Evaluation of Minimized Expected Cost Forecasts of PV Energy Generation for Multiple Deviation Cost Functions

Forecasts of PV energy generation are used by generators to offer supply to the energy market. Generators face deviation costs for over/under forecasting energy production. We derived a forecast that minimized the expected deviation cost given a probabilistic forecast of energy generation: the minimized expected cost (MEC) forecast. We evaluated the average deviation cost of the MEC forecast against five deterministic forecasts: a persistence forecast, an NWP derived forecast, the mean of the probabilistic forecast, a “rule of thumb” under-forecast of the mean, and a gradient boosted trees regression model forecast. We find the MEC forecast to have the lowest average cost when evaluated against four hypothetical deviation cost functions. The MEC forecast reduced deviation costs by between 2% and 6% compared against the two most skilled alternate forecast methods, and by between 77% and 195% compared against the persistence and NWP derived reference forecasts.