A technique for reducing computational effort in Monte-Carlo based composite reliability evaluation

The authors describe a novel technique for reducing the number of samplings in Monte-Carlo-based composite power system reliability evaluation. The proposed technique uses analytical information from simple models (such as a generation reliability model) as regression variables to reduce the variance of LOLP (loss-of-load probability) and EPNS (expected power not supplied) estimates with the complete (composite) model. Sample size reductions of up to two orders of magnitude were obtained in case studies with a 24 bus modified reliability test system and a 124 bus reduced Brazilian system. Speed-ups of 6.1 to more than 40 were obtained for the estimate of the EPNS and of 2.1 to 3.8 for the LOLP estimate. In all cases, the EPNS index required more samplings to converge to the required tolerance (5% relative uncertainty) than the respective LOLP index (for the same tolerance). As a consequence, the higher speedups prevail in the overall convergence if a uniform criterion is adopted