Machine learning-assisted seismic risk assessment of substation equipment considering material parameter uncertainty

This paper proposes a machine learning (ML)-assisted approach for seismic risk assessment of substation equipment, efficiently quantifying material parameter uncertainty. It first generates a few samples using simulation models and structural parameters of Latin hypercube sampling. A multi-input-multi-output ML model is then trained to predict response distributions based on intensity measures (IM) and structural parameters. With uncertainties, the model predicts responses’ means and standard deviations at each specified IM, thereby enabling fragility and risk derivation. Three equipment types are selected to validate the approach. Results show that the developed ML models accurately estimate structural responses, fragility, and risk across various equipment types. Sensitivity analyses reveal that the elastic modulus of the steel support is the most important material parameter. The material property dispersion of insulators has a greater impact on fragility and risk than that of steel supports. As the ML models avoid time-consuming time-history calculations for a vast number of scenarios caused by parameter dispersion, they achieve high efficiency in risk prediction, uncertainty quantification, and sensitivity analysis. Using structural features as inputs, the trained ML models are applicable to any similar equipment with varying parameters. Accordingly, a graphical user interface (GUI) is developed as a practical tool yet reliable for efficient seismic risk assessment of electrical equipment in substations. • A ML-assisted seismic risk assessment approach of substation equipment is proposed. • ML prediction of response distribution allows efficient uncertainty quantification. • Distribution- and risk-oriented importance of material parameters is identified. • Impacts of consistent and individual material parameter dispersion are quantified. • A practical GUI applicable to similar equipment types is developed.