融合批量数值仿真与机器学习的局部场地放大快速预测方法OA

Traditional numerical simulation methods based on single physical models generally suffer from limitations such as computational complexity and high resource consumption when analyzing complex site effects.To address these issues,this study takes a typical inverted trapezoidal sedimen-tary basin as the research object and integrates batch numerical simulations with machine learning tech-niques to develop an efficient prediction framework for local site amplification effects.First,3 312 standard finite element models of the sedimentary basin were constructed.Second,a large-scale data-base containing 11 types of input parameters and spectral responses(totaling 977 010 data entries)was built through systematic seismic response analysis.Third,three types of machine learning algo-rithms—convolutional neural networks,long short-term memory networks,and decision tree regres-sion—were employed to train intelligent prediction models of surface amplification effects.All data processing was automated using self-developed Python programs,with the dataset split into training and testing sets in an 8∶2 ratio to ensure model generalizability.Empirical analysis based on a typical profile of the Mygdonian Basin in Greece demonstrated that the proposed model rapidly and accurately assessed local site effects.Compared with traditional numerical simulation methods,this approach established direct mapping between inputs and outputs through machine learning algorithms,signifi-cantly reducing computational complexity.While maintaining engineering accuracy,the approach markedly improved computational efficiency,providing a reliable technical foundation for the rapid assessment of seismic responses at practical engineering sites.