Earthquake demand prediction method based on vectorization and sequence-to-sequence proxy model

Abstract

An earthquake demand prediction method based on vectorization and a sequence-to-sequence proxy model belongs to the technical field of evaluation of structural damage caused by earthquakes, and comprises the following steps: (1) generating multi-dimensional input parameters including geometry, materials and earthquake boundary conditions based on Latin hypercube sampling; (2) generating a dynamic response database of the reinforced concrete frame structure by utilizing finite element random time-history analysis based on the multi-dimensional input parameters; and (3) adopting a deep learning algorithm, training a point model, a vectorization model and a sequence-to-sequence model based on the dynamic response database, and respectively applying the point model, the vectorization model and the sequence-to-sequence model to structure response prediction and multi-target reliability analysis. By integrating deep learning technologies (including a deep neural network DNN, a long-short term memory network LSTM and a Transform architecture), efficient prediction of multi-target dynamic response and multivariable earthquake vulnerability evaluation are realized, and the defects of an existing method in the aspects of calculation efficiency, model adaptability and multi-target prediction precision are overcome.

Description

Technical Field

[0001] The present invention belongs to the technical field of structural damage assessment caused by earthquakes, and in particular relates to an earthquake demand prediction method based on vectorization and a sequence-to-sequence proxy model. Background Art

[0002] With the acceleration of global urbanization, the continuous increase in building height and structural complexity has put forward higher requirements for the seismic performance of building structures, especially reinforced concrete frame structures. However, traditional seismic analysis methods, such as finite element analysis, can provide high-precision results, but in the face of multiple uncertainties (such as the randomness of earthquake input, the volatility of material and geometric parameters) and the reliability assessment of large-scale group structures, they show problems of high computational cost and long time consumption. This severely limits the practical application of seismic analysis based on the finite element method in disaster emergency response, rapid assessment and multi-scenario prediction.

[0003] In recent years, with the rapid development of machine learning and deep learning technologies, surrogate models, as a technical means to quickly approximate the response of complex nonlinear systems, have gradually attracted attention and application in the field of structural engineering. Surrogate models can significantly reduce computational costs and achieve real-time prediction of complex models while ensuring the accuracy of the results. At present, common surrogate model technologies include polynomial regression, support vector machine, Gaussian process regression, decision tree, and deep learning methods based on neural networks. Among them, surrogate models based on deep learning have a particularly broad application prospect in structural dynamic analysis due to their powerful modeling capabilities for high-dimensional complex nonlinear problems.

[0004] Existing studies have shown that the application of machine learning technology in seismic analysis has made some progress. For example, the hydrodynamic simulation of shallow water wave equations is realized through physically driven neural networks; and various regression methods (such as Gaussian process regression and support vector regression) are used to predict the cable force in structural optimization. These studies have demonstrated the potential of machine learning technology in seismic modeling, but most of them focus on single-objective or simple multi-objective tasks, such as frequency analysis or single-point seismic vulnerability assessment.

[0005] In addition, although the research on multi-objective agent models has begun to attract attention, the existing models still have certain limitations. Taking the adaptive vectorized agent modeling framework as an example, although this method achieves vectorized output, it still relies on multiple independent machine learning models, which increases the computational complexity and makes it difficult to expand to more complex application scenarios. More importantly, most of the existing agent models are limited to fixed geometry and material properties and cannot adapt to the diversity of reinforced concrete frame structures in actual engineering. This limitation makes it difficult for existing models to meet the wide range of needs of urban building groups in seismic analysis.

[0006] At the same time, most traditional proxy models adopt a single-point output modeling approach, ignoring the correlation between multiple target outputs. This approach leads to inaccuracies in multi-target reliability assessments, limiting its application in actual complex tasks. Especially in seismic response analysis, there is a significant coupling relationship between the inter-story displacement, vertex displacement, base shear, and acceleration of the structure. The single-point output model cannot effectively capture these intrinsic connections, resulting in incomplete prediction results. In addition, existing studies mostly focus on point models rather than time series models, which further limits the comprehensive modeling capabilities of seismic input characteristics.

[0007] In response to the above problems, the introduction of deep learning methods such as long short-term memory networks (LSTM) and Transformer in recent years has provided new solutions for surrogate models. LSTM networks, through their unique memory unit design, can effectively capture the dependencies of long time series data and are widely used in language modeling, time series prediction, and structural response analysis. The Transformer architecture achieves efficient parallel computing and modeling capabilities of long-distance dependencies through the self-attention mechanism, and is particularly suitable for sequence-to-sequence tasks with more complex input features. However, the application of LSTM and Transformer in structural engineering is still in the exploratory stage, especially in complex multi-objective and time series modeling, and its potential has not been fully explored.

[0008] In summary, the current research gaps include:

[0009] Adaptive modeling of diverse structural properties: Existing models have difficulty adapting to the diverse geometric, material, and boundary condition properties of reinforced concrete frame structures.

[0010] Effective capture of time series characteristics: The temporal characteristics of earthquake inputs are not fully utilized, resulting in the inadequacy of proxy models in dynamic response prediction.

[0011] Accurate modeling of multi-objective outputs: There is a lack of surrogate model frameworks that can output multiple objective responses simultaneously.

[0012] Computational efficiency and scalability: Existing methods have high computational complexity and are difficult to meet the needs of group structure analysis and rapid disaster response.

[0013] Therefore, developing an efficient proxy model framework that can adapt to diverse structural characteristics, capture the characteristics of earthquake input time series, and support multi-objective synchronous output has become the key to solving the problem of seismic analysis of reinforced concrete frame structures. Based on this research background, this paper proposes a vectorization and sequence-to-sequence proxy model framework, which provides a new solution for rapid seismic demand prediction and vulnerability analysis of reinforced concrete frame structures. Summary of the invention

[0014] The present invention discloses an earthquake demand forecasting method based on vectorization and sequence-to-sequence proxy model. By integrating deep learning technology (including deep neural network DNN, long short-term memory network LSTM and Transformer architecture), the present invention realizes efficient prediction of multi-objective dynamic response and multivariate earthquake vulnerability assessment, overcoming the shortcomings of existing methods in computational efficiency, model adaptability and multi-objective prediction accuracy.

[0015] To achieve the above purpose, the technical solution of the present invention is:

[0016] The earthquake demand prediction method based on vectorization and sequence-to-sequence proxy model includes the following steps:

[0017] (1) Generate multidimensional input parameters including geometry, material and seismic boundary conditions based on Latin hypercube sampling;

[0018] (2) Based on multi-dimensional input parameters, a dynamic response database of reinforced concrete frame structures is generated using finite element stochastic time history analysis;

[0019] (3) A deep learning algorithm is used to train point models, vectorized models, and sequence-to-sequence models based on a dynamic response database, which are used for structural response prediction and multi-objective reliability analysis, respectively.

[0020] Preferably, in step (1), the multidimensional input parameters include parameter one based on geometric properties, parameter two based on material properties, and parameter three based on seismic boundary condition properties.

[0021] Preferably, in the step (1), the parameter one includes the floor height, single span length, single span width, column width, beam depth, column cross-sectional area, beam cross-sectional area, concrete cover thickness, and floor thickness; the parameter two includes the concrete strength grade, steel bar yield strength, steel bar stiffness ratio before and after yield, elastic modulus, and Poisson's ratio; and the parameter three includes the earthquake magnitude, fault zone length, earthquake duration, peak ground acceleration, spectral acceleration, and cumulative absolute acceleration.

[0022] Preferably, in the step (3), the point model refers to: realizing rapid prediction of single-target output response through DNN, which is suitable for single-variable analysis tasks of structural dynamic response; the vectorized model refers to: designing a network model of vectorized output structure based on DNN, simple LSTM and Transformer, so as to realize synchronous prediction of multi-target responses; the sequence-to-sequence model refers to: constructing a network model using complex LSTM and Transformer architecture to capture the time series characteristics of earthquake input and generate the time history of structural dynamic response.

[0023] Preferably, in the step (3), the vectorized model is used to achieve multi-objective synchronous output, and the output of the multi-objective response includes the maximum top displacement, maximum inter-story displacement angle, maximum acceleration and maximum base shear force of the reinforced concrete frame structure.

[0024] Preferably, in the step (3), the sequence-to-sequence model captures the time series characteristics of the earthquake input through LSTM and Transformer networks.

[0025] Preferably, in the step (3), based on the constructed point model, vectorized model and sequence-to-sequence model, the Monte Carlo method is used to perform univariate and multivariate vulnerability analysis, and a relationship curve between the earthquake intensity index and the failure probability is output.

[0026] Preferably, in step (3), the earthquake intensity index includes spectral acceleration, peak ground acceleration and cumulative absolute acceleration.

[0027] The beneficial effects of the earthquake demand prediction method based on vectorization and sequence-to-sequence proxy model of the present invention are:

[0028] 1. The present invention realizes the synchronous prediction of multi-objective responses through a vectorized model framework, which significantly improves the computational efficiency and prediction accuracy of multi-objective reliability analysis.

[0029] 2. The sequence-to-sequence model of the present invention introduces LSTM and Transformer architecture, which enables the proxy model to capture the time series characteristics of earthquake input and adapt to complex dynamic response prediction tasks.

[0030] 3. The present invention screens out the most important earthquake characteristics for reinforced concrete frame response through sensitivity analysis of input characteristics, and optimizes the input parameter combination of the model.

[0031] 4. The present invention uses the Monte Carlo method combined with the proxy model to quickly generate single-variable and multi-variable earthquake vulnerability curves, which are suitable for disaster emergency management and reliability assessment.

[0032] 5. The present invention has a wide range of application scenarios, as follows:

[0033] A. Individual structure reliability analysis: Applicable to rapid earthquake demand prediction and reliability assessment of reinforced concrete frame structures, which can provide technical support for earthquake-resistant design and performance evaluation.

[0034] B. Group structural disaster assessment: The method of the present invention has good scalability and can be further applied to community-level structural group analysis to support urban disaster prevention and control planning and emergency response.

[0035] C. Engineering decision support: By quickly generating multi-objective earthquake response and vulnerability analysis results, the present invention provides data support for engineering decision-making, resource allocation and post-disaster recovery. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 The process of generating training data for the proxy model is shown;

[0037] Figure 2 This is a schematic diagram of the architecture of the point model;

[0038] Figure 3 This is a schematic diagram of the architecture of the vectorized model;

[0039] Figure 4 Schematic diagram of the LSTM architecture of the sequence-to-sequence model;

[0040] Figure 5 Schematic diagram of the Transformer architecture of the sequence-to-sequence model;

[0041] Figure 6 This is the correlation analysis diagram between earthquake input and structural response;

[0042] Figure 7 is the result of univariate seismic vulnerability analysis;

[0043] Figure 8 is the result of multivariate seismic vulnerability analysis;

[0044] Fig. 9 This is a comparison chart of the prediction accuracy of different proxy models;

[0045] Fig.10 This is a schematic diagram of the earthquake response time history;

[0046] Fig.11 Schematic diagram of the application of agent model in community structure group analysis. DETAILED DESCRIPTION

[0047] The following description is only a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

[0048] The following embodiments may be understood as individually expressing a part of a local structure or method of the present invention, or may be understood as a combination of the embodiments to explain the connotation of a larger structure or method of the present invention.

[0049] Example 1

[0050] The earthquake demand prediction method based on vectorization and sequence-to-sequence proxy model is as follows: Figure 1-11As shown, the following steps are included:

[0051] (1) Generate multidimensional input parameters including geometry, material and seismic boundary conditions based on Latin hypercube sampling; the input parameters cover a wide range of structural and seismic characteristics to ensure the versatility and adaptability of the model;

[0052] (2) Based on multi-dimensional input parameters, a dynamic response database of reinforced concrete frame structures is generated using finite element random time history analysis. A dynamic response database covering a variety of structural configurations and seismic input conditions is constructed by performing random time history analysis using the finite element method. This database is used to train a proxy model, enabling it to efficiently predict structural dynamic responses in complex nonlinear systems.

[0053] (3) A deep learning algorithm is used to train a point model, a vectorized model, and a sequence-to-sequence model (i.e., the proxy model includes the above three models) based on the dynamic response database, which are used for structural response prediction and multi-objective reliability analysis, respectively.

[0054] Example 2

[0055] like Figure 1-11 As shown, in the step (1), the multidimensional input parameters include parameter 1 based on geometric properties, parameter 2 based on material properties, and parameter 3 based on seismic boundary condition properties;

[0056] In the step (1), the first parameter includes the floor height, single span length, single span width, column width, beam depth, column cross-sectional area, beam cross-sectional area, concrete cover thickness, and floor thickness; the second parameter includes the concrete strength grade, steel yield strength, steel yield stiffness ratio before and after steel yield, elastic modulus, and Poisson's ratio; the third parameter includes earthquake magnitude, fault zone length, earthquake duration, peak ground acceleration, spectral acceleration, and cumulative absolute acceleration. The input parameters are normalized to eliminate dimensional differences and improve the stability and accuracy of model training.

[0057] Example 3

[0058] like Figure 1-11As shown, in the step (3), the point model refers to: realizing rapid prediction of single-objective output response through DNN, which is suitable for single-variable analysis tasks of structural dynamic response; the vectorized model refers to: designing a network model of vectorized output structure based on DNN, simple LSTM and Transformer, so as to realize synchronous prediction of multi-objective responses and improve the accuracy and efficiency of multi-objective reliability assessment; the sequence-to-sequence model refers to: constructing a network model using complex LSTM and Transformer architecture to capture the time series characteristics of earthquake input and generate the time history of structural dynamic response, which is suitable for complex dynamic response prediction and higher-dimensional data modeling tasks.

[0059] Example 4

[0060] like Figure 1-11 As shown: In the step (3), the vectorized model is used to achieve multi-objective synchronous output, and the output of the multi-objective response includes the maximum top displacement, maximum inter-story displacement angle, maximum acceleration and maximum base shear force of the reinforced concrete frame structure;

[0061] like Figure 1-11 As shown: In the step (3), the sequence-to-sequence model captures the time series characteristics of earthquake input through LSTM and Transformer networks;

[0062] like Figure 1-11 As shown: In the step (3), based on the constructed point model, vectorized model and sequence-to-sequence model, the Monte Carlo method is used to perform univariate and multivariate vulnerability analysis, and the relationship curve between the earthquake intensity index and the failure probability is output.

[0063] like Figure 1-11 As shown: In the step (3), the earthquake intensity indicators include spectral acceleration, peak ground acceleration and cumulative absolute acceleration.

[0064] Example 5

[0065] like Figure 1-11 As shown: This embodiment discloses the implementation methods of various proxy models, specifically:

[0066] (1) The implementation of the point model is as follows:

[0067] The point model is based on a deep neural network (DNN) and is used to predict single-target responses. The point model structure includes: input layer: including normalized geometry, material and seismic input parameters; hidden layer: setting multiple fully connected layers, each layer includes a certain number of neurons, and the activation function is ReLU; output layer: single target output (such as the maximum inter-layer displacement angle). The model weights are optimized by stochastic gradient descent, and the early stopping strategy is used to prevent overfitting.

[0068] (2) The implementation of the vectorized model is as follows:

[0069] The vectorized model is an extension of the point model to multi-objective output, realizing the simultaneous prediction of the maximum vertex displacement, the maximum inter-story displacement angle, the maximum base shear force and the maximum acceleration. The vectorized model includes: input layer: the same as the point model; output layer: a vector including multi-objective responses (including the maximum top displacement, the maximum inter-story displacement angle, the maximum base shear force and the maximum acceleration at the same time). This multi-objective output structure can capture the correlation between the objectives and significantly improve the efficiency of multi-objective reliability analysis.

[0070] (3) The implementation of the sequence-to-sequence model is as follows:

[0071] The sequence-to-sequence model uses the long short-term memory network LSTM and Transformer architecture to achieve modeling of earthquake input time series and time history prediction of dynamic response. Among them, the LSTM model: captures the time dependency of earthquake input through memory units; the Transformer model: processes earthquake input in parallel through the self-attention mechanism, which is suitable for long sequence modeling tasks. The output result is a time series of dynamic response (such as the top displacement time history).

[0072] Example 6

[0073] like Figure 1-11 As shown, this embodiment discloses that vulnerability analysis includes the following methods:

[0074] (1) Univariate vulnerability analysis: Combined with the point model or vectorized model, the univariate vulnerability curve is calculated: the geometric and material parameters are fixed, and a simulation with gradually increasing amplitude is performed for an earthquake input variable (such as PGA); the model is sampled multiple times using the Monte Carlo method to estimate the failure probability; and the relationship between the failure probability and the earthquake intensity is plotted.

[0075] (2) Multivariate vulnerability analysis: Combined with the vectorized model or sequence-to-sequence model, the joint effect of multiple seismic input variables is considered simultaneously: spectral acceleration, peak seismic acceleration, and cumulative absolute acceleration are used as joint input variables; ensure that there is a practical correlation between the input variables (for example, a positive correlation between peak ground acceleration and spectral acceleration); use the Monte Carlo method to generate a joint failure probability surface and draw a vulnerability curve.

[0076] Example 7

[0077] like Figure 1-11 As shown, this embodiment discloses an example of specific application of the present invention:

[0078] For a reinforced concrete frame structure, the input parameters are fixed geometric and material parameters and ground motion characteristics, and its vertex displacement under earthquake action is predicted. The results show that the point model can quickly provide high-precision single-objective predictions.

[0079] For the same reinforced concrete frame structure, the same parameters are input and the synchronous prediction values ​​of vertex displacement, inter-story displacement angle, base shear force and acceleration are output. The experimental results show that the vectorized model effectively captures the correlation between the various objectives and improves the overall accuracy of multi-objective prediction.

[0080] A set of earthquake records containing time series is input, and LSTM and Transformer are used to predict the time history of vertex displacement of reinforced concrete frame structures. The results show that the LSTM model performs better in capturing the characteristics of long sequences, and the prediction accuracy of time history is higher than that of the Transformer model.

[0081] The vectorized model is used to conduct a univariate fragility analysis of the peak ground acceleration on the reinforced concrete frame structure, and its failure probability under different earthquake intensities is calculated. The fragility curve shows that the failure probability increases with the increase of the peak ground acceleration.

[0082] The multivariate fragility curve is calculated by sequence-to-sequence model under the joint input of spectral acceleration, peak ground acceleration and cumulative absolute acceleration. The results show that after considering the correlation between variables, the fragility curve is closer to the actual engineering conditions.

[0083] Working principle of the present invention:

[0084] 1. Data generation and preprocessing: Latin hypercube sampling is used to generate a multidimensional sample set covering geometry, materials, and seismic inputs, and the finite element method is used to perform random time history analysis on reinforced concrete frame structures to generate the dynamic response data set required for training the proxy model. A sensitivity analysis of the input features was performed, and it was found that spectral acceleration, cumulative absolute acceleration, and peak ground acceleration are the key parameters affecting the response of reinforced concrete frames.

[0085] 2. Agent model construction and training: According to the task requirements, point models, vectorized models and sequence-to-sequence models are constructed respectively, and random time history analysis data are used for training.

[0086] 3. Vulnerability analysis: Combined with the constructed proxy model, the Monte Carlo simulation method is used to perform univariate and multivariate seismic vulnerability analysis:

[0087] Calculation method: Based on the Gardoni model (predicted vulnerability) and the Wen model (median vulnerability), the vulnerability of reinforced concrete frame structures under different earthquake intensities is estimated.

[0088] Parameter adjustment: Distribution fitting is performed on the earthquake input parameters to keep them reasonably correlated (e.g., spectral acceleration is positively correlated with peak ground acceleration).

[0089] 4. Multi-objective output and multi-variable vulnerability analysis:

[0090] The present invention proposes a multi-objective proxy model framework based on vectorization and sequence output, which can synchronously output multi-objective responses for univariate and multivariate seismic vulnerability assessment.

[0091] Univariate fragility analysis: Calculate the fragility curve of a reinforced concrete frame structure based on a single earthquake characteristic (such as peak ground acceleration or spectral acceleration).

[0092] Multivariate vulnerability analysis: Multiple earthquake characteristics (such as spectral acceleration, peak ground acceleration and cumulative absolute acceleration) are used as input to predict the failure probability distribution of the structure under combined earthquake inputs through a proxy model.

[0093] Monte Carlo method: Combining the surrogate model and Monte Carlo simulation, the failure probability of reinforced concrete frame structures under different earthquake intensities is calculated, providing a rapid assessment tool for earthquake disaster emergency management.

[0094] Application prospects of the present invention:

[0095] 1. Individual structure reliability analysis: Applicable to rapid earthquake demand prediction and reliability assessment of reinforced concrete frame structures, which can provide technical support for seismic design and performance evaluation.

[0096] 2. Group structural disaster assessment: The method of the present invention has good scalability and can be further applied to community-level structural group analysis to support urban disaster prevention and control planning and emergency response.

[0097] 3. Engineering decision support: By quickly generating multi-objective earthquake response and vulnerability analysis results, the present invention provides data support for engineering decision-making, resource allocation and post-disaster recovery.

Claims

1. A method for earthquake demand prediction based on vectorization and sequence-to-sequence proxy model, characterized by: The steps include: (1) Generate multidimensional input parameters including geometry, material and seismic boundary conditions based on Latin hypercube sampling; (2) Based on multi-dimensional input parameters, a dynamic response database of reinforced concrete frame structures is generated using finite element stochastic time history analysis; (3) A deep learning algorithm is used to train point models, vectorized models, and sequence-to-sequence models based on a dynamic response database, which are used for structural response prediction and multi-objective reliability analysis, respectively.

2. The earthquake demand forecasting method based on vectorization and sequence-to-sequence proxy model according to claim 1, characterized in that: In the step (1), the multidimensional input parameters include parameter one based on geometric properties, parameter two based on material properties, and parameter three based on seismic boundary condition properties.

3. The earthquake demand forecasting method based on vectorization and sequence-to-sequence proxy model as claimed in claim 2, characterized in that: In the step (1), the parameter one includes the floor height, single span length, single span width, column width, beam depth, column cross-sectional area, beam cross-sectional area, concrete cover thickness, and floor thickness; the parameter two includes the concrete strength grade, steel bar yield strength, steel bar stiffness ratio before and after yield, elastic modulus, and Poisson's ratio; the parameter three includes the earthquake magnitude, fault zone length, earthquake duration, peak ground acceleration, spectral acceleration, and cumulative absolute acceleration.

4. The earthquake demand forecasting method based on vectorization and sequence-to-sequence proxy model as claimed in claim 3 is characterized by: In the step (3), the point model refers to: realizing rapid prediction of single-target output response through DNN, which is suitable for single-variable analysis tasks of structural dynamic response; the vectorized model refers to: designing a network model of vectorized output structure based on DNN, simple LSTM and Transformer, so as to realize synchronous prediction of multi-target responses; the sequence-to-sequence model refers to: constructing a network model using complex LSTM and Transformer architecture to capture the time series characteristics of earthquake input and generate the time history of structural dynamic response.

5. The earthquake demand forecasting method based on vectorization and sequence-to-sequence proxy model according to claim 4 is characterized by: In the step (3), the vectorized model is used to achieve multi-objective synchronous output, and the output of the multi-objective response includes the maximum top displacement, maximum inter-story displacement angle, maximum acceleration and maximum base shear force of the reinforced concrete frame structure.

6. The earthquake demand forecasting method based on vectorization and sequence-to-sequence proxy model according to claim 5, characterized in that: In the step (3), the sequence-to-sequence model captures the time series characteristics of the earthquake input through LSTM and Transformer networks.

7. The earthquake demand forecasting method based on vectorization and sequence-to-sequence proxy model according to claim 6 is characterized by: In the step (3), based on the constructed point model, vectorized model and sequence-to-sequence model, the Monte Carlo method is used to perform univariate and multivariate vulnerability analysis, and a relationship curve between the earthquake intensity index and the failure probability is output.

8. The earthquake demand forecasting method based on vectorization and sequence-to-sequence proxy model according to claim 7, characterized in that: In the step (3), the earthquake intensity indicators include spectral acceleration, peak ground acceleration and cumulative absolute acceleration.

Images