Predicting wind-induced structural response with LSTM in transmission tower-line system

Abstract

Wind-induced dynamic response of the nonlinear structure is critical for the structural safety and reliability. The traditional approaches for this response including observation or simulation focus on the structural health monitoring, the experiment, or finite element model development. However, all these approaches require high cost or computational investment. This paper proposes to predict the wind-induced dynamic response of the nonlinear structure with a novel deep learning approach, LSTM, and applies this in a structural lifeline system, the transmission tower-line system. By constructing the optimized LSTM architectures, the proposed method applies to both the linear structure, the single transmission tower and the nonlinear structure, the transmission tower-line system, with promising results for the dynamic and extreme response prediction. It can conclude that the layers and the hidden units have a strong impact on the LSTM prediction performance, and with proper training data set, the computational time can significantly decrease. A comparison surrogate model developed by CNN is also utilized to demonstrate the robustness of the LSTM-based surrogate model with limited data scale.