of the time-series neural network model to obtain the predicted downhole ground motion acceleration time history; Step 4: Input the predicted downhole ground motion acceleration time history into the forward reconstruction branch of the time-series neural network model to obtain the reconstructed surface ground motion acceleration time history; Step 5: Based on the predicted downhole ground motion acceleration time history, the reconstructed surface ground motion acceleration time history, and the Fourier spectrum ratio of the surface and downhole ground motion acceleration time histories, construct a joint loss function. Step 6: Train the temporal neural network model using the joint loss function to obtain the joint prediction model; Step 7: Input the ground motion acceleration time history of the target site and the site parameters of the target site into the joint prediction model to obtain the predicted downhole ground motion acceleration time history results.

2. The downhole ground motion time history inversion method based on closed-loop reconstruction constraints according to claim 1, characterized in that, In step 1, the measured site parameters include the average shear wave velocity V at a depth of 30 meters near the ground surface. S30 and shear wave velocity profile gradient parameter B 30 The shear wave velocity profile gradient parameter B 30 The shear wave velocity logarithm and depth logarithm within a 30-meter depth range near the surface of the site are obtained by linear regression and used to characterize the gradient characteristics of soil stiffness with depth. The measured ground ground acceleration time history and the measured underground ground ground acceleration time history both include east-west and north-south components. The measured site parameters are extended to have the same time series length as the measured ground ground acceleration time history, and the measured site parameters and the measured ground ground acceleration time history together form the input tensor of the time series neural network model.

3. The downhole ground motion time history inversion method based on closed-loop reconstruction constraints according to claim 1, characterized in that, In step 2, the preprocessing includes one or more of the following: filtering, truncation, zero padding, interpolation, and normalization.

4. The downhole ground motion time history inversion method based on closed-loop reconstruction constraints according to claim 1, characterized in that, In step 3, the temporal neural network model includes a shared temporal encoder, a downhole inversion decoder, and a surface forward reconstruction decoder; the shared temporal encoder uses a Long Short-Term Memory (LSTM) network, a Bidirectional Long Short-Term Memory (BiLSTM) network, a Transformer, or a combination thereof; and the temporal neural network model is trained using the Adam optimizer.

5. The downhole ground motion time history inversion method based on closed-loop reconstruction constraints according to claim 1, characterized in that, In step 5, the joint loss function includes a downhole ground motion acceleration time history prediction loss term, a surface ground motion acceleration time history reconstruction loss term, and a frequency domain physical constraint loss term.

6. The downhole ground motion time history inversion method based on closed-loop reconstruction constraints according to claim 5, characterized in that, In step 5, the downhole ground motion acceleration time history prediction loss term is determined by comparing the time-by-time difference between the measured downhole ground motion acceleration time history and the predicted downhole ground motion acceleration time history; the time-by-time difference is quantified using mean square error and calculated on the normalized time series.

7. The downhole ground motion time history inversion method based on closed-loop reconstruction constraints according to claim 5, characterized in that, In step 5, the surface ground acceleration time history reconstruction loss term is determined by comparing the time-by-time difference between the measured surface ground acceleration time history and the reconstructed surface ground acceleration time history; the time-by-time difference is quantified using mean square error and calculated on the normalized time history sequence.

8. The downhole ground motion time history inversion method based on closed-loop reconstruction constraints according to claim 5, characterized in that, In step 5, the reference Fourier spectrum ratio of the measured surface ground motion acceleration time history and the measured downhole ground motion acceleration time history under strong and weak earthquake conditions is calculated. and Calculate the benchmark ratio of the site's nonlinear to linear response. The dynamic Fourier spectrum ratio of the measured surface ground motion acceleration time history and the predicted downhole ground motion acceleration time history under strong and weak earthquake conditions. and Calculate the dynamic ratio of the site's nonlinear to linear response. The frequency domain physical constraint loss term is obtained by comparing the benchmark ratio during training. The dynamic ratio The differences are determined and quantified using mean square error.

9. A downhole seismic motion time history inversion system based on closed-loop reconstruction constraints, characterized in that, The system includes: The data acquisition module acquires the measured time histories of surface ground motion acceleration, the measured time histories of downhole ground motion acceleration, and the measured site parameters. The preprocessing module preprocesses the measured surface ground motion acceleration time history, the measured downhole ground motion acceleration time history, and the measured site parameters to obtain a dataset. The downhole data prediction module inputs the dataset into the inversion branch of the time-series neural network model to obtain the predicted downhole ground motion acceleration time history; The surface data reconstruction module inputs the predicted downhole ground motion acceleration time history into the forward reconstruction branch of the time-series neural network model to obtain the reconstructed surface ground motion acceleration time history. The loss function construction unit constructs a joint loss function based on the predicted downhole ground motion acceleration time history, the reconstructed surface ground motion acceleration time history, and the Fourier spectrum ratio of the surface and downhole ground motion acceleration time histories. The joint prediction model construction unit uses the joint loss function to train the temporal neural network model to obtain the joint prediction model; The result output unit inputs the ground motion acceleration time history of the target site and the site parameters of the target site into the joint prediction model to obtain the predicted downhole ground motion acceleration time history results.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is run by the processor, it executes the steps of the downhole ground motion time history inversion method based on closed-loop reconstruction constraints as described in any one of claims 1 to 8.

Images