Probabilistic Seismic Demand Analysis of Pile-Supported Transmission Towers on Infinite Slopes: Exploring Machine Learning Models for Optimal Intensity Measures

Many transmission towers in mountainous areas are founded on soil slopes. In these cases, the reliability of towers in supporting electricity conductors under seismic events is affected largely by the response of the slopes and soil-structure interactions in addition to the structural performance of the towers. Probabilistic seismic demand models (PSDMs) of transmission towers to reliably associate key engineering demand parameters (EDPs) to intensity measures (IMs) must capture complexities present in the seismic performance of these systems. The performance of PSDMs also depends to a large degree on the choice IMs. Currently, there are no PSDMs for transmission towers on soil slopes and optimal IMs for such systems have not been identified. The objective of this study is to address these gaps. Toward this goal, a nonlinear finite-element model of a coupled tower-soil-foundation (CTSF) system is developed in the OpenSees platform. To capture the earthquake response and sliding displacement of soil slopes, a two-dimensional soil domain is modeled with pressure sensitive soil materials. Nonlinear soil-structure interactions are captured using nonlinear soil springs implemented at the soil-pile interface. Cloud analysis is performed and the developed CTSF system is subjected to a large set of ground motions. During these analyses, the responses of the system are recorded, and a large spectrum of IMs including structure- and non-structure-specific intensity measures are analyzed in terms of efficiency, practicality, and proficiency. In establishing the PSDMs, conventional regression as well as machine learning models are explored for associating EDPs to IMs. Results of this extensive IM analysis indicate that non-structure-specific acceleration-related IMs are more optimal compared to velocity-, displacement-, and time-related IMs.