Rapid Earthquake Damage Prediction Method Of Brick Masonry Buildings Based On Machine Learning

China is one of the earthquake-prone countries around the world,and the occurrence of earthquakes will cause incalculable casualties and economic losses.In recent years,China has also carried out a lot of basic work in earthquake prevention and mitigation,including the prediction of seismic damage to buildings.However,the traditional seismic damage prediction methods often require a lot of human and material resources to prepare basic data and carry out a lot of calculation and analysis.It is difficult to assess the seismic damage of buildings in a large area rapidly.With the rapid development of artificial intelligence technology,machine learning has been gradually introduced into the field of seismic damage assessment with the advantages of efficiency and accuracy.Previous earthquake scene shows that masonry structure is the structure with the largest number of damages and the most serious earthquake damage in urban and rural buildings in China.Therefore,it is of great importance to predict the seismic damage of masonry structures.To this end,this paper will take brick masonry houses as the research object and propose two types of fast prediction methods for earthquake damage with both efficiency and accuracy based on machine learning models.This research is not only important for earthquake prevention and mitigation,but also a realistic need.The main research of the paper is as follows:Firstly,from the factors mainly affecting the seismic damage of brick masonry houses,22 seismic damage factors related to structural characteristics,ground shaking characteristics and site characteristics were selected as input variables.143 ground motions were selected and amplitude-adjusted,and the amplitude-adjusted ground motions were used to carry out the nonlinear times time history analysis on 170 typical brick masonry houses.A dataset which contains a total of 75,820 sets of data were obtained.Three rapid prediction methods for brick masonry earthquake damage based on support vector machine(SVM),random forest(RF)and artificial neural network(ANN)models were established.Few metrics were used to evaluate the proposed three methods.Based the evaluation results,the RF model-based seismic damage rapid prediction method for brick masonry houses was selected as the optimal method,and the method was further validated by arithmetic examples.Secondly,the optimization and improvement of the RF model-based seismic damage rapid prediction method for brick masonry structures were carried out.Firstly,two parameter importance analysis methods were used to analyze the importance of 22 parameters,and the best parameter combinations with different numbers of input parameters under the two analysis methods were obtained respectively.The influence of the change of the number of input parameters on the RF model was analyzed.Then,the optimal number of input parameters was determined as 9 by considering the accuracy of RF model prediction and the complexity of the model.Finally,from the perspective of easy access to parameters,a set of optimal parameter combinations is determined.An optimized RF model-based fast prediction method for brick masonry house earthquake damage is re-established based on9 parameters and further validated by arithmetic examples.Thirdly,based on the data and results of the completed seismic damage prediction projects in Lu’an,Qinhuangdao and Tangshan cities,10 structural parameters and seismic intensity of the of 160 brick masonry houses were selected as input variables,and the corresponding seismic damage prediction results were used as output variables to establish a data set consisting of 800 sets of data.Based on the LM-BP neural network model,a novel rapid earthquake damage prediction method for brick masonry houses is established.To further validate the proposed novel method,40 samples outside thetraining sets and two actual brick masonry earthquake cases were used to further test the method.The results showed that the method can achieve the goal of fast and accurate prediction of brick masonry earthquake damage.Fourthly,two platforms are established to cover the above two types of methods for rapid prediction of earthquake damage in brick masonry houses,respectively.The platforms of both types of methods use the GUI function in MATLAB.The core programs were inserted into the platform to achieve the goal that the damage state of the building can be obtained by entering the information of relevant parameters in the input part of the interface.The platform provides a convenience for further promoting the application of the two types of methods.