Data analysis and robust modelling of the impact of renewable generation on long term security of supply and demand

This paper studies rigorous statistical techniques for modelling long term reliability of demand and supply of electrical power given uncertain variability in the generation and availability of wind power and conventional generation. In doing so, we take care to validate statistical assumptions, using historical observations, as well as our intuition about the actual underlying real-world statistical process. Where assumptions could not be easily validated, we say so explicitly. In particular, we aim to improve existing statistical models through sensitivity analysis of ill-known parameters: we propose models for wind power and conventional generation, estimate their parameters from historical wind power data and conventional availability data, and finally combine them with historical demand data to build a full robust joint time-dependent model of energy not served. Bounds on some useful indices from this model are then calculated, such as expected energy not served, and expected number of continuous outage periods-the latter cannot be estimated from a purely time collapsed model because time collapsed models necessarily do not model correlations across time. We compare our careful model with a naive model that ignores deviations from normality, and find that this results in substantial differences: in this specific study, the naive model overestimates the risk roughly by a factor 2. This justifies the care and caution by which model assumptions must be verified, and the effort that must be taken to adapt the model accordingly.