Building group seismic peak response evaluation method fusing measured data and simulated data

Abstract

This invention provides a method for assessing the peak seismic response of building complexes by integrating measured and simulated data. It belongs to the interdisciplinary field of civil engineering disaster prevention and mitigation and artificial intelligence. The method establishes a building foundation database and a ground motion database. Source domain data obtained from numerical simulation is used as large-scale simulation data, and a deep neural network is trained to obtain a teacher model. Knowledge distillation is performed on the student model using the teacher model, extracting high-rise response characteristics from the teacher model using a teacher-student structure. Combined with a transfer learning mechanism, the simplified model after distillation is fine-tuned with limited target domain data to obtain a customized assessment model for the seismic response of building complexes in a specific region. This achieves cross-domain fusion of simulated and measured data, solving the problems of large distribution differences, difficulty in fusion, and insufficient model generalization performance between large-scale simulation data and limited measured data. Therefore, it significantly improves the accuracy of seismic response prediction under limited target domain data conditions.

Description

Technical Field

[0001] This invention relates to the field of interdisciplinary technology of disaster prevention and mitigation in civil engineering and artificial intelligence, and in particular to a method for assessing the peak seismic response of building complexes by integrating measured data and simulation data. Background Technology

[0002] Extensive earthquake damage surveys have shown that building collapse and destruction are the primary causes of casualties. Therefore, it is of great significance to conduct rapid and accurate real-time assessments of the overall earthquake damage status (such as peak response) of building complexes within urban areas during the critical post-earthquake rescue period.

[0003] Traditional methods for assessing earthquake damage to building complexes mainly include vulnerability methods, capacity requirement analysis methods, and time-history analysis-based methods. However, in practical applications, it is difficult to balance timeliness and accuracy. Vulnerability methods typically rely on historical earthquake damage data to establish vulnerability curves or matrices, making them highly dependent on historical data. Furthermore, they usually only consider a limited set of ground motion parameters, resulting in a limited data range and a single parameter dimension, making it difficult to accurately reflect the complexity of a specific earthquake. Capacity requirement analysis methods determine earthquake damage by analyzing the intersection of building capacity curves and earthquake demand curves, falling under the category of static analysis. Their drawback lies in their inability to effectively capture higher-order vibration modes and damage concentration characteristics, resulting in limited assessment accuracy. While time-history analysis-based methods can fully consider ground motion and building characteristics, offering higher accuracy, the computational load is relatively large. When dealing with urban-scale building complexes (tens of thousands of buildings), the computational load increases exponentially, leading to excessive time consumption and failing to meet the timeliness requirements of post-earthquake "real-time assessment."

[0004] Furthermore, data-driven methods based on deep learning (such as LSTM (Long Short-Term Memory), CNN (Convolutional Neural Network), and ANN (Artificial Neural Network)) are widely used in earthquake damage assessment due to their millisecond-level computational efficiency. These methods typically utilize massive amounts of numerical simulation data to establish a mapping relationship between ground motion and building response, but they suffer from "model bias" and "data silos." Specifically, if a model is trained using purely simulated data, the model's construction relies heavily on simulation data. However, simulation models are often based on idealized assumptions (such as material constitutive models), inevitably deviating from the real physical world. This results in models trained solely on simulation data often lacking reliability in practical applications and failing to accurately predict building responses during real earthquakes. Conversely, while models trained using purely measured data are reliable, strong earthquake observation data is extremely scarce and covers a very limited range of building types. Data shows that models trained solely on measured data have an earthquake damage prediction accuracy of only 72.15%, exhibiting poor generalization ability and failing to address the vast urban building clusters lacking monitoring equipment.

[0005] Furthermore, existing technical solutions are beginning to utilize transfer learning techniques to transfer physical knowledge or simulation data to measured data, thereby improving prediction performance under limited data conditions. For example, adversarial domain adaptation methods are used to achieve label-free seismic damage diagnosis, or transfer learning is used to predict the seismic response of high-speed railway bridge-track systems. However, the application of such solutions in structural engineering is mostly limited to damage diagnosis and response prediction of individual buildings or specific structures (such as bridges), lacking research at the level of urban building clusters. Due to the high heterogeneity of urban building clusters (covering different structural types, number of stories, and construction dates), existing individual transfer methods are difficult to directly extend to solve assessment problems involving large-scale, multi-type buildings with vastly different characteristic distributions, and thus cannot form a general regional prediction capability.

[0006] Therefore, how to conduct high-precision assessment of earthquake damage to building complexes is an urgent problem to be solved. Summary of the Invention

[0007] In view of this, the purpose of the present invention is to provide a method for assessing the peak seismic response of building complexes by integrating measured data and simulation data, so as to alleviate at least some of the above-mentioned technical problems.

[0008] In a first aspect, embodiments of the present invention provide a method for evaluating the peak seismic response of building complexes by integrating measured data and simulation data. The method includes: establishing a building foundation database and a ground motion database; wherein the building foundation database contains building source parameters, including foundation parameters for various building types; the ground motion database contains ground motion source parameters, including multiple historical earthquake data; performing numerical simulations based on the building foundation database and the ground motion database to obtain earthquake response source parameters, and constructing source domain data containing building source parameters, ground motion source parameters, and earthquake response source parameters; wherein the earthquake response source parameters include maximum response label information; and utilizing the source domain data... A teacher model is obtained by training a deep neural network; a student model is constructed, and the student model is distilled using the teacher model to obtain a simplified model; the network structure complexity of the student model is lower than that of the teacher model; target domain data is obtained, including measured seismic response data of building clusters within the target area and corresponding target ground motion data; based on the target domain data, the simplified model is fine-tuned through a transfer learning mechanism to obtain an evaluation model; real-time ground motion parameters of the area to be evaluated are obtained and input into the evaluation model so that the evaluation model outputs the peak seismic response evaluation results of the building clusters in the area to be evaluated.

[0009] Optionally, a deep neural network is trained using source domain data to obtain a teacher model, including: using seismic motion source parameters and building source parameters as inputs to the deep neural network, using seismic response source parameters as outputs of the deep neural network, establishing a mapping relationship between the inputs and outputs of the deep neural network, and using the trained deep neural network as a teacher model.

[0010] Optionally, knowledge distillation is performed on the student model using the teacher model to obtain a simplified model after distillation. This includes: keeping the parameters of the teacher model fixed and simultaneously inputting the target source domain data into both the teacher model and the student model; wherein the target source domain data is obtained by filtering the source domain data based on the basic parameters of the building clusters within the target area; and optimizing the parameters of the student model by minimizing the loss function until the student model converges, so as to transfer the mapping relationship learned by the teacher model to the student model and obtain the simplified model after distillation.

[0011] Optionally, the method further includes: creating a loss function; wherein the loss function includes: the difference loss between the output of the student model and the seismic response source parameters, and the distillation loss between the output of the student model and the output of the teacher model.

[0012] Optionally, the basic parameters include information on building structure type, building height, number of building floors, and building construction year; the historical earthquake data has complete acceleration time history and includes various seismic intensity indices calculated based on the seismic energy, duration, and frequency of historical earthquakes.

[0013] Optionally, numerical simulations are performed based on a building foundation database and a ground motion database to obtain seismic response source parameters. This includes: for each building in the building foundation database, the acceleration time history of each historical earthquake data in the ground motion database is sequentially input to perform urban-scale seismic response numerical simulations, and the maximum response label information corresponding to each historical earthquake data is calculated. The maximum response label information includes the maximum inter-story drift angle, the maximum top floor acceleration, and the maximum floor acceleration value. The maximum response label information corresponding to multiple buildings in the building foundation database for multiple historical earthquake data is used as the seismic response source parameters.

[0014] Optionally, acquiring target domain data includes: acquiring the foundation parameters, measured ground ground acceleration data of the building base, and measured seismic response data of the monitored buildings in the building complex within the target area; wherein, the measured seismic response data includes the maximum acceleration of the top floor of the building; calculating the characteristic parameters of the measured ground ground acceleration data of the building base as the target ground motion data of the monitored buildings; and using the measured seismic response data and corresponding target ground motion data of multiple monitored buildings in the building complex within the target area as target domain data.

[0015] Optionally, based on the target domain data, the simplified model is fine-tuned through a transfer learning mechanism to obtain an evaluation model, including: keeping some layers of the simplified model unchanged, and fine-tuning the remaining layers of the simplified model according to the target domain data and the transfer learning mechanism until the convergence condition is met, and using the fine-tuned simplified model as the evaluation model.

[0016] Secondly, embodiments of the present invention also provide a system for assessing the peak seismic response of building complexes by integrating measured data and simulation data. This system includes: The module is used to establish a building foundation database and a ground motion database. The building foundation database contains building source parameters, which include foundation parameters for various building types. The ground motion database contains ground motion source parameters, which include multiple historical earthquake data. The module is used to perform numerical simulations based on the building foundation database and the ground motion database to obtain seismic response source parameters and construct source domain data containing building source parameters, ground motion source parameters and seismic response source parameters; among them, the seismic response source parameters contain the maximum response label information corresponding to multiple historical earthquake data. The training module is used to train a deep neural network using source domain data to obtain a teacher model; The distillation module is used to build a student model and perform knowledge distillation on the student model using the teacher model to obtain a simplified model after distillation. The network structure complexity of the student model is lower than that of the teacher model. The acquisition module is used to acquire target domain data; the target domain data includes: measured seismic response data of building clusters within the target area and corresponding target ground motion data; The fine-tuning module is used to fine-tune the simplified model based on the target domain data through a transfer learning mechanism to obtain the evaluation model; The evaluation module is used to obtain real-time ground motion parameters of the area to be evaluated and input the real-time ground motion parameters into the evaluation model so that the evaluation model can output the peak seismic response evaluation results of the building complex in the area to be evaluated.

[0017] Thirdly, embodiments of the present invention also provide 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 computer program to implement the steps of the method described in the first aspect.

[0018] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the method described in the first aspect.

[0019] The embodiments of the present invention bring the following beneficial effects: This invention provides a method for assessing the peak seismic response of building complexes by integrating measured and simulated data. First, a building foundation database and a ground motion database are established. Numerical simulations are then performed based on these databases to obtain seismic response source parameters, constructing a source domain data set containing building source parameters, ground motion source parameters, and seismic response source parameters. Next, a deep neural network is trained using the source domain data to obtain a teacher model. A student model is then constructed, and the teacher model is used to perform knowledge distillation on the student model to obtain a simplified model. Next, target domain data is acquired, and based on this data, the simplified model is fine-tuned using a transfer learning mechanism to obtain an evaluation model. Finally, real-time ground motion parameters of the area to be evaluated are obtained and input into the evaluation model, enabling the model to output the peak seismic response assessment results for the building complex in the area to be evaluated. The aforementioned assessment method, on the one hand, utilizes limited target domain data to fine-tune a simplified model built from massive source domain data, thereby overcoming the problem of low accuracy caused by insufficient training samples of purely measured data and improving the assessment model's accuracy. On the other hand, by first distilling and simplifying the model's complexity and then correcting the model's physical biases through transfer learning mechanisms, it not only significantly improves the assessment model's accuracy but also meets the needs of real-time post-earthquake assessment. After an earthquake, the assessment model can quickly assess the safety status of building complexes in the area to be assessed using a small amount of real-time ground motion data, improving the accuracy of earthquake response prediction and enhancing the scientific nature of post-earthquake emergency response.

[0020] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained through the structures particularly pointed out in the description and the drawings.

[0021] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0022] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific 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 from these drawings without creative effort.

[0023] Figure 1 A flowchart illustrating a method for evaluating the peak seismic response of building complexes by integrating measured and simulated data, provided in an embodiment of the present invention; Figure 2A schematic diagram comparing the determination coefficients and mean squared errors of an evaluation model and a non-transfer model provided in an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the accuracy of building earthquake damage prediction in an embodiment of the present invention. Figure 4 A schematic diagram of a building complex seismic peak response assessment system that integrates measured data and simulation data, provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions 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.

[0025] To facilitate understanding of this embodiment, the embodiments of the present invention will be described in detail below.

[0026] Example 1 This invention provides a method for assessing the peak seismic response of building complexes by integrating measured and simulated data, such as... Figure 1 As shown, the method includes the following steps: Step S102: Establish the building foundation database and the seismic motion database.

[0027] The building database contains building source parameters, including basic parameters for various building types. These parameters include building structure type, building height, number of floors, and construction year information. In practical applications, building structure types include various types such as timber, steel, and reinforced concrete frame structures. The number of floors covers the range of buildings with available measured data. The construction year information is divided into multiple intervals based on specific time points, such as each interval including a whole year, which can be adjusted according to actual conditions. Therefore, various types of building models can be constructed based on the combination of these four dimensions of basic parameters: building structure type, building height, number of floors, and construction year information.

[0028] Furthermore, the seismic motion database contains seismic source parameters, which include multiple historical earthquake data. Each historical earthquake data has a complete acceleration time history and includes various seismic intensity indices calculated based on the seismic energy, duration, and frequency of the historical earthquakes. In this embodiment of the invention, 19 seismic intensity indices are preferred, specifically including: peak ground acceleration, peak ground velocity, peak ground displacement, maximum velocity to maximum acceleration ratio, cumulative absolute acceleration, cumulative absolute velocity, cumulative absolute displacement, Arias intensity, root mean square of acceleration, root mean square of velocity, root mean square of displacement, characteristic intensity, effective duration, bracketed duration, peak spectral acceleration, peak spectral velocity, acceleration spectral intensity, Hausner intensity, and dominant period, which are confirmed through corresponding numerical calculation formulas.

[0029] Step S104: Numerical simulation is performed based on the building foundation database and the ground motion database to obtain the seismic response source parameters, and source domain data containing building source parameters, ground motion source parameters and seismic response source parameters is constructed.

[0030] Specifically, large-scale numerical simulations are conducted based on building foundation databases and ground motion databases to calculate the seismic response of buildings under different ground motion inputs, obtaining seismic response source parameters. These seismic response source parameters include maximum response label information, and a simulation response database containing building source parameters, ground motion source parameters, and seismic response source parameters is constructed as source domain data. Therefore, by defining large-scale numerical simulation data as source domain data, a model with powerful feature extraction capabilities can be trained using massive amounts of source domain data.

[0031] Step S106: Train a deep neural network using source domain data to obtain a teacher model.

[0032] Specifically, seismic motion source parameters and building source parameters are used as inputs to a deep neural network, while seismic response source parameters are used as outputs. A mapping relationship between the inputs and outputs of the deep neural network is established, and the trained deep neural network is used as a teacher model. The deep neural network is preferably a multi-head attention deep neural network, which learns the nonlinear mapping relationship between the inputs and outputs. Because the multi-head attention mechanism can automatically capture the nonlinear correlation weights between different seismic motion source parameters and building responses (i.e., seismic response source parameters), the final evaluation model maintains extremely high prediction accuracy even when facing heterogeneous building complexes with complex structural types and large age spans in cities, further improving the evaluation accuracy of the model.

[0033] Step S108: Construct a student model, and use the teacher model to perform knowledge distillation on the student model to obtain a simplified model after distillation.

[0034] In this model, the network structure complexity of the student model is lower than that of the teacher model; for example, the student model has fewer layers than the teacher model. Therefore, knowledge distillation can be performed on the student model using the teacher model, transferring the feature mappings learned by the teacher model to the student model, resulting in a simplified, distilled model. Thus, before transfer learning, a teacher-student network architecture is introduced. The complex and accurate teacher model guides the training of the simple student model. This not only achieves model compression and reduces computational complexity, meeting the needs of real-time post-earthquake assessment, but also forces the student model to learn richer hidden knowledge contained in the teacher model through the transfer of soft labels, laying a more robust feature foundation for subsequent transfer of experimental data.

[0035] Step S110: Obtain target domain data.

[0036] The target domain data includes: measured seismic response data of building complexes within the target area and corresponding target ground motion data; that is, measured seismic response data and corresponding target ground motion data of at least some buildings within the target area are collected to construct a measured response database as target domain data, thereby breaking the limitation of traditional schemes that rely on only a single data source and further improving the evaluation accuracy of the evaluation model.

[0037] Step S112: Based on the target domain data, the simplified model is fine-tuned through a transfer learning mechanism to obtain the evaluation model.

[0038] The transfer learning mechanism refers to using target domain data containing real monitoring information to fine-tune the parameters of a distilled simplified model, achieving knowledge transfer between the source and target domain data. This allows the model to adapt to the distribution characteristics of the measured data, thereby improving the accuracy of predicting real seismic responses. Furthermore, the transfer learning mechanism breaks down the barriers between simulation and measured data, effectively integrating the breadth of knowledge from simulation data with the precision of knowledge from measured data. While retaining the advantage of the broad coverage of simulation data (i.e., source domain data) (e.g., covering 501 building types), it introduces the real physical characteristics of the measured data, reducing the systematic bias of the simulation model and improving the reliability of the prediction results. This makes the evaluation model suitable for real-time prediction of seismic responses of building clusters at the urban scale.

[0039] Therefore, in the process of evaluating model generation, the model complexity is first simplified through knowledge distillation, and then the physical bias of the model is corrected through transfer learning. This model optimization mechanism, which combines knowledge distillation and transfer learning, improves the evaluation accuracy of the evaluation model.

[0040] Step S114: Obtain real-time ground motion parameters of the area to be evaluated, and input the real-time ground motion parameters into the evaluation model so that the evaluation model outputs the peak seismic response evaluation results of the building complex in the area to be evaluated.

[0041] In practical applications, after an earthquake, without requiring time-consuming building-by-building time-history analysis, the assessment model can quickly evaluate the safety status of a cluster of buildings in the area to be assessed using a small amount of real-time ground motion data. It can even output the peak response assessment results for all buildings in the urban area (including unmonitored buildings) within seconds, thereby improving the accuracy and timeliness of earthquake damage prediction at a lower cost and enhancing the scientific rigor of post-earthquake emergency response. It should be noted that the aforementioned real-time ground motion parameters include 19 ground motion intensity indicators (covering energy, duration, and frequency).

[0042] This invention provides a method for assessing the peak seismic response of building complexes by integrating measured and simulated data. On one hand, it fine-tunes a simplified model built from massive source domain data using limited target domain data, overcoming the problem of low accuracy caused by insufficient training samples from purely measured data, thus improving the assessment accuracy. On the other hand, by first distilling and simplifying the model's complexity, and then correcting physical biases through transfer learning, it not only significantly improves the assessment accuracy but also meets the needs of real-time post-earthquake assessment. After an earthquake, the assessment model can quickly assess the safety status of building complexes in the area to be assessed using a small amount of real-time ground motion data, improving the accuracy of earthquake response prediction and enhancing the scientific rigor of post-earthquake emergency response.

[0043] In one implementation, knowledge distillation is performed on the student model using the teacher model to obtain a simplified model after distillation. This includes: keeping the parameters of the teacher model fixed and simultaneously inputting target source domain data into both the teacher model and the student model; wherein the target source domain data is obtained by filtering source domain data based on the basic parameters of building clusters within the target area, such as filtering source domain data based on building height; and optimizing the parameters of the student model by minimizing the loss function until the student model converges, so as to transfer the mapping relationship learned by the teacher model to the student model and obtain the simplified model after distillation.

[0044] The loss function is created; in this embodiment of the invention, the loss function includes: the difference loss between the output of the student model and the seismic response source parameters, and the distillation loss between the output of the student model and the output of the teacher model; the loss function L The specific expression is as follows: (1) in, This represents the output of the student model. Indicates the source parameters of the earthquake response. This represents the output of the teacher model. n Indicates the number of samples. This represents the supervision weight coefficient.

[0045] Therefore, by transferring the mapping relationships from the teacher model to the student model using knowledge distillation, the model structure is simplified due to the lower network complexity of the student model compared to the teacher model. This allows for a significant improvement in computational efficiency while maintaining high accuracy. After an earthquake, once real-time ground motion parameters are obtained, peak seismic response simulations for thousands of buildings in the affected area can be completed within seconds, achieving real-time post-earthquake assessment. This provides highly timely data support for post-earthquake emergency response decisions and enhances the scientific rigor of the emergency response.

[0046] In one implementation, numerical simulation is performed based on a building foundation database and a ground motion database to obtain seismic response source parameters. This includes: for each building in the building foundation database, the acceleration time history of each historical earthquake data in the ground motion database is sequentially input to perform urban-scale seismic response numerical simulation, and the maximum response label information corresponding to each historical earthquake data is calculated; wherein, the maximum response label information includes the maximum inter-story drift angle, the maximum top floor acceleration, and the maximum floor acceleration value; the maximum response label information corresponding to multiple buildings in the building foundation database for multiple historical earthquake data is used as the seismic response source parameters.

[0047] Therefore, by calculating the maximum response label information corresponding to each building in each historical earthquake data point, and matching the building, historical earthquake data, and maximum response label information, earthquake response source parameters are formed. This generates massive amounts of source domain data based on the building source parameters, ground motion source parameters, and earthquake response source parameters. Furthermore, abandoning traditional simple linear regression, a multi-head attention deep neural network is employed, integrating 19 ground motion intensity indicators (covering energy, duration, and frequency) with multi-dimensional building attributes (building structure type, building height, number of floors, and building construction date) to establish a multi-dimensional feature fusion-based urban-level building cluster earthquake response mapping mechanism.

[0048] In one embodiment, acquiring target domain data includes: acquiring the foundation parameters, measured ground motion acceleration data of the building base, and measured seismic response data of the monitored buildings in the building complex within the target area; wherein the measured seismic response data includes the maximum acceleration of the top floor of the building; calculating the characteristic parameters of the measured ground motion acceleration data of the building base as the target ground motion data of the monitored buildings; and using the measured seismic response data and the corresponding target ground motion data of multiple monitored buildings in the building complex within the target area as target domain data.

[0049] Among these methods, the building strong earthquake observation network can be used to collect the foundation parameters of the monitored buildings, the measured ground ground acceleration data of the building base, and the measured seismic response data; in order to maximize the use of existing building strong earthquake observation network data, so that a small amount of monitoring data can serve a wider range of unmonitored areas and reduce the dependence on large-scale hardware deployment.

[0050] In summary, the method for assessing the peak seismic response of building complexes by integrating measured and simulated data provided in this invention first establishes a building foundation database and a ground motion database, then trains a teacher model, obtains a student model through knowledge distillation, and finally uses target domain data for transfer learning to obtain a fine-tuned assessment model. Simultaneously, large-scale numerical simulation data is defined as source domain data, and scarce measured data is used as target domain data. A model with strong feature extraction capabilities is first trained using massive amounts of source domain data, and then the model's parameters are fine-tuned using limited target domain data (i.e., measured data). This achieves fine-tuning of a simplified model built from massive amounts of source domain data using limited target domain data. This not only solves the problems of underfitting and poor generalization ability caused by insufficient measured data, but also corrects the systematic deviation between the pure simulation model and the real physical environment, achieving effective integration of virtual and real knowledge. It overcomes the problem of low accuracy (e.g., 72.15%) of the assessment model due to insufficient training samples of pure measured data, thus improving the assessment accuracy of the assessment model.

[0051] Furthermore, compared to training a simplified model using only measured target domain data, this embodiment of the invention also fine-tunes the simplified model through a transfer learning mechanism, significantly improving the prediction accuracy of the evaluation model. For example, as... Figure 2 As shown, the horizontal axis represents the measured value of the maximum top-level acceleration, and the vertical axis represents the predicted value of the maximum top-level acceleration. The non-transfer model is the model trained only using source domain data (i.e., the teacher model). This method represents the evaluation model obtained by fine-tuning the distilled student model through a transfer learning mechanism. The simplified model is fine-tuned through a transfer learning mechanism, and the determination coefficient (R²) of the evaluation model is used. 2 The mean square error (MSE) of the evaluation model increased from 0.868 to 0.909, while the mean square error (MSE) of the evaluation model decreased from 0.146 to 0.100, thus significantly improving the prediction accuracy of the evaluation model and consequently improving the accuracy of earthquake damage prediction.

[0052] In addition, such as Figure 3As shown, the transfer model refers to the evaluation model obtained by fine-tuning the simplified model through the transfer learning mechanism provided in this embodiment of the invention; the non-transfer model is the model trained only using source domain data (i.e., the teacher model), and the experimental model refers to the model obtained by training the deep neural network only using experimental data (the architecture of the experimental model is consistent with that of the student model). It can be seen that, compared with the non-transfer model and the experimental model, the transfer model improves the accuracy of building earthquake damage prediction from 72.15% to 81.01%, thereby significantly improving the credibility of the evaluation results output by the evaluation model.

[0053] Furthermore, due to the high generalization ability of the assessment model, even for buildings in the city without monitoring equipment, high-precision response prediction can be achieved by simply inputting their basic attributes and regional ground motion parameters. This breaks down data silos, achieves full coverage assessment at the urban area level, and solves the assessment problem of buildings without monitoring.

[0054] In one implementation, a simplified model is fine-tuned based on target domain data using a transfer learning mechanism to obtain an evaluation model. This includes: keeping some layers of the simplified model unchanged, and fine-tuning the remaining layers of the simplified model according to the target domain data and the transfer learning mechanism until convergence is achieved, and using the fine-tuned simplified model as the evaluation model.

[0055] Specifically, the simplified model is a multi-layered deep neural network model. During the fine-tuning of the simplified model using target domain data and a transfer learning mechanism, a portion of the simplified model's layers are kept constant, while the remaining layers are fine-tuned. For example, if the simplified model includes a first, second, and third layer, the first and second layers are kept unchanged during fine-tuning, while the third layer is fine-tuned based on the target domain data and the transfer learning mechanism. This process is repeated for all cases until a convergence condition is met. The preferred convergence condition is the minimum mean MSE during cross-validation. Finally, the fine-tuned simplified model is used as the evaluation model. Therefore, by correcting the physical biases of the model through the transfer learning mechanism, the evaluation accuracy of the evaluation model is improved, making it suitable for real-time prediction of seismic responses in urban-scale building complexes.

[0056] Example 2 Based on the above method embodiments, this invention also provides a system for evaluating the peak seismic response of building complexes that integrates measured data and simulation data, such as... Figure 4 As shown, the system includes, in sequence, a setup module 41, a construction module 42, a training module 43, a distillation module 44, an acquisition module 45, a fine-tuning module 46, and an evaluation module 47; the functions of each module are as follows: Module 41 is used to establish a building foundation database and a ground motion database. The building foundation database contains building source parameters, which include foundation parameters for various building types. The ground motion database contains ground motion source parameters, which include multiple historical earthquake data. Module 42 is used to perform numerical simulations based on the building foundation database and the ground motion database to obtain earthquake response source parameters and construct source domain data containing building source parameters, ground motion source parameters and earthquake response source parameters; wherein, the earthquake response source parameters contain the maximum response label information corresponding to multiple historical earthquake data; Training module 43 is used to train a deep neural network using source domain data to obtain a teacher model; Distillation module 44 is used to construct a student model and perform knowledge distillation on the student model using the teacher model to obtain a simplified model after distillation; wherein, the network structure complexity of the student model is lower than that of the teacher model. The acquisition module 45 is used to acquire target domain data; wherein, the target domain data includes: measured seismic response data of building groups within the target area and corresponding target ground motion data; The fine-tuning module 46 is used to fine-tune the simplified model based on the target domain data through a transfer learning mechanism to obtain the evaluation model; The evaluation module 47 is used to obtain real-time ground motion parameters of the area to be evaluated and input the real-time ground motion parameters into the evaluation model so that the evaluation model outputs the peak seismic response evaluation results of the building complex in the area to be evaluated.

[0057] The seismic peak response assessment system for building complexes provided in this invention integrates measured and simulated data. On one hand, it uses limited target domain data to correct a simplified model built from massive source domain data, thus overcoming the problem of low accuracy caused by insufficient training samples from purely measured data and improving the assessment model's accuracy. On the other hand, by first distilling and simplifying the model's complexity and then correcting physical biases through transfer learning, it not only significantly improves the assessment model's accuracy but also meets the needs of real-time post-earthquake assessment. After an earthquake, the assessment model can quickly assess the safety status of building complexes in the area to be assessed using a small amount of real-time ground motion data, improving the accuracy of earthquake damage prediction and enhancing the scientific rigor of post-earthquake emergency response.

[0058] Optionally, the training module 43 is also used to: use the seismic motion source parameters and building source parameters as inputs to the deep neural network, use the seismic response source parameters as outputs to the deep neural network, establish a mapping relationship between the inputs and outputs of the deep neural network, and use the trained deep neural network as a teacher model.

[0059] Optionally, the distillation module 44 is also used to: keep the parameters of the teacher model fixed and simultaneously input the target source domain data into the teacher model and the student model; wherein the target source domain data is obtained by filtering the source domain data based on the basic parameters of the building clusters in the target area; and optimize the parameters of the student model by minimizing the loss function until the student model converges, so as to transfer the mapping relationship learned by the teacher model to the student model and obtain the simplified model after distillation.

[0060] Optionally, the system further includes: creating a loss function; wherein the loss function includes: the difference loss between the output of the student model and the seismic response source parameters, and the distillation loss between the output of the student model and the output of the teacher model.

[0061] Optionally, the basic parameters include information on building structure type, building height, number of building floors, and building construction year; the historical earthquake data has complete acceleration time history and includes various seismic intensity indices calculated based on the seismic energy, duration, and frequency of historical earthquakes.

[0062] Optionally, numerical simulations are performed based on a building foundation database and a ground motion database to obtain seismic response source parameters. This includes: for each building in the building foundation database, the acceleration time history of each historical earthquake data in the ground motion database is sequentially input to perform urban-scale seismic response numerical simulations, and the maximum response label information corresponding to each historical earthquake data is calculated. The maximum response label information includes the maximum inter-story drift angle, the maximum top floor acceleration, and the maximum floor acceleration value. The maximum response label information corresponding to multiple buildings in the building foundation database for multiple historical earthquake data is used as the seismic response source parameters.

[0063] Optionally, the acquisition module 45 is further configured to: acquire the foundation parameters, measured ground motion acceleration data of the building base, and measured seismic response data of the monitored buildings in the building complex within the target area; wherein the measured seismic response data includes the maximum acceleration of the top floor of the building; calculate the characteristic parameters of the measured ground motion acceleration data of the building base as the target ground motion data of the monitored buildings; and use the measured seismic response data and the corresponding target ground motion data of multiple monitored buildings in the building complex within the target area as target domain data.

[0064] Optionally, the fine-tuning module 46 is also used to: keep some layers of the simplified model unchanged, and fine-tune the remaining layers of the simplified model according to the target domain data and the transfer learning mechanism until the convergence condition is met, and use the fine-tuned simplified model as the evaluation model.

[0065] The building cluster seismic peak response assessment system that integrates measured data and simulation data provided in this embodiment of the invention has the same technical features as the building cluster seismic peak response assessment method that integrates measured data and simulation data provided in the above embodiments, so it can also solve the same technical problems and achieve the same technical effects.

[0066] This invention also provides an electronic device, including a processor and a memory. The memory stores machine-executable instructions that can be executed by the processor. The processor executes the machine-executable instructions to implement the above-mentioned method for evaluating the peak seismic response of building complexes by integrating measured data and simulation data.

[0067] See Figure 5 As shown, the electronic device includes a processor 100 and a memory 101. The memory 101 stores machine-executable instructions that can be executed by the processor 100. The processor 100 executes the machine-executable instructions to implement the above-mentioned method for assessing the peak seismic response of building complexes by integrating measured data and simulation data.

[0068] Furthermore, Figure 5 The electronic device shown also includes a bus 102 and a communication interface 103, with the processor 100, the communication interface 103 and the memory 101 connected via the bus 102.

[0069] The memory 101 may include high-speed random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one communication interface 103 (which can be wired or wireless), such as the Internet, wide area network, local area network, metropolitan area network, etc. The bus 102 may be an ISA (Industrial Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Enhanced Industry Standard Architecture) bus, etc. These buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, Figure 5 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.

[0070] Processor 100 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of processor 100 or by instructions in software form. Processor 100 may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this invention. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this invention can be directly manifested as execution by a hardware decoding processor, or execution by a combination of hardware and software modules in the decoding processor. The software module can reside in a readily available storage medium in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory 101, and the processor 100 reads the information from memory 101 and, in conjunction with its hardware, completes the steps of the method described in the foregoing embodiments.

[0071] This embodiment also provides a computer-readable storage medium storing a computer program, which, when run by a processor, executes the above-described method for evaluating the peak seismic response of building complexes by fusing measured and simulated data.

[0072] The computer program product of the method, system and electronic device for evaluating the peak seismic response of building complexes that integrates measured data and simulation data provided in the embodiments of the present invention includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the methods described in the preceding method embodiments. For specific implementation, please refer to the method embodiments, which will not be repeated here.

[0073] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system and apparatus described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0074] Furthermore, in the description of the embodiments of the present invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in the present invention based on the specific circumstances.

[0075] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0076] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0077] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, 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, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for assessing the peak seismic response of building complexes by integrating measured and simulated data, characterized in that, The method includes: Establish a building foundation database and a ground motion database; wherein, the building foundation database contains building source parameters, which include foundation parameters for various building types; the ground motion database contains ground motion source parameters, which include multiple historical earthquake data. Numerical simulations are performed based on the building foundation database and the seismic motion database to obtain seismic response source parameters, and source domain data containing the building source parameters, the seismic motion source parameters, and the seismic response source parameters are constructed; wherein, the seismic response source parameters include maximum response label information; A deep neural network is trained using the source domain data to obtain a teacher model; A student model is constructed, and the student model is subjected to knowledge distillation using the teacher model to obtain a simplified model after distillation; wherein the network structure complexity of the student model is lower than that of the teacher model. Acquire target domain data; wherein, the target domain data includes: measured seismic response data of building clusters within the target area and corresponding target ground motion data; Based on the target domain data, the simplified model is fine-tuned through a transfer learning mechanism to obtain an evaluation model; Real-time ground motion parameters of the area to be evaluated are obtained and input into the evaluation model so that the evaluation model outputs the peak seismic response evaluation results of the building complex in the area to be evaluated.

2. The method according to claim 1, characterized in that, The step of training a deep neural network using the source domain data to obtain a teacher model includes: The seismic motion source parameters and the building source parameters are used as inputs to the deep neural network, and the seismic response source parameters are used as outputs to establish a mapping relationship between the inputs and outputs of the deep neural network. The trained deep neural network is then used as the teacher model.

3. The method according to claim 2, characterized in that, The step of using the teacher model to perform knowledge distillation on the student model to obtain a simplified model after distillation includes: Keeping the parameters of the teacher model fixed, the target source domain data is simultaneously input into both the teacher model and the student model; wherein, the target source domain data is obtained by filtering the source domain data based on the basic parameters of the building clusters within the target area; The parameters of the student model are optimized by minimizing the loss function until the student model converges, so that the mapping relationship learned by the teacher model can be transferred to the student model, resulting in a simplified model after distillation.

4. The method according to claim 3, characterized in that, The method further includes: Create the loss function; wherein the loss function includes: the difference loss between the output of the student model and the seismic response source parameters, and the distillation loss between the output of the student model and the output of the teacher model.

5. The method according to claim 1, characterized in that, The basic parameters include building structure type, building height, number of floors, and building construction year information; the historical earthquake data has complete acceleration time history and includes various ground motion intensity indicators calculated based on the earthquake energy, duration, and frequency of historical earthquakes.

6. The method according to claim 5, characterized in that, The numerical simulation based on the building foundation database and the seismic motion database yields seismic response source parameters, including: For each building in the building foundation database, the acceleration time history of each historical earthquake data in the seismic motion database is sequentially input to perform a city-scale seismic response numerical simulation and calculate the maximum response label information corresponding to each historical earthquake data; wherein, the maximum response label information includes the maximum inter-story drift angle, the maximum top floor acceleration, and the maximum floor acceleration value; The maximum response label information corresponding to multiple buildings in the building foundation database and multiple historical earthquake data is used as the earthquake response source parameter.

7. The method according to claim 1, characterized in that, The acquisition of target domain data includes: Acquire the foundation parameters, measured ground motion acceleration data of the building base, and measured seismic response data of the monitored buildings in the building complex within the target area; wherein, the measured seismic response data includes the maximum acceleration at the top floor of the building; The characteristic parameters of the measured ground motion acceleration data of the building foundation are calculated as the target ground motion data of the monitored building. The measured seismic response data of multiple monitored buildings in the building complex within the target area and the corresponding target ground motion data are used as the target domain data.

8. The method according to claim 1, characterized in that, The step of fine-tuning the simplified model based on the target domain data using a transfer learning mechanism to obtain an evaluation model includes: The number of some layers in the simplified model is kept constant, and the remaining number of layers in the simplified model is fine-tuned according to the target domain data and the transfer learning mechanism until the convergence condition is met. The fine-tuned simplified model is then used as the evaluation model.

9. 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 computer program, it implements the steps of the method described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, performs the steps of the method described in any one of claims 1-8.

Images