Building earthquake damage monitoring method based on deep learning and sparse monitoring data

Abstract

The invention provides a building earthquake damage monitoring method based on deep learning and sparse monitoring data. According to the method, modeling is carried out on a seismic response mode of a structure along a height direction through spline interpolation, data dimensions of global seismic response of buildings with different floor numbers are unified through equidistant sampling, and then monitoring data features of two sensors installed on different floors are extracted through a multi-scale convolutional neural network. And combining monitoring data features with building features extracted based on a multi-layer perceptron, and inputting the combined features into an attention-enhanced long-short-term memory neural network for feature fusion, so as to reconstruct earthquake responses of all layers of the building, thereby realizing earthquake damage monitoring of the building. The method can be applied to any building with a layer structure.

Description

Technical Field

[0001] This invention relates to the field of building seismic damage monitoring technology, and in particular to a building seismic damage monitoring method based on deep learning and sparse monitoring data. Background Technology

[0002] After an earthquake, the structural performance of buildings typically deteriorates, and they may even face serious damage or collapse. In the engineering field, the damage status of buildings is usually assessed based on the seismic response of the building structure. Efficient and low-cost building response monitoring methods can be used to quickly assess the post-earthquake damage status and functional loss of buildings, which helps in the initial screening of refuge sites and the rational allocation of post-earthquake disaster relief resources, thereby minimizing economic losses and casualties.

[0003] Currently, seismic damage monitoring of buildings is typically based on sensor monitoring data or surrogate models. Accelerometers or displacement sensors can accurately record the dynamic response data of buildings during earthquakes. If sensors are installed on all floors of a building, the inter-story drift angles and floor accelerations can be acquired in real time, thereby determining the damage state and functional losses of each floor. While this method offers sufficient accuracy, the installation and maintenance costs of the related equipment are high, thus limiting its use to seismic damage monitoring of only a few important buildings. In most cases, sensors are installed on only a few floors of the building, and the seismic response of other floors is reconstructed using modal interpolation. When the building is in a nonlinear state, the mode shapes change significantly, resulting in lower accuracy for modal interpolation. Model-updated methods are suitable for nonlinear situations. They update the mechanical parameters of the building's finite element model using optimization algorithms until the error between the simulation results and the monitoring data is less than a preset threshold, allowing for the reconstruction of the seismic response of all floors. However, each parameter update requires time-history analysis of the finite element model, and multiple updates are usually needed to achieve sufficient accuracy, making this method computationally expensive and time-consuming.

[0004] The surrogate model is a data-driven approach that uses machine learning algorithms to establish a mapping between sensor monitoring data and building seismic response. It is computationally efficient and suitable for nonlinear scenarios, but existing surrogate models lack generalization ability across structures. This is because in real-world scenarios, the installation locations and number of floors of sensors on different buildings vary, and once a surrogate model is established, its input characteristics and output dimensions cannot be changed. Therefore, existing surrogate models are only suitable for reconstructing the response of specific individual buildings and are difficult to apply to seismic damage monitoring of groups of buildings.

[0005] In summary, existing solutions struggle to balance cost, accuracy, and efficiency. Improving existing proxy models to enhance their generalization ability to buildings could enable low-cost, high-accuracy, and high-efficiency reconstruction of building seismic responses, allowing for rapid post-earthquake assessment of building damage and remaining functionality. Summary of the Invention

[0006] The purpose of this invention is to address the problems of low accuracy of existing modal interpolation methods, low computational efficiency of finite element methods, insufficient generalization ability of surrogate models for structures, and unsuitability for monitoring seismic damage in groups of buildings. This invention proposes a building seismic damage monitoring method based on deep learning and sparse monitoring data.

[0007] This invention is achieved through the following technical solution: This invention proposes a building seismic damage monitoring method based on deep learning and sparse monitoring data, the method comprising: Step 1: Collect measured or numerical simulation data of building seismic response, and process the seismic response time history to unify the sampling frequency and duration of the data; Step 2: Preprocess the data: Model the response pattern of each building along the height direction under each earthquake based on spline interpolation, and unify the data dimension of earthquake response of different floors by means of equidistant sampling. Step 3: Using building parameters, acceleration monitoring data of the two floors, and sensor location information as features, and the seismic response time history of the building to be reconstructed as a label, establish a dataset and divide it into training set, test set, and validation set; Step 4: Develop a feature extraction and fusion module and build a proxy model. Train the proxy model using the training set, and optimize the network architecture and hyperparameters of the proxy model based on the Optuna algorithm and the validation set to build a mapping relationship from features to labels. Step 5: Input the features of the test set samples into the surrogate model to obtain the output, and then perform data post-processing: Model the response pattern of each sample along the height direction based on spline interpolation, and sample at the height of each floor to derive the seismic response of all floors, and then evaluate the reconstruction accuracy of the seismic response. Step 6: Assess the damage status of the building based on the building damage criteria and the earthquake response of the reconstruction, compare it with the actual damage status of the building, and evaluate the accuracy of earthquake damage monitoring.

[0008] Furthermore, in step 1, the specific scheme for unifying the sampling frequency and duration of the data includes: (1) Resample at a frequency of 50 Hz; (2) If the duration is less than 50 seconds, use zero padding at the end of the data to make the duration reach 50 seconds; (3) If the duration exceeds 50 seconds and the peak ground acceleration occurs 25 seconds before / after, only the first / last 50 seconds of the data are retained; (4) In other cases, retain the data for 25 seconds before and after the time when the peak ground acceleration occurs.

[0009] Furthermore, step 2 specifically includes: For the inter-story drift angle at each moment According to the building floor number x 1. Seismic response of all floors of the building y 1 is used as both the independent and dependent variables, and the spline interpolation fitting function is: y = f 1( x Based on the maximum number of floors in the database; N m Determine the number of sampling points N , N Take as N m 2 / 3 of the value N When the value is a decimal, it is rounded to the nearest integer; a sampling sequence is generated according to Equation 1. x 2. Step by step in time f 1( x Upsampling To expand the data dimensions of seismic response from N s × l Transform into N × l , N s and l These are the number of building floors and the length of the IDR time history, respectively. (1).

[0010] Furthermore, in step 3, the building parameters include the number of floors, floor height, fundamental period, yield displacement, yield shear force, peak displacement, peak shear force, ultimate displacement, ultimate shear force, and stress pinching coefficient of the building analysis model; the monitoring data is the acceleration response time history of any two floors of the building; and the sensor location information is the ratio of its installation height to the total height of the building.

[0011] Furthermore, in step 3, the dataset is randomly divided into a training set, a test set, and a validation set in a ratio of 8:1:1, which are used for training, optimization, and accuracy evaluation of the surrogate model, respectively.

[0012] Furthermore, in step 4, a dynamic feature extraction module is constructed based on a multi-scale convolutional neural network to fully extract the features of the monitoring data; a static feature extraction module is constructed based on a multilayer perceptron to map building parameters and sensor location information to a feature space of a predetermined dimension; and a feature fusion module is developed based on an attention-enhanced long short-term memory network to fully integrate dynamic and static features. The above modules are combined to construct a proxy model.

[0013] Further, in step 5, the features of the test set samples are input into the surrogate model to obtain the output, and then post-processed to obtain the IDR time history of all floors of the building. The specific steps are as follows: (1) For the surrogate model output at each time step The sampling sequences in step 2 are respectively x 2. The outputs of the surrogate model are used as independent and dependent variables, respectively. The spline interpolation fitting function is: y = f 2( x (2) By building floor number x 1 is used as a new sampling sequence, step-by-step in time. f 2( x Upsampling was used to obtain the inter-story drift angles of all floors at each time point. This reduces the data dimension of the proxy model's output from N × l Transform into N s × l Then evaluate the IDR time history derived from the proxy model. With finite element analysis results The correlation coefficient, and the formula for calculating the correlation coefficient is: (2) In the formula, cov and σ These are the covariance and standard deviation, respectively.

[0014] Further, in step 6, the damage status of each floor of the test set building is evaluated based on the IDR thresholds corresponding to various damage states of the building and the IDR peak values ​​output by the surrogate model. The damage status includes basically intact (ND), slightly damaged (SD), moderately damaged (MD), severely damaged (ED), and completely destroyed (CD). The damage status of the floor with the most severe damage is taken as the overall damage status of the building. Then, the confusion matrix between the simulation results of the surrogate model and the finite element analysis results is calculated to evaluate the accuracy of earthquake damage monitoring.

[0015] The present invention also proposes an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the building earthquake damage monitoring method based on deep learning and sparse monitoring data.

[0016] The present invention also proposes a computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the steps of the building earthquake damage monitoring method based on deep learning and sparse monitoring data.

[0017] The beneficial effects of this invention are: 1. It combines low cost and high efficiency. Only two acceleration sensors need to be installed on the building to complete the earthquake damage monitoring task. The trained surrogate model can complete the earthquake response reconstruction and structural damage assessment process within one second.

[0018] 2. It has strong applicability, requiring only monitoring data from two sensors and having no restrictions on the sensor installation location. It is suitable for any sensor layout, so there is no need to remove or modify existing sensors. Attached Figure Description

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

[0020] Figure 1 This is a flowchart of the building earthquake damage monitoring method based on deep learning and sparse monitoring data as described in this invention.

[0021] Figure 2 This is a schematic diagram of the building analysis model.

[0022] Figure 3 A flowchart for data preprocessing.

[0023] Figure 4 This is a network architecture diagram for the proxy model.

[0024] Figure 5 This is a flowchart for data post-processing.

[0025] Figure 6 The bar chart shows the correlation coefficient between the time history of the inter-story drift ratio output by the surrogate model and the actual value, and a comparison chart of two cases.

[0026] Figure 7 This is a schematic diagram of the confusion matrix corresponding to the assessment results of building damage status at the floor level and overall level. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] This invention addresses the problems of low accuracy in existing modal interpolation methods, low computational efficiency in finite element methods, and insufficient generalization ability of surrogate models for structural applications, making them unsuitable for monitoring seismic damage in clustered buildings. It proposes a building seismic damage monitoring method based on deep learning and sparse monitoring data. This method models the seismic response pattern of the structure along its height using spline interpolation and unifies the data dimension of the global seismic response of buildings with different numbers of stories through equidistant sampling. Then, it extracts monitoring data features from two sensors installed on different floors using a multi-scale convolutional neural network. These monitoring data features are then merged with building features extracted using a multilayer perceptron and input into an attention-enhanced long short-term memory neural network for feature fusion to reconstruct the seismic response of all floors of the building, thereby achieving seismic damage monitoring. This invention is applicable to any building with a multi-story structure.

[0029] Specifically, in combination Figures 1 to 7 This invention proposes a building seismic damage monitoring method based on deep learning and sparse monitoring data, the method comprising: Step 1: Collect measured or numerical simulation data of building seismic response and process the seismic response time history, such as resampling and zero-filling, to unify the sampling frequency and duration of the data. Step 2: Preprocess the data: Model the response pattern of each building along the height direction under each earthquake based on spline interpolation, and unify the data dimension of earthquake response of different floors by means of equidistant sampling. Step 3: Using building parameters, acceleration monitoring data of two floors, and sensor location information as features (input to the surrogate model), and the seismic response time history of the building to be reconstructed as the label (output of the surrogate model), establish a dataset and divide it into training set, test set, and validation set; Step 4: Develop a feature extraction and fusion module and build a proxy model. Train the proxy model using the training set, and optimize the network architecture and hyperparameters of the proxy model based on the Optuna algorithm and the validation set to build a mapping relationship from features to labels. Step 5: Input the features of the test set samples into the surrogate model to obtain the output, and then perform data post-processing: Model the response pattern of each sample along the height direction based on spline interpolation, and sample at the height of each floor to derive the seismic response of all floors, and then evaluate the reconstruction accuracy of the seismic response. Step 6: Assess the damage status of the building based on the building damage criteria and the earthquake response of the reconstruction, compare it with the actual damage status of the building, and evaluate the accuracy of earthquake damage monitoring.

[0030] In step 1, the specific scheme for unifying the sampling frequency and duration of the data includes: (1) Resample at a frequency of 50 Hz; (2) If the duration is less than 50 seconds, use zero padding at the end of the data to make the duration reach 50 seconds; (3) If the duration exceeds 50 seconds and the peak ground acceleration occurs 25 seconds before / after, only the first / last 50 seconds of the data are retained; (4) In other cases, retain the data for 25 seconds before and after the time when the peak ground acceleration occurs.

[0031] Step 2 specifically includes: For the inter-story drift angle at each moment According to the building floor number x 1. Seismic response of all floors of the building y 1 is used as both the independent and dependent variables, and the spline interpolation fitting function is: y = f 1( x Based on the maximum number of floors in the database; N m Determine the number of sampling points N , N Take as N m 2 / 3 of the value N When the value is a decimal, it is rounded to the nearest integer; a sampling sequence is generated according to Equation 1. x 2. Step by step in time f 1( x Upsampling To expand the data dimensions of seismic response from N s × l Transform into N × l , N s and l These are the number of building floors and the length of the IDR time history, respectively. (1).

[0032] In step 3, the building parameters include the number of floors, floor height, fundamental period, yield displacement, yield shear force, peak displacement, peak shear force, ultimate displacement, ultimate shear force, and stress pinching coefficient of the building analysis model; the monitoring data is the acceleration response time history of any two floors of the building; and the sensor location information is the ratio of its installation height to the total height of the building.

[0033] In step 3, the dataset is randomly divided into a training set, a test set, and a validation set in a ratio of 8:1:1, which are used for training, optimization, and accuracy evaluation of the surrogate model, respectively.

[0034] In step 4, a dynamic feature extraction module is constructed based on a multi-scale convolutional neural network to fully extract the features of the monitoring data. A static feature extraction module is constructed based on a multilayer perceptron to map building parameters and sensor location information to a feature space of a predetermined dimension. A feature fusion module is developed based on an attention-enhanced long short-term memory network to fully integrate dynamic and static features. The above modules are combined to construct a proxy model.

[0035] In step 5, the features of the test set samples are input into the surrogate model to obtain the output, and then post-processed to obtain the IDR time history of all floors of the building. The specific steps are as follows: (1) For the surrogate model output at each time step The sampling sequences in step 2 are respectively x 2. The outputs of the surrogate model are used as independent and dependent variables, respectively. The spline interpolation fitting function is: y = f 2( x (2) By building floor number x 1 is used as a new sampling sequence, step-by-step in time. f 2( x Upsampling was used to obtain the inter-story drift angles of all floors at each time point. This reduces the data dimension of the proxy model's output from N × l Transform into N s × l Then evaluate the IDR time history derived from the proxy model. With finite element analysis results The correlation coefficient, and the formula for calculating the correlation coefficient is: (2) In the formula, cov and σ These are the covariance and standard deviation, respectively.

[0036] In step 6, the damage status of each floor of the test set building is evaluated based on the IDR thresholds corresponding to various damage states of the building and the IDR peak values ​​output by the surrogate model. The damage status includes basically intact (ND), slightly damaged (SD), moderately damaged (MD), severely damaged (ED), and completely destroyed (CD). The damage status of the floor with the most severe damage is taken as the overall damage status of the building. Then, the confusion matrix between the simulation results of the surrogate model and the finite element analysis results is calculated to evaluate the accuracy of earthquake damage monitoring.

[0037] Example This invention proposes a building seismic damage monitoring method based on deep learning and sparse monitoring data, the method comprising the following steps: Step 1: This invention is validated using numerical simulation data. The building numerical analysis model used is a multi-degree-of-freedom shear (MDOF) model, which is commonly used for earthquake damage simulation of low- and mid-rise buildings. A schematic diagram of this model is shown below. Figure 2 One hundred MDOF models corresponding to concrete frame structures were established using the open-source finite element software OpenSees. Then, 200 ground motions were used as seismic excitations to perform dynamic analysis on all MDOF models to obtain the inter-story drift ratio (IDR) time history of all floors of the building. The IDR time history and the floor acceleration data monitored by the sensors were processed to unify the sampling frequency and duration. The specific implementation plan is as follows: (1) Resampling was performed at a frequency of 50 Hz; (2) If the duration was less than 50 seconds, zero padding was used at the end of the data to make the duration reach 50 seconds; (3) If the duration exceeded 50 seconds and the peak ground acceleration occurred 25 seconds before / after, only the first / last 50 seconds of the data were retained; (4) In other cases, the data for 25 seconds before and after the time when the peak ground acceleration occurred were retained.

[0038] Step 2: Preprocess the seismic response data to unify the data dimensions of seismic responses for buildings with different numbers of stories. The specific implementation method is as follows: (1) For the inter-story drift angle at each time point According to the building floor number x 1. Seismic response of all floors of the building y 1 is used as both the independent and dependent variable, and the function is fitted based on spline interpolation. y = f 1( x (2) Based on the maximum number of floors in the database. N m Determine the number of sampling points N , N Take as N m 2 / 3 of the value N When the value is a decimal, it is rounded to the nearest integer; (3) Generate the sampling sequence according to Equation 1. x 2. Step by step in time f 1( x Upsampling To expand the data dimensions of seismic response from N s × l Transform into N × l , N s and l These refer to the number of building floors and the length of the IDR time history, respectively. See the detailed process flow below. Figure 3 .

[0039] (1) Step 3: Select the features of the dataset, i.e., the input to the surrogate model, including building parameters, monitoring data, and sensor location information. Building parameters include the number of floors, floor height, fundamental period, yield displacement, yield shear force, peak displacement, peak shear force, ultimate displacement, ultimate shear force, and stress pinching coefficient of the building analysis model. Monitoring data consists of the acceleration response time histories of any two floors of the building. Sensor location information is the ratio of their installation height to the total height of the building. Determine the label for the surrogate model, i.e., the output of the surrogate model, which is the inter-story displacement ratio time histories of all floors of the building. Then, create the dataset and randomly divide it into a training set, a test set, and a validation set in an 8:1:1 ratio for training, optimization, and accuracy evaluation of the surrogate model, respectively.

[0040] Step 4: Construct a dynamic feature extraction module based on a multi-scale convolutional neural network to fully extract features from the monitoring data. Construct a static feature extraction module based on a multilayer perceptron to map building parameters and sensor location information to a feature space of predetermined dimensions. Then, develop a feature fusion module based on an attention-enhanced long short-term memory network to fully integrate dynamic and static features. Combine these modules to construct a proxy model. See the specific architecture below. Figure 4 The agent model is trained using the training set, with the goal of optimizing the agent model's performance on the validation set. Hyperparameters such as network depth, network width, learning rate, and training epochs of the agent model are optimized based on the Optuna algorithm.

[0041] Step 5: Input the features of the test set samples into the surrogate model to obtain the output, and then perform post-processing to obtain the IDR time history of all floors of the building. The specific steps are as follows: (1) For the output of the surrogate model at each time step The sampling sequences in step 2 are respectively x 2. The outputs of the surrogate model are used as independent and dependent variables, respectively, and a spline interpolation function is fitted. y = f 2( x (2) By building floor number x 1 is used as a new sampling sequence, step-by-step in time. f 2( x Upsampling was used to obtain the inter-story drift angles of all floors at each time point. This reduces the data dimension of the proxy model's output from N × l Transform into N s × l The detailed process can be found in the following document. Figure 5 Then evaluate the IDR time history derived from the proxy model. With finite element analysis results The correlation coefficient is shown in equation (2).

[0042] (2) In the formula, cov and σ These are the covariance and standard deviation, respectively.

[0043] Step 6: See Figure 6 and Figure 7 Based on the IDR thresholds corresponding to various damage states of the building and the peak IDR output by the surrogate model, the damage state of each floor of the building in the test set (basically intact ND, slightly damaged SD, moderately damaged MD; severely damaged ED; completely destroyed CD) is evaluated, and the damage state of the most severely damaged floor is taken as the overall damage state of the building. Then, the confusion matrix between the simulation results of the surrogate model and the finite element analysis results is calculated to evaluate the accuracy of earthquake damage monitoring. The confusion matrix uses three indicators, Accuracy, Recall and Precision, to evaluate the accuracy of earthquake damage monitoring. The calculation method is shown in equations (3)-(5): (3) (4) (5) TP: Positive class and classified as positive; FN: Positive class but classified as false; FP: False class but classified as positive; TN: False class and classified as false.

[0044] The present invention also proposes an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the building earthquake damage monitoring method based on deep learning and sparse monitoring data.

[0045] The present invention also proposes a computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the steps of the building earthquake damage monitoring method based on deep learning and sparse monitoring data.

[0046] The memory in this application embodiment can be volatile memory or non-volatile memory, or it can include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct rambus RAM (DR RAM). It should be noted that the memory used in the methods described in this invention is intended to include, but is not limited to, these and any other suitable types of memory.

[0047] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., high-density digital video discs (DVDs)), or semiconductor media (e.g., solid-state disks (SSDs)).

[0048] In implementation, each step of the above method can be completed by integrated logic circuits in the processor's hardware or by instructions in software. The steps of the method disclosed in the embodiments of this application can be directly implemented by a hardware processor, or by a combination of hardware and software modules in the processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, detailed descriptions are omitted here.

[0049] It should be noted that the processor in the embodiments of this application can be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method embodiments can be completed by the integrated logic circuitry in the processor's hardware or by instructions in software form. The processor can be a general-purpose processor, 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 application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied as execution by a hardware decoding processor, or as a combination of hardware and software modules in the decoding processor. The software modules can be located in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory, and the processor reads the information in the memory and, in conjunction with its hardware, completes the steps of the above methods.

[0050] The above provides a detailed description of the building earthquake damage monitoring method based on deep learning and sparse monitoring data proposed in this invention. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A building seismic damage monitoring method based on deep learning and sparse monitoring data, characterized in that, The method comprises: Step 1, collecting measured or numerically simulated building seismic response data, processing the seismic response time history to unify the sampling frequency and duration of the data; Step 2, data preprocessing: based on spline interpolation, the response mode of each building along the height direction under each earthquake is modeled, and the data dimension of the seismic response of different floors is unified by equidistant sampling method; Step 3, taking building parameters, acceleration monitoring data of two floors and sensor position information as features, and taking the building seismic response time history to be reconstructed as the label, a data set is established, and is divided into training set, test set and validation set; Step 4, developing a feature extraction and fusion module and constructing a surrogate model, training the surrogate model through the training set, optimizing the network architecture and hyperparameters of the surrogate model based on the Optuna algorithm and the validation set, to construct the mapping relationship from features to labels; Step 5, input the features of the test set samples into the surrogate model to get the output, and then perform data post-processing: based on spline interpolation, the response mode of each sample along the height direction is modeled, and sampling is performed at the height of each floor, thereby deriving the seismic response of all floors, and then evaluating the reconstruction accuracy of the seismic response; Step 6, according to the building damage criterion and the reconstructed seismic response, the damage state of the building is evaluated, and compared with the actual damage state of the building, the accuracy of the seismic damage monitoring is evaluated.

2. The method of claim 1, wherein, In step 1, the specific scheme for unifying the sampling frequency and duration of the data includes: (1) resample at a frequency of 50 Hz; (2) if the duration is less than 50 seconds, pad zeros at the end of the data to make the duration reach 50 seconds; (3) if the duration exceeds 50 seconds and the peak ground acceleration occurs in the first / last 25 seconds, only keep the first / last 50 seconds of the data; (4) in other cases, keep the data 25 seconds before and after the time when the peak ground acceleration occurs.

3. The method of claim 1, wherein, The step 2 specifically comprises: inter-story drift angle at each time step , the building floor number x 1 and the seismic response of all floors of the building y 1 are taken as the independent variable and dependent variable respectively, and the spline interpolation fitting function is y = f 1( x );the number of sampling points is determined according to the maximum number of floors of the building in the database N m , N , N take 2 / 3 of the value N m , and the value of N is rounded off when it is a decimal number; the sampling sequence x 2 is generated according to formula 1, and f 1( x ) is sampled to obtain at each time step, so as to convert the data dimension of the seismic response from N s × l to N × l , N s and l are the number of floors of the building and the length of the IDR time history respectively; （1）。 4. The method of claim 1, wherein, In step 3, the building parameters include the number of floors, floor height, fundamental period, yield displacement, yield shear, peak displacement, peak shear, ultimate displacement, ultimate shear and stress pinch coefficient of the building analysis model; the monitoring data are the acceleration response time history of any two floors of the building; the sensor position information is the ratio of its installation height to the total height of the building.

5. The method of claim 1, wherein, In step 3, the data set is randomly divided into training set, test set and validation set in the ratio of 8:1:1, which are respectively used for training, optimization and precision evaluation of the surrogate model.

6. The method of claim 1, wherein, In step 4, based on the multi-scale convolutional neural network, a dynamic feature extraction module is constructed to fully extract the features of the monitoring data, based on the multilayer perception mechanism, a static feature extraction module is constructed to map the building parameters and sensor position information to a predetermined dimensional feature space, based on the attention enhanced long short-term memory network, a feature fusion module is developed to fully fuse dynamic and static features, and the above modules are combined to construct the surrogate model.

7. The method of claim 1, wherein, In step 5, the features of the test set samples are input into the surrogate model to obtain the output, which is then post-processed to obtain the IDR time history of all floors of the building, the specific steps are as follows: (1) for each time of the surrogate model output , respectively, the sampling sequence x 2 in step 2 and the output of the surrogate model are taken as the independent variable and the dependent variable, and the spline interpolation fitting function is y = f 2( x ); (2) taking the building floor number x 1 as a new sampling sequence, sampling on f 2( x ) step by step to obtain the inter-story drift angle of all floors at each time, so as to convert the data dimension of the surrogate model output result from N × l to N s × l ; then evaluate the correlation coefficient of the IDR time history derived by the surrogate model and the finite element analysis result , and the calculation formula of the correlation coefficient is: （2） where cov and σ are the covariance and standard deviation, respectively.

8. The method of claim 1, wherein, In step 6, the damage states of each floor of the test set building are evaluated according to the IDR threshold corresponding to each damage state of the building and the IDR peak value output by the surrogate model, and the damage states include the following: ND (nearby damage), SD (slight damage), MD (moderate damage), ED (extensive damage) and CD (collapse); the damage state of the most severely damaged floor is taken as the overall damage state of the building, and then the confusion matrix between the simulation results of the surrogate model and the finite element analysis results is calculated to evaluate the accuracy of the seismic damage monitoring. 9.An electronic device comprising a memory and a processor, the memory storing a computer program, wherein, The processor implements the steps of the method of any one of claims 1-8 when executing the computer program.

10. A computer readable storage medium for storing computer instructions, characterized in that, The computer instructions implement the steps of the method of any one of claims 1-8 when executed by the processor.

Images