Seismic Fragility Analysis of Wind Turbine-Reinforced Concrete Tower (WTRCT) Based on Machine Learning Method

WTRCTs are critical load-bearing components prone to extreme loads. The finite element-based seismic fragility analysis method is characterized by several limitations, including high computational costs, difficulties in accurately representing the complex features of ground motion with low-dimensional inputs, and uncertainties in ground motion intensity indices. Consequently, this study employs three machine learning (ML) methods, namely MLP, RF, and SVM, to conduct efficient and rapid seismic fragility analysis on WTRCTs. Class I and Class II intensity indexes are proposed. The multidimensional complex nonlinear learning of the seismic response law of WTRCTs is realized. The results show that the ML earthquake damage prediction model based on class I has higher prediction accuracy, with the minimum R2 of 0.9409 in the test set, while the corresponding value for the model based on class II is 0.8525. The identification of ground motion intensity indexes is realized by an iterative method, and the ability to interpret maximum horizontal displacement angle (θmax) by IPGV, IPGD, and IHI is better than that of other class I intensity indexes. Based on the cloud mapping method for seismic fragility curves, the RF model considering 3D inputs and the SVM model considering 8D inputs have significant similarities with FEA in terms of damage severity. The accuracy and feasibility of using ML methods for seismic fragility analysis are further verified.