Bayesian optimisation with Gaussian processes of the external LPS of switchyards
Abstract
This paper will present a method for optimizing the geometry of the lightning protection system (LPS) of an outdoor high-voltage switchgear, which is based on Bayesian optimization using the Gaussian process. This is a statistical optimization method that is additionally combined here with a stochastic approach to analyzing the exposure of an outdoor switchgear to direct lightning strikes. Bayesian optimization implies that there is no analytical expression that describes the objective function, but only a (stochastic) process that, based on the inputs (arrestor arrangement and geometry), produces a certain output (the level of protection of the facility), while the process itself is viewed as a black box. Namely, the analysis of the protection of a facility by the LPS system is viewed more as an iterative process that converges towards satisfying some criterion and less as an exact analytical calculation. The application of the Bayesian optimization method for the outdoor LPS system will be demonstrated on the example of a specific facility.
