Seismic Optimization Design Of RC Frame Structures Based On Machine Learning

It is difficult to find the complex nonlinear relationship between the structural design parameters and the structural seismic performance indexes by obtaining the structural seismic performance indexes(such as the maximum structural displacement Angle between floors,etc.)through limited tests or finite element simulation.However,when the traditional method combining finite element simulation and optimization method is used for seismic optimization design of structure,the problems such as complex calculation process,low efficiency and unstable iteration result are often caused by large model scale and repeated iteration.In recent years,intelligent algorithm has become a powerful tool to study the seismic performance of structures.In this paper,a large number of 3D models of RC frame structures are established using Open Sees to analyze the factors affecting the seismic performance of structures under strong earthquakes.Then,based on the simulation data,machine learning is used to predict the seismic performance index of the structure through the structural design parameters.Then,based on NGBoost prediction model and NSGA-Ⅱalgorithm,the multi-objective optimization preliminary design of RC frame structure is carried out,so that the preliminary optimized design scheme can be given quickly before the establishment of finite element model,which greatly reduces the workload of the subsequent optimization design.The specific research contents and conclusions are as follows:(1)In this paper,considering the errors of material parameters in the actual construction process of RC frame structure,a 3D model of RC frame structure with 3000 different design parameters is established through Open Sees finite element program,and then 10 three-way ground motions are reasonably selected as the external excitation of the structure according to the specification.The factors influencing the seismic performance of the structure are studied by time history analysis.The results show that the maximum interstory displacement Angle of the structure increases with the increase of the number of layers,the storey height and the span of the structure,thus reducing the seismic performance of the structure,and the increase of the axial compression ratio makes the maximum interstory displacement Angle of the structure decrease first and then increase.Subsequently,this paper makes a further analysis of the reasons for this result.(2)Based on the large amount of data obtained from finite element simulation,the prediction model of seismic performance of RC frame structures is established by using RF,XGBoost,NGBoost and SVM,which are four machine learning algorithms,which are an efficient data analysis and learning tool.The results show that all the four prediction models have better performance,and NGBoost model has the best prediction effect,with R~2 values of0.999 and 0.980 in the training set and test set,respectively.(3)Based on the NGBoost model with the best seismic performance prediction,combined with the non-dominated sorting genetic algorithm NSGA-Ⅱ,a multi-objective optimization framework for RC frame structure was established.It can give the structure design optimization method with the height,span and material strength as the design variables,other design parameters are certain,can meet the requirements of the design code as the constraint conditions,make the structure has better seismic performance and reduce the cost.This method can quickly give the preliminary optimized design scheme before establishing the finite element model of RC frame structure,thus greatly reducing the workload of the subsequent optimization design.