Method and system for determining seismic energy and ground motion amplification factor under complex topographic geological conditions based on energy method and artificial intelligence

Abstract

The application discloses a method and system for determining seismic energy and ground motion amplification coefficient under complex terrain geological conditions based on energy method and artificial intelligence, and belongs to the technical field of earthquake engineering and geotechnical engineering. The method comprises the following steps: determining the energy of a seismic source; calculating the energy attenuation on the seismic wave propagation path; calculating the energy transformation of different geological interfaces; extracting terrain feature parameters and constructing a training database to train a deep neural network to determine the energy distribution coefficient of a complex combined terrain; predicting the energy distribution coefficient of each point of a target site; combining the incident energy of bedrock to calculate the ground motion amplification coefficient and perform energy conservation correction; and finally outputting an amplification coefficient distribution map. The application combines physical mechanisms with artificial intelligence, can efficiently and accurately predict the ground motion amplification effect of complex combined terrains (such as mountain groups and valleys), and the result conforms to energy conservation, so that the application can provide a scientific basis for seismic zoning and engineering seismic design.

Description

Technical Field

[0001] This invention relates to the fields of earthquake engineering and geotechnical engineering technology, and more specifically, to a method and system for determining earthquake energy and seismic amplification factor under complex terrain and geological conditions based on energy methods and artificial intelligence. Background Technology

[0002] Seismic amplification refers to the phenomenon where seismic waves, upon reaching the Earth's surface, exhibit significantly enhanced amplitude, velocity, and acceleration due to the influence of local topography and geological conditions. This is one of the main causes of severe earthquake damage. Accurately determining the seismic amplification factor is crucial for the seismic design of major engineering projects, seismic microzonation, and rapid earthquake damage assessment.

[0003] Currently, methods for determining the amplification factor of topographic ground motion mainly include empirical methods, numerical simulation methods, and traditional energy methods. First, empirical methods are typically based on simple one-dimensional soil layer models or single mountain models, providing overly general amplification factors that fail to reflect the true seismic response of complex topographic combinations (such as the coupling between multiple peaks or the combination of mountains and sedimentary basins), thus limiting their applicability and accuracy. Second, numerical simulation methods, such as the finite element method or spectral element method, while capable of accurately simulating the propagation process of seismic waves in complex media, are computationally extremely costly, computationally difficult, and time-consuming for large-scale models containing high-frequency components and complex topographic and geological conditions, making them unsuitable for large-scale rapid assessments. Finally, traditional energy methods typically only consider the geometric diffusion and absorption attenuation of seismic waves in a single medium, lacking a quantitative description of the multiple reflections, refractions, diffractions, and energy redistribution mechanisms of seismic waves on complex terrain surfaces (especially combined terrains), and cannot explain the energy convergence and divergence phenomena caused by topography from a physical mechanism perspective.

[0004] Therefore, there is an urgent need for a method to determine the seismic motion amplification factor that can fully consider the physical mechanisms of seismic wave propagation and energy distribution (such as energy conservation and transformation) and efficiently handle the nonlinear and high-dimensional mapping relationships brought about by complex terrain, so as to make up for the shortcomings of existing technologies. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and system for determining earthquake energy and ground motion amplification factor under complex terrain and geological conditions based on energy method and artificial intelligence. This method achieves efficient and accurate prediction of ground motion amplification factor by combining physical mechanism and data-driven approach.

[0006] To solve the above-mentioned technical problems, the technical solution proposed in this application is as follows:

[0007] This invention provides a method for determining seismic energy and seismic amplification factor under complex terrain and geological conditions based on energy methods and artificial intelligence, comprising the following steps: Step 1: Determine the source energy. Based on the seismic hazard analysis results of the target area, determine the source parameters of the set earthquake, and calculate the energy radiation spectrum of the source at different frequencies based on the source parameters. Step 2: Calculation of energy attenuation along the propagation path of seismic waves. Based on the energy radiation spectrum, calculate the energy attenuation caused by geometric diffusion, medium absorption and transmission loss at different rock layer interfaces during the propagation of seismic waves from the source to the engineering bedrock surface at the bottom of the target site, and obtain the incident energy field reaching the engineering bedrock surface. Step 3: Calculation of energy transformation at different geological interfaces. Based on the detailed geological model of the target site, the transfer matrix method, equivalent linearization method or pre-trained interface energy distribution prediction model are used to calculate the energy transmission and distribution of seismic waves as they travel from the engineering bedrock surface upward through the near-surface geological structure to the nearest bedrock surface below the surface, thus obtaining the incident energy reaching the nearest bedrock surface below the surface. Step 4: Determine the energy distribution coefficient of complex terrain combinations, construct and train an artificial intelligence model for predicting terrain energy distribution coefficients. The artificial intelligence model takes parameters characterizing terrain features as input and the energy distribution coefficients at corresponding frequencies as output. Step 5: Predict the energy distribution coefficient of the target site. Perform a three-dimensional terrain model of the target site, extract the terrain feature parameters of all grid points, and input the terrain feature parameters into the trained artificial intelligence model. The model quickly outputs the energy distribution coefficient of each grid point at different frequencies. Step 6: Determine the ground motion amplification factor. Based on the relationship between energy and amplitude, determine the ground motion amplification factor of each grid point according to the ratio of the energy distribution factor of the target site grid point to the energy distribution factor of the flat reference point. Step 7: Results output and application. The calculated magnification factor distribution map is overlaid with the site geological map and topographic map to form a seismic motion parameter microzonation map for seismic fortification.

[0008] Furthermore, in the first step, the parameters of the earthquake are set, including magnitude, focal depth, fault type, and fault geometry, which are determined comprehensively through regional seismic tectonic background, historical seismic activity patterns, and seismic zoning map data. Based on the focal parameters, a focal model is used to describe the earthquake rupture process, and the total energy radiated from the focal source is calculated. The proportion of energy carried by different types of seismic waves is distinguished, and finally, the energy radiation spectrum of the focal source at different frequencies is output.

[0009] Furthermore, the second step specifically includes: calculating the energy attenuation caused by geometric diffusion based on the actual distance from the seismic source to the target site; calculating the energy loss of the seismic wave as it passes through different rock layers based on the medium quality factor and wave velocity data provided by the regional crustal structure model; calculating the energy transmission coefficient of the seismic wave as it passes through different rock layer interfaces based on the wave impedance differences of each medium layer to obtain the transmitted energy; and combining the geometric diffusion, medium absorption, and interface transmission loss to obtain the incident energy field reaching the engineering bedrock surface at the bottom of the target site.

[0010] Furthermore, the third step specifically includes: establishing a detailed geological model of the target site, including the depth of the top surface of the bedrock, the thickness of each overlying soil layer, wave velocity, and density information; when seismic waves are incident from the engineering bedrock surface to the bottom surface of the overburden layer, calculating the energy transmitted into the overburden layer based on the difference in wave impedance between the bedrock and the overburden layer and the angle of the incident wave; for multiple overburden layers, using the equivalent linearization method or the transfer matrix method to calculate the effective energy of the seismic waves after multiple reflections and transmissions between the interfaces of each layer, finally transmitted to the nearest bedrock surface below the ground surface; if there are multiple different geological zones within the site, calculating the transmission energy of each zone separately.

[0011] Furthermore, the fourth step specifically includes the following sub-steps: 4.1 Terrain Feature Parameterization: Obtain a high-precision digital elevation model of the target site, perform grid subdivision on the digital elevation model data, extract a series of parameters that can characterize the terrain features and their influence on seismic wave scattering for each grid point, and normalize all parameters. 4.2 Training Database Construction: Construct a database containing a large number of samples, each sample containing input features and output labels. The input features are the terrain feature parameters, and the output labels are the energy allocation coefficients at the corresponding frequencies. Divide the samples into training sets, validation sets, and test sets. 4.3 Neural Network Model Construction and Training: A deep neural network is designed, whose structure includes an input layer, multiple hidden layers, and an output layer. The number of neurons in the input layer corresponds to the number of terrain feature parameters, and the number of neurons in the output layer corresponds to the energy allocation coefficient at a specific frequency. The neural network is trained using the training set and validation set, and the model accuracy is evaluated on the test set. 4.4 Model Validation and Optimization: For terrain types with large prediction errors, the sources of error are analyzed, and optimization is carried out by increasing the training samples of this type of terrain, adjusting the network structure, or introducing more relevant feature parameters to obtain the final artificial intelligence model.

[0012] Furthermore, the terrain feature parameters extracted in step 4.1 include: basic geometric parameters, local undulation parameters, combined terrain parameters, and statistical parameters; the basic geometric parameters include elevation, slope, aspect, plane curvature, and profile curvature; the local undulation parameters include local elevation difference, surface roughness, and terrain relief; the combined terrain parameters include the distance to the nearest mountain peak, the distance to the nearest valley line, the relative azimuth of adjacent peaks, ridgeline density, and local terrain openness; the statistical parameters include the mean, variance, and skewness of elevation within a certain surrounding area.

[0013] Furthermore, in step 4.2, the sample sources for the training database include samples generated by high-precision numerical simulation and strong earthquake observation data samples. The high-precision numerical simulation involves selecting typical complex terrain combinations, using the spectral element method or finite element method to simulate seismic wave propagation, inputting plane waves of different frequencies, calculating the velocity time history of each surface node, and then obtaining the surface kinetic energy. The surface kinetic energy is then compared with the incident kinetic energy of the bedrock to obtain the true energy distribution coefficient. The strong earthquake observation data involves collecting records from strong earthquake arrays deployed on complex terrain during actual earthquakes, calculating the energy ratio between surface points and bedrock reference points, and using this as the true energy distribution coefficient.

[0014] Furthermore, the sixth step specifically includes: Select a flat reference point: Choose an area near the target site that is open and has no obvious topographic undulations as a flat reference point. For any grid point within the target site, its surface kinetic energy is obtained by multiplying the incident energy of the bedrock below that point by the terrain energy distribution coefficient corresponding to that point; the surface kinetic energy of a flat reference point is obtained by multiplying the incident energy of the bedrock below that reference point by the energy distribution coefficient of that reference point. When the difference in bedrock incident energy at each point is within a preset range, it is approximately considered to be equal to that at the reference point, and the amplification factor is the square root of the ratio of the energy distribution coefficient of the complex terrain point to the energy distribution coefficient of the reference point. Energy conservation corrections were applied to the initially obtained amplification coefficients to ensure that the total surface kinetic energy of the entire site matched the total incident energy of the bedrock, resulting in the corrected amplification coefficient distribution.

[0015] Furthermore, the interface energy distribution prediction model in the third step is constructed through the following steps: 3.1 Constructing a training sample set: For different types of soil-rock interfaces, the wave impedance, incident wave angle, incident wave frequency, and corresponding energy transmission coefficient and reflection coefficient of the media on both sides of the interface are obtained through high-precision numerical simulation or field measurement to form a sample database. 3.2 Design of Neural Network Model: Construct a deep neural network. The input layer corresponds to parameters such as wave impedance, incident angle, and frequency on both sides of the interface, and the output layer corresponds to the energy transmission coefficient and reflection coefficient. 3.3 Training and Validation: The neural network is trained using the training sample set, and the model parameters are adjusted using the validation set until the prediction accuracy meets the requirements; 3.4 The trained model is used to predict the energy distribution coefficient of each geological interface in the actual site.

[0016] On the other hand, this application also claims protection for a system for implementing the aforementioned method for determining seismic energy and seismic amplification factor under complex terrain and geological conditions based on energy methods and artificial intelligence, comprising: The terrain data processing module is used to acquire and process digital elevation model data of the target site and extract terrain feature parameters; The artificial intelligence prediction module has a built-in pre-trained artificial intelligence model for predicting terrain energy distribution coefficients. The artificial intelligence model takes the terrain feature parameters as input and outputs the energy distribution coefficients of each grid point at different frequencies. The amplification factor calculation module is used to determine the surface ground motion amplification factor based on the energy distribution factor, combined with the calculation results of source energy, propagation path attenuation and geological interface transformation. The mapping module is used to overlay the magnification factor distribution map with the site geological map and topographic map to generate a seismic motion parameter microzonation map.

[0017] Furthermore, this application also claims protection for an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the method for determining seismic energy and seismic amplification factor under complex terrain and geological conditions based on energy method and artificial intelligence.

[0018] Compared with the prior art, the present invention has achieved the following beneficial technical effects: This invention proposes a novel method for determining seismic motion amplification factors by combining the complete energy transfer chain of seismic waves from the seismic source to the Earth's surface (including source radiation, path attenuation, and geological interface transformation) with an artificial intelligence prediction model. This method retains the advantages of the energy method, such as its clear physical mechanism and compliance with the law of energy conservation, while also efficiently handling the highly nonlinear scattering and focusing effects caused by complex terrain combinations (such as mountain ranges, valleys, and basins) through an AI model. Compared to traditional empirical methods, this method can more accurately characterize the spatial amplification effect of complex terrain; compared to pure numerical simulation methods, this method improves computational efficiency by several orders of magnitude while maintaining accuracy, meeting the engineering application needs of large-scale seismic microzonation. The final output amplification factor distribution map, corrected for energy conservation, provides a more scientific and reliable decision-making basis for the site selection of major projects, seismic assessment of existing buildings, and seismic risk assessment. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 The flowchart illustrates a method for determining seismic energy and ground motion amplification factor under complex terrain and geological conditions based on energy methods and artificial intelligence, as provided in this embodiment of the invention.

[0021] Figure 2 This is a framework diagram of a system for determining seismic energy and ground motion amplification factor under complex terrain and geological conditions based on energy methods and artificial intelligence, provided for embodiments of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] like Figure 1 As shown in the figure, the method for determining seismic energy and seismic amplification factor under complex terrain and geological conditions based on energy method and artificial intelligence provided by the present invention includes the following seven steps: Step 1: Determine the source energy. Based on the seismic hazard analysis results of the target area, determine the source parameters of the set earthquake, and calculate the energy radiation spectrum of the source at different frequencies based on the source parameters. In one embodiment of this application, the method first determines the source energy. The calculation of the seismic amplification factor begins at the source of the seismic energy. Specifically, based on the seismic hazard analysis results of the target area, one or a group of representative design earthquakes need to be set. The parameters of the design earthquakes include magnitude (e.g., moment magnitude). M w The hypocenter parameters include focal depth, fault type (e.g., strike-slip, thrust, normal fault), and fault geometry (strike, dip, slip angle). These parameters are determined comprehensively using data such as regional seismic tectonic background, historical seismic activity patterns, and seismic zoning maps to ensure that the hypothesized earthquake represents a potential future seismic event in the region that could threaten the engineering site. Based on the determined focal parameters, commonly used seismological focal models (e.g., the Brune model or ω-type model) are employed. 2 A model is used to describe the earthquake rupture process and calculate the total energy radiated from the seismic source. This step requires distinguishing the energy proportions carried by different types of seismic waves (P-waves, S-waves, and surface waves), as S-waves typically carry the most energy and have the greatest impact on surface engineering structures, making them the focus of subsequent analysis. The final output is the energy radiation spectrum of the seismic source at different frequencies. E source ( f ), which serves as the input for the entire energy transfer chain.

[0024] Step 2: Calculation of energy attenuation along the propagation path of seismic waves. Based on the energy radiation spectrum, calculate the energy attenuation caused by geometric diffusion, medium absorption and transmission loss at different rock layer interfaces during the propagation of seismic waves from the source to the engineering bedrock surface at the bottom of the target site, and obtain the incident energy field reaching the engineering bedrock surface. Furthermore, in one embodiment of this application, the method further includes calculating energy attenuation along the propagation path of the seismic wave. As the seismic wave propagates from the source to the bottom of the target site (engineering bedrock surface), its energy attenuates due to various physical mechanisms. This step requires detailed consideration of geometric diffusion and medium absorption along the propagation path. Geometric diffusion occurs because the seismic wave front expands with increasing propagation distance, resulting in a gradual decrease in energy distributed per unit area. For body waves, the energy density is inversely proportional to the square of the propagation distance. Based on the actual distance R from the source to the target site, the energy attenuation coefficient G(R) caused by geometric diffusion can be calculated. For example, for S-waves in a homogeneous medium, the geometric diffusion factor is typically taken as 1 / R. Medium absorption attenuation occurs because the crustal medium is not perfectly elastic; during propagation, some of the seismic wave energy is converted into heat energy dissipation due to friction within the medium. This attenuation is related to the frequency of the seismic wave.f Closely related: high-frequency components attenuate faster, while low-frequency components propagate further. This is based on the medium quality factor Q provided by the regional crustal structure model. f Based on wave velocity data, the energy loss of seismic waves as they pass through different rock layers can be calculated layer by layer. The attenuation is usually expressed as exp( πfR / ( Qv (), where v is the wave velocity. Furthermore, seismic waves undergo reflection and transmission when passing through the interfaces of different rock layers. Each interface causes some energy to be reflected back to the lower layer, with only the transmitted portion continuing to propagate upwards. The energy transmission coefficient T needs to be calculated layer by layer based on the differences in wave impedance (the product of density ρ and wave velocity v) of each medium. interface For perpendicular incidence, the transmittance is 4. ρ 1 v 1 ρ 2 v 2 / ( ρ 1 v 1+ ρ 2 v 2) 2 Taking into account the geometric diffusion, medium absorption, and interfacial transmission losses mentioned above, the incident energy field reaching the bottom of the target site (engineering bedrock surface) is finally obtained. E base ( f , x , y )= E source ( f ) G ( R ) exp( πfR / ( Qv )) ∏ T Since the site is usually within a certain range, there may be slight spatial differences in the incident energy at the bottom (such as the source directionality effect), and these differences need to be retained as input for subsequent calculations.

[0025] Step 3: Calculation of energy transformation at different geological interfaces. Based on the detailed geological model of the target site, the transfer matrix method, equivalent linearization method or pre-trained interface energy distribution prediction model are used to calculate the energy transmission and distribution of seismic waves as they travel from the engineering bedrock surface upward through the near-surface geological structure to the nearest bedrock surface below the surface, thus obtaining the incident energy reaching the nearest bedrock surface below the surface. In one embodiment of this application, the method further includes energy transformation calculations at different geological interfaces. After seismic waves reach the bottom of the target site (engineering bedrock surface), they need to continue upwards through near-surface geological structures (such as weathered layers and sedimentary layers) to reach the surface. This stage involves energy distribution across multiple geological interfaces. First, a detailed geological model of the target site needs to be established, including the depth of the top bedrock surface and the thickness of each overlying soil layer. h i Wave speed v i ,density ρ i Information such as... When seismic waves incident from the bedrock to the bottom of the overburden, energy distribution occurs due to the difference in wave impedance between the bedrock and the overburden: some energy is reflected back to the bedrock, and some is transmitted into the overburden. Transmitted energy... T The proportion of the site depends on the difference in physical properties between the media on both sides of the interface and the angle of the incident wave. For oblique incidence, the more complex Zoeppritz equations are required to calculate the transmission and reflection coefficients. If the overburden is multi-layered, seismic waves will undergo multiple reflections and transmissions between the interfaces, forming complex energy propagation paths. The effective energy ultimately transmitted to the bottom of the surface can usually be calculated using the equivalent linearization method (considering the nonlinear characteristics of the soil) or the transfer matrix method (Thomson-Haskell method). E near-surface ( f , x , y The output of this step is the incident energy reaching the nearest bedrock surface (or the reference surface of engineering interest) below the Earth's surface. E in ( f , x , y )= E base ( f , x , y ) T The energy from the site will be directly used as input for subsequent topographic effect analysis. Additionally, if the site contains multiple different geological zones (e.g., overburden layers of varying thicknesses), the transmission energy for each zone needs to be calculated separately. T site, to obtain spatial changes E in ( f , x , y ).

[0026] In a preferred embodiment of the present invention, in the third step, a pre-trained interface energy distribution prediction model can be introduced to quickly and accurately predict the energy transmission coefficient and reflection coefficient of each geological interface, thereby further improving the computational efficiency, especially suitable for large-scale sites or multi-scheme comparison scenarios that require rapid evaluation.

[0027] The construction process of this interface energy distribution prediction model is as follows: First, a training sample set is constructed. For common soil-rock interface types in engineering, including bedrock-weathered layer interfaces, weathered layer-soil interfaces, and interfaces between different soil layers, high-precision numerical simulation methods or field measurement data are used to obtain a large number of interface parameters and their corresponding energy distribution coefficients under different working conditions. Each sample contains input features and output labels. Input features include the wave impedance, incident wave angle, and incident wave frequency of the media on both sides of the interface. Output labels are the energy transmission coefficient and energy reflection coefficient. Wave impedance is defined as the product of the medium density and the wave velocity, the incident wave angle is the angle between the seismic wave propagation direction and the interface normal, and the frequency is the vibration frequency of the seismic wave. For the case of oblique incidence, if wave conversion needs to be considered, the output labels can be further refined into P-wave transmission coefficient, P-wave reflection coefficient, S-wave transmission coefficient, and S-wave reflection coefficient.

[0028] Secondly, a deep neural network model is designed. The number of neurons in the input layer corresponds to the number of input features, such as impedance ratio, incident angle, and frequency. The hidden layers consist of 3 to 5 fully connected layers, with the number of neurons in each layer set according to the sample size. The activation function is ReLU. The number of neurons in the output layer corresponds to the number of energy coefficients to be predicted, and the activation function is Sigmoid to ensure that the output value is between 0 and 1.

[0029] Then, the neural network is trained using the constructed sample set. The samples are divided into training, validation, and test sets. Mean squared error is used as the loss function, and the Adam optimizer updates the parameters. Overfitting is prevented by monitoring the validation set. After training, the model accuracy is evaluated on the test set to ensure that the mean relative error meets a preset threshold.

[0030] For multi-layered overburden structures, a trained interface energy distribution prediction model is applied to each interface. The calculation begins at the bottom interface, multiplying the incident wave energy by the energy transmitted from the previous interface, and then using the prediction model for the current interface to obtain the energy transmitted to the next layer. This process is repeated layer by layer until the effective energy transmitted to the nearest bedrock surface below the Earth's surface is obtained. When the incident wave angle is small, wave conversion can be ignored to simplify the calculation; when wave type conversion needs to be considered, a multi-output neural network is used to simultaneously predict the energy distribution coefficients of various wave types.

[0031] In practical applications, those skilled in the art can flexibly choose the traditional transfer matrix method, the equivalent linearization method, or the aforementioned artificial intelligence prediction model for calculation, depending on the complexity and accuracy requirements of the target site. Regardless of the method used, the output of the third step is the incident energy reaching the nearest bedrock surface below the Earth's surface, which will serve as the input for subsequent topographic effect analysis.

[0032] Step 4: Determine the energy distribution coefficient of complex terrain combinations, construct and train an artificial intelligence model for predicting terrain energy distribution coefficients. The artificial intelligence model takes parameters characterizing terrain features as input and the energy distribution coefficients at corresponding frequencies as output. A key embodiment of this application relates to the determination of energy distribution coefficients in complex terrain combinations. After seismic wave energy reaches the vicinity of the Earth's surface, complex terrain (especially combinations of multiple mountains) undergoes a complex redistribution of energy, manifested as the convergence or dissipation of kinetic energy. This is the core innovation of this invention: using artificial intelligence methods to predict terrain energy distribution coefficients. κ ( f , x , y ), defined as the sum of the actual kinetic energy at a point on the Earth's surface and the incident kinetic energy of the bedrock below that point (i.e. E in ( f , x , y The ratio of )). The method specifically includes the following sub-steps: 4.1 Terrain Feature Parameterization: First, a high-precision digital elevation model (DEM) of the target site is acquired, typically requiring a resolution of 5–10 meters. The DEM data is then gridded, with each grid point representing a surface location. For each grid point, a series of parameters characterizing the terrain features and their impact on seismic wave scattering are extracted, including basic geometric parameters (such as elevation Z, slope S, aspect A, plane curvature, and profile curvature) and local undulation parameters (such as local elevation difference Δ). H local, surface roughness σ Z , terrain relief Δ H relief), combined with terrain parameters (such as distance to the nearest mountain peak). D peak Distance to the nearest valley line D valley Relative azimuth of adjacent mountain peaks θ ridge Ridge line density ρ ridge Local terrain openness O ) and statistical parameters (such as the mean elevation within a certain surrounding area) μ Z ,variance σZ 2 skewness S k All parameters are normalized to ensure a consistent value range (e.g., mapped to [0, 1] or [...]). [1, 1]), which facilitates neural network training.

[0033] 4.2 Training Database Construction: To train an AI model capable of accurately predicting terrain energy distribution coefficients, a database containing a large number of samples needs to be constructed. The samples are primarily sourced in two ways: one is through high-precision numerical simulation, which involves selecting a large number of typical and complex terrain combinations (such as single-peak, double-peak, multi-peak, ridges, valleys, saddles, river valleys, basins, etc.) and using numerical methods such as the spectral element method (SEM) or the finite element method (FEM) to simulate seismic wave propagation. Different frequencies are input for each terrain type. f For plane waves (covering the frequency range of interest in the project, such as 0.1-20Hz), calculate the velocity time history of each surface node. v ( t ), thereby obtaining surface kinetic energy. E surface ∝∫ v ( t ) 2 dt Combining surface kinetic energy with incident bedrock kinetic energy E in By comparison, the true energy distribution coefficient is obtained. κ true ( f , x , y )= E surface / E in Each terrain type can generate tens of thousands of sample points, forming a numerical simulation sample set. Secondly, strong earthquake observation data is collected, specifically records from strong earthquake arrays deployed in complex terrains both domestically and internationally during actual earthquakes, particularly arrays with simultaneous bedrock reference point records. The energy ratio between surface points and bedrock points is calculated as the true energy distribution coefficient. κ true Although limited in quantity, this type of data can reflect the complex incident conditions of real seismic waves and can serve as an important supplement and validation for numerical simulation samples. Each sample contains two parts: input features X (the aforementioned terrain parameters) and output labels y (corresponding frequencies). f Energy distribution coefficient under κ true The samples are divided into training, validation, and test sets for model training and evaluation.

[0034] 4.3 Neural Network Model Construction and Training: Design a deep neural network (DNN) with an input layer, multiple hidden layers, and an output layer. The number of neurons in the input layer corresponds to the number of terrain feature parameters. N input Hidden layers typically consist of 3 to 6 layers, with the number of neurons in each layer determined by the sample size (e.g., 64 to 256). ReLU (Rectified Linear Unit) activation is used to accelerate convergence and prevent gradient vanishing. The output layer consists of a single neuron, corresponding to the energy distribution coefficient κ at a specific frequency f. pred Linear activation is employed. During training, the output κ is predicted using the input features X of the training set samples. pred and with real labels κ true Compare and calculate the loss function (e.g., mean squared error, MSE). Use the Adam optimizer to update parameters while monitoring the validation set loss. Stop training early when the validation set loss stops decreasing to prevent overfitting. After training, evaluate the model accuracy on the test set to ensure the mean relative error meets requirements (e.g., less than 10%). Pay special attention to the prediction performance on terrains with multiple mountainous formations; supplement training samples as needed.

[0035] 4.4 Model Validation and Optimization: For terrain types with large prediction errors, the sources of error are analyzed. Optimization can be achieved by increasing the training samples for this type of terrain, adjusting the network structure (such as increasing the number of layers or neurons), or introducing more relevant feature parameters. Ultimately, an AI model with strong generalization ability and high prediction accuracy is obtained, denoted as [model name missing]. κ pred = F AI ( X | f ).

[0036] Step 5: Predict the energy distribution coefficient of the target site. Perform a three-dimensional terrain model of the target site, extract the terrain feature parameters of all grid points, and input the terrain feature parameters into the trained artificial intelligence model. The model quickly outputs the energy distribution coefficient of each grid point at different frequencies. In one embodiment of this application, the terrain feature parameters are further specified. The extracted parameters include basic geometric parameters (such as elevation, slope, aspect, plane curvature, and profile curvature), local undulation parameters (such as local elevation difference, surface roughness, and terrain relief), combined terrain parameters (such as distance to the nearest mountain peak, distance to the nearest valley line, relative azimuth of adjacent peaks, ridgeline density, and local terrain openness), and statistical parameters (such as the mean, variance, and skewness of elevation within a certain surrounding area). These parameters comprehensively characterize the influence of terrain on seismic wave scattering from different scales and perspectives, providing rich input information for the AI ​​model.

[0037] In one embodiment of this application, the construction of the training database is further specified. The sample sources for the training database include samples generated through high-precision numerical simulations such as the spectral element method or the finite element method, as well as real earthquake observation data obtained from strong-motion array records. Numerical simulations can generate a large number of samples under various theoretical topographic conditions, covering a comprehensive parameter space; while strong-motion observation data provides valuable samples under real seismic inputs, used to verify and supplement the deficiencies of numerical simulations, ensuring that the trained model has good generalization ability and physical realism.

[0038] Subsequently, in one embodiment of this application, the energy distribution coefficient of the target site is predicted. For the actual complex terrain site to be evaluated, a three-dimensional terrain model of the target site is first performed. Based on its DEM data, the terrain feature parameters of all grid points (x, y) are extracted according to the aforementioned method. X ( x , y (Keep this consistent with training). Then, the feature parameters for each grid point... X ( x , y Input the trained AI model F AI The model quickly outputs the values ​​of each grid point at different frequencies. f Energy distribution coefficient under κ pred (f, x, y) = F AI (X(x, y) | f) Since the number of site grid points can be very large (e.g., tens of thousands to hundreds of thousands), batch input can be used, and with the help of GPU acceleration, the entire prediction process usually only takes a few minutes. The prediction result is a three-dimensional array (spatial coordinates × frequency), reflecting the local enhancement (κ>1) or attenuation (κ<1) effect of complex terrain on seismic wave energy, including the focusing effect of single-point terrain and the coupling effect between multiple mountains.

[0039] Step 6: Determine the ground motion amplification factor. Based on the relationship between energy and amplitude, determine the ground motion amplification factor of each grid point according to the ratio of the energy distribution factor of the target site grid point to the energy distribution factor of the flat reference point. Next, in one embodiment of this application, the amplification factor for ground motion is determined. Amplification factor A ( f , x , y Seismic energy is defined as the ratio of the seismic motion amplitude (such as peak ground velocity, PGV) at a point on the surface of complex terrain to the seismic motion amplitude at a reference point on flat terrain. This is based on the principle that seismic wave energy is proportional to the square of its amplitude (kinetic energy ∝ amplitude). 2 This ratio is equal to the square root of the ratio of the surface kinetic energies of the two points. First, a flat reference point needs to be selected. Choose an area near the target site with open terrain and no significant topographic relief as the flat reference point. The surface of this area should be exposed bedrock or have only a thin layer of overlying bedrock, and its topographic energy distribution coefficient should be [not specified]. κ ref ( f This can be achieved through numerical simulation or simplified theoretical calculations (such as...). κ ref ( f ≈1) Determine. If the reference point has a covering layer of a certain thickness, its transmission energy needs to be calculated according to the method in step three. T site，ref This ensures the reliability of the benchmark. For any grid point (x, y) within the target site, its surface kinetic energy... E surface ( f , x , y It consists of two parts: the incident energy of the bedrock below this point. E in ( f , x , y (From the calculation results of steps two and three) multiplied by the terrain energy distribution coefficient corresponding to that point κ pred ( f , x , y (Based on AI predictions in step five). That is... E surface ( f , x , y )= E in ( f , x , y ) κ pred (f , x , y Similarly, the surface kinetic energy at a flat reference point. E surface，ref ( f Energy incident from the bedrock below the reference point E in，ref ( f Multiply by the energy distribution coefficient at the reference point κ ref ( f This is obtained by considering that, due to the limited area of ​​the target site, the incident energy of the bedrock at each point is usually very small, and can be approximated as... E in ( f , x , y )≈ E in，ref ( f At this point, the magnification factor... A ( f , x , y It mainly depends on the relative magnitude of the terrain energy distribution coefficient, i.e. A ( f , x , y )= E surface ( f , x , y ) / E surface，ref ( f )≈ κ pred ( f , x , y ) / κ ref ( f If the reference point energy distribution coefficient κ ref ( f If ) = 1, then the magnification factor is 1. A ( f , x , y Directly taking the energy distribution coefficient of complex terrain points κ pred ( f , x , y The square root of the amplification factor is used. To ensure the physical rationality of the energy distribution across the entire site, an energy conservation correction is needed to be applied to the initially obtained amplification factor. The principle of the correction is to ensure the total surface kinetic energy of the entire site. Esurface ( f , x , y ) dxdy Total incident energy on bedrock (after deducting radiation damping losses) E in ( f , x , y ) dxdy This ensures a proper match and avoids energy non-conservation due to AI prediction errors or approximations. The corrected amplification factor distribution. A corrected ( f , x , y This is more consistent with actual physical processes.

[0040] Step 7: Results output and application. The calculated magnification factor distribution map is overlaid with the site geological map and topographic map to form a seismic ground motion parameter microzonation map for seismic design. Finally, in one embodiment of this application, the results are output and applied. The calculated magnification factor distribution map is overlaid with the site geological map and topographic map to form a seismic motion parameter microzonation map. The map clearly delineates different areas such as high magnification zones (e.g., peaks), medium-high magnification zones (e.g., saddles), medium magnification zones (e.g., foothills), and low magnification zones (e.g., valleys), and is marked with different colors. These results can be directly applied to site selection for major projects, seismic assessment and reinforcement of existing buildings, earthquake insurance and risk assessment, and post-earthquake emergency rescue.

[0041] See Figure 2 This application also provides a system for implementing the above-described method. In one embodiment, the system includes a terrain data processing module, an artificial intelligence prediction module, a magnification factor calculation module, and a mapping module. The terrain data processing module acquires and processes digital elevation model (DEM) data of the target site, extracts terrain feature parameters for each grid point according to the aforementioned method, and normalizes the parameters. The artificial intelligence prediction module is connected to the terrain data processing module and receives the terrain feature parameters. This module has a built-in artificial intelligence model trained through the steps described in the fourth step above, used to predict terrain energy distribution coefficients. This model takes the terrain feature parameters as input and outputs the energy distribution coefficients of each grid point at different frequencies. κ pred ( f , x , yThe amplification factor calculation module is connected to the artificial intelligence prediction module and receives the energy distribution coefficient. This module is used to calculate the source energy, propagation path attenuation, and geological interface transformation results (i.e., bedrock incident energy) based on the energy distribution coefficient, combined with the results obtained from the first to the third steps. E in ( f , x , y And, following the method described in step six above, determine the surface seismic amplification factor. A corrected ( f , x , y The mapping module is connected to the magnification factor calculation module, receives the magnification factor, and uses it to overlay the magnification factor distribution map with the site geological map and topographic map to generate a visualized seismic motion parameter microzonation map, and outputs it in the form of an image file or a Geographic Information System (GIS) layer.

[0042] This application also provides an electronic device. In one embodiment, the electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it can implement all the steps of the aforementioned method for determining seismic energy and seismic amplification factor under complex terrain and geological conditions based on energy methods and artificial intelligence. This electronic device can be a server, personal computer, or embedded system, or other device with computing capabilities, used to execute the method of this invention, providing efficient computational support for engineering applications.

[0043] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for determining seismic energy and seismic motion amplification factor under complex terrain and geological conditions based on energy methods and artificial intelligence, characterized in that, Includes the following steps: Step 1: Determine the source energy. Based on the seismic hazard analysis results of the target area, determine the source parameters of the set earthquake, and calculate the energy radiation spectrum of the source at different frequencies based on the source parameters. Step 2: Calculation of energy attenuation along the propagation path of seismic waves. Based on the energy radiation spectrum, calculate the energy attenuation caused by geometric diffusion, medium absorption and transmission loss at different rock layer interfaces during the propagation of seismic waves from the source to the engineering bedrock surface at the bottom of the target site, and obtain the incident energy field reaching the engineering bedrock surface. Step 3: Calculation of energy transformation at different geological interfaces. Based on the detailed geological model of the target site, the transfer matrix method, equivalent linearization method or pre-trained interface energy distribution prediction model are used to calculate the energy transmission and distribution of seismic waves as they travel from the engineering bedrock surface upward through the near-surface geological structure to the nearest bedrock surface below the surface, thus obtaining the incident energy reaching the nearest bedrock surface below the surface. Step 4: Determine the energy distribution coefficient of complex terrain combinations, construct and train an artificial intelligence model for predicting terrain energy distribution coefficients. The artificial intelligence model takes parameters characterizing terrain features as input and the energy distribution coefficients at corresponding frequencies as output. Step 5: Predict the energy distribution coefficient of the target site. Perform a three-dimensional terrain model of the target site, extract the terrain feature parameters of all grid points, and input the terrain feature parameters into the trained artificial intelligence model. The model quickly outputs the energy distribution coefficient of each grid point at different frequencies. Step 6: Determine the ground motion amplification factor. Based on the relationship between energy and amplitude, determine the ground motion amplification factor of each grid point according to the ratio of the energy distribution factor of the target site grid point to the energy distribution factor of the flat reference point. Step 7: Results output and application. The calculated magnification factor distribution map is overlaid with the site geological map and topographic map to form a seismic motion parameter microzonation map for seismic fortification.

2. The method for determining seismic energy and seismic amplification factor under complex terrain and geological conditions based on energy method and artificial intelligence as described in claim 1, characterized in that, In the first step, the parameters of the earthquake are set, including magnitude, focal depth, fault type, and fault geometry, which are determined by comprehensively considering regional seismic tectonic background, historical seismic activity patterns, and seismic zoning map data. Based on the focal parameters, a focal model is used to describe the earthquake rupture process, and the total energy radiated from the focal source is calculated. The proportion of energy carried by different types of seismic waves is distinguished, and finally, the energy radiation spectrum of the focal source at different frequencies is output.

3. The method for determining seismic energy and seismic amplification factor under complex terrain and geological conditions based on energy method and artificial intelligence as described in claim 1, characterized in that, The second step specifically includes: calculating the energy attenuation caused by geometric diffusion based on the actual distance from the seismic source to the target site; calculating the energy loss of the medium absorbed when the seismic wave passes through different rock layers based on the medium quality factor and wave velocity data provided by the regional crustal structure model; calculating the energy transmission coefficient of the seismic wave when passing through different rock layer interfaces based on the wave impedance differences of each medium layer to obtain the transmitted energy; and combining the geometric diffusion, medium absorption, and interface transmission loss to obtain the incident energy field reaching the engineering bedrock surface at the bottom of the target site.

4. The method for determining seismic energy and seismic amplification factor under complex terrain and geological conditions based on energy method and artificial intelligence as described in claim 1, characterized in that, The third step specifically includes: establishing a detailed geological model of the target site, including the depth of the top surface of the bedrock, the thickness of each overlying soil layer, wave velocity, and density information; when seismic waves are incident from the engineering bedrock surface to the bottom surface of the overburden, calculating the energy transmitted into the overburden based on the difference in wave impedance between the bedrock and the overburden and the angle of the incident wave; for multiple overburden layers, using the equivalent linearization method or the transfer matrix method to calculate the effective energy of the seismic waves after multiple reflections and transmissions between the interfaces of each layer, finally transmitted to the nearest bedrock surface below the ground surface; if there are multiple different geological zones within the site, calculating the transmission energy of each zone separately.

5. The method for determining seismic energy and seismic amplification factor under complex terrain and geological conditions based on energy method and artificial intelligence as described in claim 1, characterized in that, The fourth step specifically includes the following sub-steps: 4.1 Terrain Feature Parameterization: Obtain a high-precision digital elevation model of the target site, perform grid subdivision on the digital elevation model data, extract a series of parameters that can characterize the terrain features and their influence on seismic wave scattering for each grid point, and normalize all parameters. 4.2 Training Database Construction: Construct a database containing a large number of samples, each sample containing input features and output labels. The input features are the terrain feature parameters, and the output labels are the energy allocation coefficients at the corresponding frequencies. Divide the samples into training sets, validation sets, and test sets. 4.3 Neural Network Model Construction and Training: A deep neural network is designed, whose structure includes an input layer, multiple hidden layers, and an output layer. The number of neurons in the input layer corresponds to the number of terrain feature parameters, and the number of neurons in the output layer corresponds to the energy allocation coefficient at a specific frequency. The neural network is trained using the training set and validation set, and the model accuracy is evaluated on the test set. 4.4 Model Validation and Optimization: For terrain types with large prediction errors, the sources of error are analyzed, and optimization is carried out by increasing the training samples of this type of terrain, adjusting the network structure, or introducing more relevant feature parameters to obtain the final artificial intelligence model.

6. The method for determining seismic energy and seismic amplification factor under complex terrain and geological conditions based on energy method and artificial intelligence as described in claim 5, characterized in that, The terrain feature parameters extracted in step 4.1 include: basic geometric parameters, local undulation parameters, combined terrain parameters, and statistical parameters; the basic geometric parameters include elevation, slope, aspect, plane curvature, and profile curvature; the local undulation parameters include local elevation difference, surface roughness, and terrain relief; the combined terrain parameters include the distance to the nearest mountain peak, the distance to the nearest valley line, the relative azimuth of adjacent peaks, ridgeline density, and local terrain openness; the statistical parameters include the mean, variance, and skewness of elevation within a certain surrounding area.

7. The method for determining seismic energy and seismic amplification factor under complex terrain and geological conditions based on energy method and artificial intelligence as described in claim 5, characterized in that, In step 4.2, the sample sources for the training database include samples generated by high-precision numerical simulation and strong earthquake observation data samples. The high-precision numerical simulation involves selecting typical complex terrain combinations, using the spectral element method or finite element method to simulate seismic wave propagation, inputting plane waves of different frequencies, calculating the velocity time history of each surface node, and then obtaining the surface kinetic energy. The surface kinetic energy is compared with the incident kinetic energy of the bedrock to obtain the true energy distribution coefficient. The strong earthquake observation data involves collecting records from strong earthquake arrays deployed on complex terrain during actual earthquakes, calculating the energy ratio between surface points and bedrock reference points, and using this as the true energy distribution coefficient.

8. The method for determining seismic energy and seismic amplification factor under complex terrain and geological conditions based on energy method and artificial intelligence as described in claim 1, characterized in that, The sixth step specifically includes: Select a flat reference point: Choose an area near the target site that is open and has no obvious topographic undulations as a flat reference point. For any grid point within the target site, its surface kinetic energy is obtained by multiplying the incident energy of the bedrock below that point by the terrain energy distribution coefficient corresponding to that point; the surface kinetic energy of a flat reference point is obtained by multiplying the incident energy of the bedrock below that reference point by the energy distribution coefficient of that reference point. When the difference in bedrock incident energy at each point is within a preset range, it is approximately considered to be equal to that at the reference point, and the amplification factor is the square root of the ratio of the energy distribution coefficient of the complex terrain point to the energy distribution coefficient of the reference point. Energy conservation corrections were applied to the initially obtained amplification coefficients to ensure that the total surface kinetic energy of the entire site matched the total incident energy of the bedrock, resulting in the corrected amplification coefficient distribution.

9. The method for determining seismic energy and seismic amplification factor under complex terrain and geological conditions based on energy method and artificial intelligence as described in claim 1, characterized in that, The interface energy distribution prediction model in the third step is constructed through the following steps: 3.1 Constructing a training sample set: For different types of soil-rock interfaces, the wave impedance, incident wave angle, incident wave frequency, and corresponding energy transmission coefficient and reflection coefficient of the media on both sides of the interface are obtained through high-precision numerical simulation or field measurement to form a sample database. 3.2 Design of Neural Network Model: Construct a deep neural network. The input layer corresponds to parameters such as wave impedance, incident angle, and frequency on both sides of the interface, and the output layer corresponds to the energy transmission coefficient and reflection coefficient. 3.3 Training and Validation: The neural network is trained using the training sample set, and the model parameters are adjusted using the validation set until the prediction accuracy meets the requirements; 3.4 The trained model is used to predict the energy distribution coefficient of each geological interface in the actual site.

10. A system for implementing the method for determining seismic energy and seismic amplification factor under complex terrain and geological conditions based on energy methods and artificial intelligence as described in claim 1, characterized in that, include: The terrain data processing module is used to acquire and process digital elevation model data of the target site and extract terrain feature parameters; The artificial intelligence prediction module has a built-in pre-trained artificial intelligence model for predicting terrain energy distribution coefficients. The artificial intelligence model takes the terrain feature parameters as input and outputs the energy distribution coefficients of each grid point at different frequencies. The amplification factor calculation module is used to determine the surface ground motion amplification factor based on the energy distribution factor, combined with the calculation results of source energy, propagation path attenuation and geological interface transformation. The mapping module is used to overlay the magnification factor distribution map with the site geological map and topographic map to generate a seismic motion parameter microzonation map.

11. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method for determining seismic energy and seismic amplification factor under complex terrain and geological conditions based on energy method and artificial intelligence, as described in any one of claims 1 to 9.

Images