基于差分进化-人工神经网络的沉积河谷地震动放大效应预测模型

探讨了基于差分进化-人工神经网络构建沉积河谷地震响应代理模型的可行性。首先建立沉积河谷对地震波散射的求解方法，以半圆形、V形沉积河谷为例，以入射波条件、沉积内外介质属性、场地形状为特征参数，以沉积河谷地震动放大系数为预测目标参数，构建数据集；其次，建立沉积河谷地震动放大效应人工神经网络、差分进化-人工神经网络算法预测模型，对比两种算法计算精度和稳定性，并进行了特征参数敏感性分析。结果表明:人工神经网络能较好地预测沉积河谷地震动放大效应，使差分进化-人工神经网络预测模型的精度和稳定性显著提高；入射波频率是影响沉积河谷地震动放大系数的主要原因，沉积内外介质密度比的影响较小。本研究结论可对地震作用下更为复杂的局部场地效应预测和评估提供参考。 

Prediction model of seismic amplification effect in sedimentary valley based on differential evolution-artificial neural network

Sedimentary valley has obvious amplification effect on ground motions, which has an increase on the engineering damage. However, the propagation mechanism of seismic wave in sedimentary valley is complex, resulting in high nonlinearity and high coupling of influences of incident wave and site parameters on seismic amplification effect. First, based on the boundary element method, the scattering of seismic waves by sedimentary valley is solved. Prediction models of seismic amplification effect of semicircular and V-shaped sedimentary valley are established, with incident wave conditions, material properties and valley shapes as characteristic parameters and the seismic amplification factor of sedimentary valley as target parameters, and the dataset is constructed; Second, the calculation accuracy and stability of artificial neural network （ANN） and its optimization algorithm, i.e., differential evolution, are compared, and the sensitivity of characteristic parameters is analyzed. The results show that the ANN can predict the amplification effect of sedimentary valley, and the accuracy and stability of differential evolution-ANN prediction model are significantly improved; The incident wave frequency is the main influence factor of the seismic amplification coefficient of sedimentary valley, and the density ratio of internal and external medium has little effect. The conclusions can provide references for more complex local site effect prediction and assessment.