Super-large-scale building group earthquake damage assessment method based on multi-batch parallel computing

Abstract

The invention discloses a super-large-scale building group earthquake damage assessment method based on multi-batch parallel computing, relates to the technical field of earthquake damage response prediction, and aims to solve the problem of low earthquake response prediction efficiency caused by the problems of slow information reading, large storage occupation and the like of existing super-large samples (such as million-level, ten million-level and even hundreds of million-level samples). Through multi-batch block parallel computing and computer resource load-storage collaborative optimization, earthquake response calculation and earthquake damage evaluation of the super-large-scale building are realized, task allocation can be automatically performed according to the availability of computing resources, the information reading speed is improved, resource idle or overload is avoided, and the calculation efficiency of the super-large-scale building is improved. The problem of large storage occupation is solved, and then the seismic response prediction efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of earthquake disaster response prediction, and particularly to a method for earthquake disaster assessment of ultra-large-scale building groups based on multi-batch parallel computing. Background Art

[0002] In recent decades, earthquakes have caused heavy casualties and property losses. As the main place for people's production and living, cities have concentrated a large amount of population and wealth. Once a major earthquake occurs, it will have a significant impact on the city. Therefore, urban disaster prevention and mitigation is of vital importance. The rapid calculation and intelligent simulation of the earthquake response of urban building groups can provide important support for pre-disaster urban disaster prevention and mitigation planning, post-disaster emergency rescue and damage assessment.

[0003] For the seismic response analysis of ultra-large-scale building groups, there are currently mainly vulnerability analysis methods, capacity demand spectrum methods, and finite element time history analysis methods. The vulnerability analysis method uses one or two indicators to describe ground motion, making the accuracy of this method very limited. The capacity demand spectrum method uses a capacity spectrum to describe the seismic capacity of a structure and a demand spectrum to characterize the damage ability of ground motion. However, it is difficult to consider the time-domain characteristics of ground motion and damage accumulation in the demand spectrum, which limits its application. The time history analysis method can reflect the characteristics of the structure, the time-domain characteristics of ground motion, and the characteristics of damage accumulation. However, due to problems such as the large number of building structures in the building group, difficulty in obtaining detailed building information, difficulty in high-fidelity finite element modeling, and time-consuming refined structural response analysis, this method is difficult to be widely used in actual processes. To solve the above problems, it is currently a common practice to use a simplified multi-degree-of-freedom layer model (MDOF) to calculate the structural response of regional building groups. However, for ultra-large-scale building groups, using a multi-degree-of-freedom multi-layer model for response calculation is still time-consuming, and a more efficient calculation strategy is urgently needed. In recent years, machine learning or deep learning has been widely used due to its excellent data fitting ability. A large number of scholars have used machine learning or deep learning methods such as support vector machines, Kriging models, and neural networks to conduct research on the intelligent simulation of the structural response of building groups. For building groups on a large scale, due to differences in structural types, number of floors, plane layouts, construction years, etc., as well as the variability of ground motion, training machine learning models or deep learning models requires a large number of samples.For example, Chulyoung Kang et al. conducted 56,962,000 time-history response analyses in the linear elastic stage for training a deep neural network model (Kang C, Kim T, Kwon OS, Song J. Deep neural network-based regional seismic loss assessment considering correlation between EDP residuals of building structures. Earthquake Engineering & Structural Dynamics. 2023;52:3414-34.); Zekun Xu et al. used 10,563,732 earthquake response samples (2788 structures and 3789 ground motions) to train a neural network model for rapid regional structural response prediction (Zekun X, Jun C, Jiaxu S, Mengjie X. Regional-scale nonlinear structural seismic response prediction by neural network. Engineering Failure Analysis. 2023;154:107707.); Guoqing Zhang et al. selected 19,643,000 earthquake response samples (38,000 structures and 1,499 ground motions) to train XGBoost models, random forest models, and neural network models for structural parameter inversion (Guoqing Z, Kun L, Weiping W, Changhai Z, Chenyu Z, Bochang Z. Rapid seismic damage assessment of building portfolio based on fusion of surrogate model and monitored data. International Journal of Disaster Risk Reduction. 2025;119:105293.). For training machine learning models or deep learning models, the rapid calculation of preliminary samples remains a major challenge. Therefore, when using simplified multi-degree-of-freedom layer models or machine learning-based artificial intelligence simulation methods for earthquake damage assessment of building groups, the rapid calculation of ultra-large samples is the key problem to be solved.

[0004] At present, some studies have adopted the method of parallel computing to further realize the calculation of large-scale structural responses. Han Bo et al. analyzed the seismic responses of a medium-sized city (about more than 7,000 structures) using a single GPU (Han Bo, Lu Xinzheng, Xu Zhen, Li Yi. Seismic damage simulation of urban building groups based on high-performance GPU computing [J]. Journal of Natural Disasters, 2012, 21(5): 16-22.); Xu Zhen et al. carried out research on the rapid calculation of the responses of urban-scale building groups based on a GPU cluster and analyzed the seismic responses of 100,000 buildings (Xu Zhen, Sun Taowen, Yuan Jingyu. Computational acceleration method for seismic damage analysis of general urban building groups based on GPU cluster [P]. Beijing: CN201710740626.X, 2018-01-16.). The above studies targeted relatively small-scale building samples. However, for ultra-large samples (such as samples in the millions, tens of millions, or even hundreds of millions), due to problems such as slow information reading and large storage occupancy, the efficiency of seismic response prediction is low. Summary of the Invention

[0005] The object of the present invention is to provide a seismic damage assessment method for ultra-large-scale building groups based on multi-batch parallel computing, aiming at the problem of low seismic response prediction efficiency caused by problems such as slow information reading and large storage occupancy for existing ultra-large samples (such as samples in the millions, tens of millions, or even hundreds of millions).

[0006] The technical solution adopted by the present invention to solve the above technical problems is as follows:

[0007] A seismic damage assessment method for ultra-large-scale building groups based on multi-batch parallel computing includes the following steps:

[0008] Input the urban building structure information and ground motion information to be evaluated into a trained neural network to obtain the output damage state;

[0009] The trained neural network is obtained through the following steps:

[0010] Step 1: Obtain urban building structure information;

[0011] Step 2: Complete the data of urban building structure information;

[0012] Step 3: Obtain ground motion data;

[0013] Step 4: Combine the urban building structure information and ground motion data to obtain a calculation sample;

[0014] Step 5: Use the calculation sample to perform response calculation to obtain the maximum inter-story drift angle and maximum floor acceleration of the urban building structure;

[0015] The specific steps of Step 5 are as follows:

[0016] Step 5-1: Store the calculation samples in the calculation sample file parameters, and read the number of CPU cores of the computer. Then, determine the range of the number of samples in each batch according to the number of CPU cores. The range of the number of samples is the number of CPU cores * a set value, and the value range of the set value is 0.1×10^4 to 5×10^4;

[0017] Step 5-2: According to the range of the number of samples, evenly split the calculation sample file parameters into m batches, namely parameters_1 to parameters_m;

[0018] Step 5-3: According to the number of CPU cores, evenly split each batch of calculation sample files parameters_i into n sample blocks Block, where n is the number of CPU cores, i is the batch number, and i = 1, 2..., m;

[0019] Step 5-4: Generate a structural automatic modeling and calculation program file Calculate_i for the sample blocks belonging to the same batch;

[0020] Step 5-5: Create a thread pool through the ThreadPoolExecutor multithreading library. The thread pool contains n threads. At the same time, allocate a sample block to each thread in the thread pool, and allocate the corresponding program file Calculate_i to the thread i belonging to the same thread pool. Then, all threads perform response calculations synchronously through the corresponding program files to obtain the maximum inter-story drift angle and the maximum floor acceleration of the urban building structure. The process of each thread performing response calculations is as follows:

[0021] Thread i starts the external OpenSees program through subprocess. Thread i then enters a blocked waiting state. The external OpenSees program runs the calculation program file Calculate_i, and then calls CPU core i to perform structural response calculations;

[0022] The specific structural automatic modeling in the structural automatic modeling and calculation program file Calculate_i is as follows:

[0023] A: Determine the type of mechanical model according to the building type and building height;

[0024] B: According to the building type, building function, contour information, and construction age in the urban building structure information, and in combination with the building structure load-related specifications corresponding to the construction age, determine the floor mass corresponding to the mechanical model;

[0025] C: Determine the inter-story force-displacement relationship of the mechanical model according to the seismic code corresponding to the building type and construction year. The inter-story force-displacement relationship is represented by a backbone curve model and a hysteretic model. The backbone curve model is a trilinear backbone curve model, and the hysteretic model is a single-parameter hysteretic model;

[0026] D: According to the model type determined in A and combined with the story mass and the inter-story force-displacement relationship, establish a mechanical model accordingly;

[0027] Step 6: Use the maximum inter-story drift ratio and the maximum floor acceleration of the urban building structure to determine the damage states of the building structure and non-structural components.

[0028] Further, the urban building structure information includes geographical coordinates, outline, height, number of floors, building type, construction year, and building function information.

[0029] Further, the urban building structure information in Step 1 is obtained through geographic information systems, satellite remote sensing, unmanned aerial vehicle oblique photography, government open data, crowdsourcing data, Internet data, and POI commercial data services.

[0030] Further, the specific steps of Step 2 are as follows:

[0031] Obtain historical urban building structure information. Use the geographical coordinates, outline, height, number of floors, building type, and building function in the historical urban building structure information as inputs, and the construction year in the historical urban building structure information as the output. Train the model and use the trained model for data completion.

[0032] Further, the ground motion data includes ground motion number, amplitude modulation coefficient, sampling interval, and number of sampling points.

[0033] Further, the ground motion data is obtained by acquiring ground motion records through a seismic monitoring network and performing baseline correction and Butterworth filtering on the ground motion records.

[0034] Further, the ground motion data is obtained through the following steps:

[0035] Step 3-1: Obtain the site category and design earthquake grouping information of the target building complex, and according to the site category and design earthquake grouping information, obtain the characteristic period of the site soil;

[0036] Step 3-2: According to the characteristic period of the site soil, obtain the corresponding ground motion record in the ground motion database or the anti-collapse design standard for building structures T / CECS 392-2021;

[0037] Step 33: After performing ground motion amplitude adjustment, resampling, and data truncation processing on the ground motion records, ground motion data is obtained.

[0038] Further, the ground motion database is PEER, KiK-net, or CESMD.

[0039] Further, the specific method for determining the type of mechanical model according to the building type and building height is as follows:

[0040] For concrete frame structures, steel structures, masonry structures, and wood structures, a lumped mass story shear model is selected as the type of their mechanical model;

[0041] For concrete shear wall structures or frame-shear wall structures, a lumped mass flexure-shear coupling model is selected as the type of their mechanical model.

[0042] Further, the damage states of the building structure and non-structural components in Step 6 are determined by the inter-story drift angle of structural damage and the maximum floor acceleration limit values recommended by the FEMA2012d Multi-hazard loss estimation methodology HAZUS-MH 2.1 engineering building module specification.

[0043] The beneficial effects of the present invention are:

[0044] This application realizes the seismic response calculation and seismic damage assessment of ultra-large-scale buildings through multi-batch and block parallel computing and computer resource load-storage collaborative optimization. This application can automatically allocate tasks according to the availability of computing resources, improving the information reading speed, avoiding resource idleness or overload, solving the problem of large storage occupancy, and thus improving the efficiency of seismic response prediction.

[0045] In addition, this application can also handle the seismic response prediction and seismic damage assessment of structures under different design codes corresponding to different structural types (such as concrete frame structures, steel frame structures, fortified masonry structures, unfortified masonry structures, shear wall structures, wood structures, etc.) and different construction years. Description of the Drawings

[0046] Figure 1 is the overall flow chart of this application;

[0047] Figure 2 is the schematic diagram of CPU block multi-core parallel computing;

[0048] Figure 3 is the schematic diagram of a multi-degree-of-freedom story model. Detailed Embodiments

[0049] It should be specifically noted that, without conflict, the various embodiments disclosed in this application can be combined with each other.

[0050] Specific Embodiment 1: A method for earthquake damage assessment of ultra-large building complexes based on multi-batch parallel computing described in this embodiment includes the following steps:

[0051] Step 1. This step is for data collection and preprocessing. This step mainly involves the collection and processing of urban building information, the collection and processing of ground motion data, and the merging of structural ground motion information, etc. First, the basic information of the urban building complex is obtained through means such as satellite remote sensing, UAV aerial photography, and GIS, mainly including the geographical coordinate information, contour information, construction age, building function, etc. of the buildings. Then, the obtained data is completed, and the data that is unreasonable is corrected. After that, the ground motion data is collected from the seismic network or ground motion database and undergoes baseline correction and filtering processing. Finally, the obtained building structure information and ground motion information are merged to obtain a calculation sample file. Each line in the calculation sample file represents a combination of a structure and a ground motion information. For the convenience of subsequent task allocation and response storage, each sample in the calculation sample file has a unique sample number.

[0052] Step 2. This step is for the parallel computing of structural responses. This step includes data deployment in the parallel computing stage, multi-core parallel computing, the processing and storage of structural response data, etc.

[0053] Data deployment in the parallel computing stage: The main program evenly divides the calculation sample file into several batches, and the number of calculation samples in each batch should not be too many or too few (recommended: 0.1 * 10,000 to 50,000 times the number of CPU cores). Too many samples in each batch will increase the later indexing time, and too few samples in each batch will make the number of batches too large, thus increasing the storage and I / O operation time. Then, according to the available number of CPU cores, each batch of calculation samples is evenly divided into several sample blocks (Blocks).

[0054] Multi-core parallel computing: According to the available number of CPU cores, the main program creates a thread pool through ThreadPoolExecutor, and assigns a sample block (Block) and the corresponding OpenSees response calculation program to each thread. Then each thread starts an external OpenSees process through subprocess, and then the thread enters a blocked waiting state. Each external OpenSees process calls a CPU core for response calculation. After the response calculation is completed, the OpenSees process automatically saves the detailed seismic response of the structure to the result folder and then ends the process. Each CPU core calculates a sample block (Block), making each core of the CPU load balanced.

[0055] Processing and saving of structural response data: After the calculation of all samples in a batch is completed, the key information (the maximum inter-story drift angle and the maximum floor acceleration) in all response files is extracted, and all the key information of a batch is saved to a response file, and the structural seismic responses automatically saved by the OpenSees process are deleted, so as to optimize the storage. Next, parallel calculation, response extraction and saving of samples in the next batch are carried out until the calculation of all batches of samples is completed.

[0056] Step 3. This step is to perform structural damage judgment, earthquake damage assessment, and establishment of a machine learning model based on the structural responses calculated in the previous step. First, the maximum inter-story drift angle of each floor of the structure and the maximum floor acceleration are extracted as needed, and then the damage conditions of structural and non-structural components are determined according to the damage limits recommended by the FEMA code, and further earthquake damage assessment is carried out. After that, a machine learning model is established through structural information, ground motion information, and the obtained structural responses or structural damage information.

[0057] The overall flow chart of this application is as Figure 1 shown. The specific implementation steps are as follows:

[0058] Step 1: The main content of this step is the acquisition and preprocessing of calculation samples, and its schematic diagram is as Figure 1 shown in the first part of. Specifically, it includes the collection and processing of urban building information, the collection and processing of ground motion data, and the merging of structural ground motion information. The detailed content of Step 1 is as follows:

[0059] Step 1.1: Obtain the building structure information of major cities across the country through means such as Geographic Information System (GIS), satellite remote sensing, UAV oblique photography, government open data, crowdsourcing data and Internet data, and POI commercial data services. The structural information includes geographical coordinates, outline, height, number of floors, building type, construction year, and building function information.

[0060] Step 1.2: Perform data correction. For data whose structural length and width do not conform to the 0.3m modulus, make it into a modulus of 0.3m; use means such as machine learning and deep learning for data completion. For example, the construction year of some structures is missing,

[0061] Take geographical coordinates, outline, height, number of floors, building type, and building function in historical data as inputs, and the construction year as the output, train the model, and use the trained model for data completion,

[0062] Step 1.3: When the response calculation of the ultra-large-scale building complex serves the post-earthquake damage assessment, the ground motion records that have occurred are obtained through the seismic monitoring network. Then, baseline correction and Butterworth filtering are performed on the ground motion data.

[0063] Step 1.4: When the response calculation of the ultra-large-scale building complex serves the pre-earthquake damage prediction or serves the training of machine learning models and deep learning models, the characteristic period of the site soil is determined according to information such as the site category and design earthquake grouping of the target building complex. Then, according to the characteristic period of the site soil, the corresponding ground motion records are obtained from existing ground motion databases such as the PEER, KiK-net, CESMD databases or the "Code for Anti-collapse Design of Building Structures T / CECS 392-2021". After that, preprocessing such as ground motion amplitude adjustment, resampling, and data truncation is performed to obtain the ground motion data.

[0064] The ground motion data includes the ground motion number

[0065] Step 1.5: The structural information and ground motion information are merged to generate calculation samples. Each calculation sample contains the basic information of the structure such as geographical coordinates, outline, height, number of floors, building type, construction age, building function, etc., and the ground motion information such as ground motion number, amplitude adjustment coefficient, sampling interval, number of sampling points, etc. All calculation samples are stored in the calculation sample file parameters.txt to prepare for subsequent parallel calculation.

[0066] Step 2: The main content of this step is the parallel calculation of structural response, and its schematic diagram is as Figure 2 shown. Specifically, it includes data deployment, CPU multi-core parallel calculation, processing and saving of structural response. The detailed content of Step 2 is as follows:

[0067] Step 2.1: The main program first automatically reads the number of CPU cores of the computer, and then installs a reasonable range of the number of samples per batch (recommended: 0.1 million to 50,000 times the number of CPU cores), and evenly splits the calculation sample file into m batches of parameters_1.txt to parameters_m.txt. Secondly, according to the number of CPU cores used, each batch of calculation sample files parameters_i.txt is evenly split into n sample blocks (Block), where n is the number of CPU cores, so that one CPU core corresponds to one sample block (Block). Since each sample block (Block) in a batch has the same size and similar calculation amount, CPU load balancing is achieved.

[0068] Step 2.2: Generate a structural automatic modeling and calculation program file Calculate_i.tcl for the sample block Block_i belonging to the same batch.

[0069] Step 2.3: The main program creates a thread pool through the ThreadPoolExecutor multithreading library. The thread pool contains n threads. At the same time, assign a sample block Block_i and a calculation program file Calculate_i.tcl to each thread i. Then thread i starts the external OpenSees program through subprocess, and thread i then enters a blocked waiting state. The external OpenSees program runs the calculation program file Calculate_i.tcl and then calls CPU core i to perform structural response calculations. Other threads also proceed synchronously, so there are n external OpenSees programs performing response calculations simultaneously. After each external OpenSees program finishes the calculation, the time history response and envelope values are automatically saved to the result file.

[0070] When calculating each sample, the program automatically establishes a simplified numerical model of the structure according to the read structural information. The modeling logic is as follows:

[0071] ①: Select an appropriate simplified model as its mechanical model according to the building type and building height. Specifically, for concrete frame structures, steel structures, masonry structures, and wood structures, select the lumped mass story shear model as the type of its mechanical model; for concrete shear wall structures or frame-shear wall structures, select the lumped mass flexure-shear coupling model as the type of its mechanical model.

[0072] ②: Determine the story mass corresponding to the simplified mechanical model according to the building type, building function, contour information, and construction era, combined with the building structure load-related specifications corresponding to the construction era.

[0073] ③: Estimate the story force-displacement relationship (including the backbone curve model and the hysteretic model) corresponding to the simplified mechanical model according to the building type and the seismic code corresponding to the construction era; common backbone curve models include the bilinear and the trilinear model recommended by the HAZUS report (FEMA, 2012). Since the trilinear model is more accurate and the computational effort is not large, the trilinear backbone curve model is selected here. The story hysteretic model affects the inelastic energy dissipation of the structure. Here, a relatively simple single-parameter hysteretic model is selected, which only requires one parameter. This hysteretic model was proposed by Steelman and Hajja. The trilinear backbone curve model and the story hysteretic model are as Figure 3 shown.

[0074] Establish a mechanical model according to the model type selected in 1, combined with the parameter story mass and the story force-displacement relationship.

[0075] Step 2.4: After the calculation of a batch of samples is completed, extract important data such as the maximum inter-story drift ratio and the maximum floor acceleration from the structural responses according to requirements, delete the useless information, and save all the responses of a batch together to achieve storage optimization.

[0076] Step 2.5: Loop through Steps 2.2 to 2.5 to calculate the next batch of samples in sequence and process and store the structural response data.

[0077] Step 3: The main content of this step is the discrimination of the structural damage level and the establishment of subsequent machine learning models. It specifically includes data deployment, CPU multi-core parallel computing, the processing and saving of structural responses. The damage state of each floor of the structure and the overall damage state are judged by the maximum inter-story drift ratio of each floor of the structure. The damage state of non-structural components is judged by the maximum floor acceleration of each floor of the structure. The damage states of the structure and non-structural components can be determined by the structural damage inter-story drift ratio and the maximum floor acceleration limits recommended by the specification (FEMA2012d Multi-hazard loss estimation methodology HAZUS-MH2.1 engineering building module). Then, a machine learning model or a deep learning model is established based on the structural information, ground motion information, and the obtained structural responses, damage states, etc.

[0078] Step 4: This step is a specific application case of this application. To train the machine learning XGBoost model for rapid calculation of structural responses, 6,500 structural samples were selected through Latin hypercube sampling. In addition, 3,022 ground motion samples were selected, and 19,643,000 calculation samples were generated by matching the structures and ground motions. To achieve rapid calculation of earthquake responses on a scale of tens of millions, the parallel computing idea of the present invention was adopted for parallel computing on the computing platform (CPU: Intel(R) Xeon(R) Gold 6462C), and the computing time was 0.8 days. When parallel computing was not used, it took 41.29 days to complete the calculation. On the computing platform (CPU: Intel(R) Xeon(R) Gold 6462C), compared with the conventional calculation method, the computing efficiency of this application was increased by more than 50 times. It should be noted that in actual use, the computing efficiency depends on the number of CPU cores of the computing platform. The above application case proves the feasibility of this application for rapid calculation on ultra-large-scale building groups. The samples obtained by parallel computing were applied to the establishment of the machine learning XGBoost model, and the corresponding results have been published in the journal International Journal of Disaster Risk Reduction in the field of earth sciences (Guoqing Z, Kun L, Weiping W, Changhai Z, Chenyu Z, Bochang Z. Rapid seismic damage assessment of building portfolio based on fusion of surrogate model and monitored data. International Journal of Disaster Risk Reduction. 2025;119:105293.).

[0079] Compared with the existing structural response calculation methods, on the basis of considering different structural types, this application can also consider the differences in design codes and performance degradation under different construction years. The application scenario of this application is wider and the applicability is stronger, and it can realize the response calculation of building groups under multiple scenarios.

[0080] Compared with the existing methods for calculating the response of building groups through parallel computing, the present application avoids the problem of retrieval time consumption in the later-stage calculation of ultra-large batches of samples by performing multi-batch calculations on the calculation samples, and improves the fault tolerance ability. The present application adopts block parallel computing, automatically and evenly distributes tasks to all CPU cores according to the number of CPU cores, realizes load balancing, avoids some CPU cores from being idle, and improves the overall efficiency. In the response data processing stage, the present application extracts the key response data of the structure, avoids the memory occupation of invalid data, and optimizes the storage. Thus, the structural response calculation of ultra-large building groups is efficiently and smoothly realized.

[0081] It should be noted that the specific implementation manners are only explanations and illustrations of the technical solutions of the present invention, and the scope of the right protection cannot be limited thereby. Those that are only partial changes made according to the claims and the specification of the present invention should still fall within the protection scope of the present invention.

Claims

1. A method for earthquake damage assessment of super-large-scale building complexes based on multi-batch parallel computing, characterized in that The following steps are involved: Input the urban building structure information and earthquake information to be evaluated into the trained neural network to obtain the output damage status; The trained neural network is obtained by the following steps: Step 1: Obtain urban building structure information; Step 2: Complete the data of urban building structure information; Step 3: Obtaining ground motion data; Step 4: Combine urban building structure information and seismic data to obtain calculation samples; Step 5: Use the calculation sample to perform response calculation to obtain the maximum inter-story displacement angle and maximum floor acceleration of the urban building structure; The specific steps of step five are: Step 51: Store the calculation samples in the calculation sample file parameters, and read the number of CPU cores of the computer, and then determine the range of the number of samples in each batch according to the number of CPU cores. The range of the number of samples is the number of CPU cores*the set value, and the range of the set value is 10,000 to 50,000; Step 52: According to the range of sample numbers, the calculation sample file parameters is evenly split into m batches, namely parameters_1 to parameters_m; Step 53: According to the number of CPU cores, each batch of calculation sample files parameters_i is evenly split into n sample blocks Block, where n is the number of CPU cores and i is the batch number, i = 1, 2..., m; Step 54: Generate a structure automatic modeling and calculation program file Calculate_i for the sample blocks belonging to the same batch; Step 55: Create a thread pool through the ThreadPoolExecutor multi-thread library. The thread pool contains n threads. At the same time, a sample block is allocated to each thread in the thread pool, and the corresponding program file Calculate_i is allocated to thread i belonging to the same thread pool. Then all threads synchronously perform response calculations through the corresponding program files to obtain the maximum inter-story displacement angle and maximum floor acceleration of the urban building structure. The process of response calculation for each thread is: Thread i starts the external OpenSees program through subprocess, and then thread i enters the blocked waiting state. The external OpenSees program runs the calculation program file Calculate_i, and then calls CPU core i to perform structural response calculation; The automatic structural modeling and the automatic structural modeling in the calculation program file Calculate_i are specifically as follows: A: Determine the type of mechanical model based on the building type and building height; B: According to the building type, building function, outline information and construction year in the urban building structure information, and combined with the relevant specifications of the building structure load corresponding to the construction year, determine the layer quality corresponding to the mechanical model; C: According to the seismic code corresponding to the building type and the construction year, the inter-layer force-displacement relationship of the mechanical model is determined. The inter-layer force-displacement relationship is represented by a skeleton line model and a hysteresis model. The skeleton line model is a trilinear skeleton line model, and the hysteresis model is a single-parameter hysteresis model. D: Based on the model type determined in A, the mechanical model is established by combining the layer quality and the interlayer force-displacement relationship; Step 6: Use the maximum inter-story drift angle and maximum floor acceleration of urban building structures to determine the damage state of building structures and non-structural components.

2. The method for earthquake damage assessment of ultra-large-scale buildings based on multi-batch parallel computing according to claim 1 is characterized in that The urban building structure information includes geographic coordinates, outline, height, number of floors, building type, construction year and building function information.

3. The method for earthquake damage assessment of ultra-large-scale building complexes based on multi-batch parallel computing according to claim 2 is characterized in that The urban building structure information in step 1 is obtained through geographic information systems, satellite remote sensing, drone oblique photography, government open data, crowdsourcing data, Internet data and POI commercial data services.

4. The method for earthquake damage assessment of ultra-large-scale building complexes based on multi-batch parallel computing according to claim 3 is characterized in that The specific steps of step 2 are: Obtain historical urban building structure information, take the geographic coordinates, outline, height, number of floors, building type and building function in the historical urban building structure information as input, and the construction year in the historical urban building structure information as output, train the model, and use the trained model to complete the data.

5. The method for earthquake damage assessment of ultra-large-scale building complexes based on multi-batch parallel computing according to claim 4 is characterized in that The seismic data includes a seismic number, an amplitude modulation coefficient, a sampling interval, and a number of sampling points.

6. The method for earthquake damage assessment of ultra-large-scale buildings based on multi-batch parallel computing according to claim 5 is characterized in that The seismic data is obtained by obtaining seismic records through a seismic monitoring network and performing baseline correction and Butterworth filtering on the seismic records.

7. The method for earthquake damage assessment of ultra-large-scale buildings based on multi-batch parallel computing according to claim 5 is characterized in that The earthquake data is obtained by the following steps: Step 31: Obtain the site category and design earthquake grouping information of the target building complex, and obtain the characteristic period of the site soil according to the site category and design earthquake grouping information; Step 32: According to the characteristic period of the site soil, obtain the corresponding seismic records in the seismic database or the building structure anti-collapse design standard T / CECS 392-2021; Step 33: After performing seismic motion amplitude modulation, resampling and data truncation processing on the seismic motion record, the seismic motion data is obtained.

8. The method for earthquake damage assessment of ultra-large-scale buildings based on multi-batch parallel computing according to claim 7 is characterized in that The earthquake motion database is PEER, KiK-net or CESMD.

9. The method for earthquake damage assessment of ultra-large-scale buildings based on multi-batch parallel computing according to claim 8 is characterized in that The type of mechanical model is determined according to the building type and building height as follows: For concrete frame structures, steel structures, masonry structures and wood structures, the concentrated mass layer shear model is selected as the type of its mechanical model; For concrete shear wall structures or frame shear wall structures, select the concentrated mass bending-shear coupling model as the type of its mechanical model.

10. The method for earthquake damage assessment of ultra-large-scale building complexes based on multi-batch parallel computing according to claim 1, characterized in that The damage states of the building structure and non-structural components in step six are determined by the structural damage inter-story displacement angle and the maximum floor acceleration limit recommended by FEMA2012d Multi-hazadloss estimation methodology HAZUS-MH 2.1engineering building module specification.

Images