Soil Seismic Response Prediction Model Method Based On Machine Learning Algorithms

The seismic response of the site has always been a concern of the earthquake engineering community,and it is an important basis for the seismic safety and economic evaluation of the engineering site.At present,the research on the seismic response of the site mainly relies on the method of numerical simulation,but the research shows that the existing numerical simulation method can not accurately simulate the seismic response of the site,especially the simulation of the soft soil site has large errors.Therefore,the use of new methods and new methods to study site seismic response is a new research direction,which can break through the limitations of traditional numerical simulation methods,improve the accuracy of seismic response analysis,and provide technical support for site seismic safety evaluation and structural seismic design.Based on the Ki K-net network database and existing research results,this paper selects more than 40 stations in the Ki K-net network of historical earthquake records and other relevant information to carry out a research on the site seismic response prediction model based on machine learning algorithms.The prediction model of soil deformation grade and the prediction model of surface peak acceleration under the action of earthquake load were established,corresponding analysis software was formed,a typical section was constructed,and the shear strain distribution law of the site was preliminarily studied.The main work and research results obtained in this paper are summarized as follows:1.Based on the Ki K-net strong earthquake network in Japan,more than 40 strong earthquake stations and historical seismic records of the stations were selected.Combined with the relevant information of the soil profile provided by the strong earthquake station,the site characteristic parameters that were easy to obtain and commonly used in engineering are selected.A data set related to the soil deformation grade and a data set related to the peak acceleration of the ground surface were established.2.Through quantitative analysis of the contribution of different site and ground motion characteristic parameters to the soil shear strain,6 influencing factors were selected,and they formed two characteristic combinations as machine learning parameters.Through the establishment of five-parameter earthquake load prediction model for soil shear strain levels,the differences in performance metrics of different prediction models were compared and analyzed.Random forest classification model was used to explore the deformation response of typical site sections,and the relationship between various site deformation degrees and PGA and buried depth was given.3.By selecting 6 site features that affect surface acceleration and input ground motion features as machine learning parameters,different surface peak acceleration prediction models were established,and the differences in performance metrics of different prediction models were compared and analyzed.From three different perspectives: overall data,site category,and ground motion intensity,several machine learning prediction models and numerical simulation results were compared and analyzed.The results show that the prediction accuracy of machine learning models is better than numerical simulation.4.Through user interface design and environment packaging by Using Py Qt5 and Qt Designer tools,the encapsulation of the prediction model of soil deformation level and the prediction model of surface peak acceleration proposed in this paper under the action of earthquake load was completed.The two softwares are easy to operate and have a friendly interface,which can be used by researchers and engineering technicians.