Prediction framework of slope topographic amplification on seismic acceleration based on machine learning algorithms

Slope topographic amplification of peak seismic acceleration leads to more severe seismic damage to nearby buildings, so its quantitative prediction is required for engineering applications. However, prior quantitative studies are fewer and use traditional regression methods, which require subjective assumptions and have lower accuracy. To solve this problem, artificial intelligence regression algorithms were firstly attempted to establish a predictive model for slope topographic amplification. In this model, slope inclination, slope height, and frequency of seismic waves were taken as parameters, and amplification ratio was taken as prediction target. Then, the multivariate nonlinear relationship between the prediction target and the sample parameters was established using artificial intelligence regression algorithms. Compared with the traditional prediction method, the determination coefficient of the present model is improved by 17.84%–32.60%, and the root mean square error is reduced by 30.05%–77.36%. In addition, the effect of different regression algorithms on the prediction model was investigated, and the influence of each parameter on the topographic amplification was analyzed. Finally, the proposed prediction model was applied to three practical earthquake cases, which confirmed that it can fill the gap in quantitatively predicting the slope topographic amplification and provide a guide for seismic design in engineering.