Uncertainty Forecasting Practices for the Next Generation Power System Corinna M¨ ohrlen * , Ricardo J. Bessa † , Gregor Giebel ‡ and Jess U. Jørgensen * † INESC Technology and Science (INESC , TEC), Porto, Portugal, Email: rbessa@inesctec.pt ‡ Danish Technical University, Department of Wind Energy, Roskilde, Denmark, Email: grgi@dtu.dk Abstract—Probabilistic forecasts in general and ensemble forecasting in particular may contain a paradigm shift in the way renewable energy forecasts have been used and evaluated in the past 20 years, where deterministic forecasting has been established and been practiced in all power markets, where the level of wind power penetration increased over a few percent of total energy consumption. In the next generation power system with large amounts of intermittent and renewable energy sources (RES), where more than a quarter of the energy is delivered by RES, deterministic methods are no longer suf- ficient. Handling the uncertainties that come with the variable weather driven generation from RES is a key requirement for forecasting tools. Probabilistic forecasting methods offer such new ways of handling uncertainties that are inherent in the generation of power from renewable sources. In this paper we demystify the use of uncertainty forecasts by providing some important definitions, showing a number of applications with best practices cases and pitfalls when choosing a solution that fits the current and future development of an end-user. I. I NTRODUCTION Although uncertainty terms are part of our day-to-day communication and language, communication and applica- tion of uncertainty in weather forecasting and the power in- dustry’s decision making is still in its infancy on many levels. Research in psychology and cognitive decision-making has proven over the past decade that uncertainty information not only helps decision making, it also reduces the distrust in forecasts when they fail from time to time. In the world meteorological organization’s (WMO) guide- lines on ensemble prediction [1], the WMO actually warns about ignoring uncertainty in forecasts, even if an end-user receives a deterministic forecasts. The WMO argues that if a forecaster issues a deterministic forecast the underlying uncertainty is still there, and the forecaster has to make a best guess at the likely outcome. Unless the forecaster fully understands the decision that the user is going to make based on the forecast, and the impact of different outcomes, the forecaster’s best guess may not be well tuned to the real needs of the user. It is this gap in the basic understanding of uncertainty inherent in forecasts that lead to wrong assumptions among end-users with little or no experience in basic meteorology or atmospheric science. Mistrust in forecasts and forecasting methods including uncertainty methodologies often stem from a wrong expectation on the quality of forecasts for a specific problem. If uncertainty forecasts should find their way into the power industry’s weather related decision making, a deeper understanding of weather uncertainty, the way weather ser- vices produce uncertainty of weather forecasts, and how such forecasts are to be translated into end-user applications is required [2]. In the following sections, we try to shed some light into the gaps and pitfalls and highlight some of the many advantages of applying uncertainty forecasts in power system applications. II. UNCERTAINTY FORECASTS: A BRIEF REVIEW One of the gaps of understanding uncertainty in the power industry and among those end-users with an interest in uncertainty forecasts due to higher wind power and solar power penetration levels is the definition of uncertainty and the corresponding methodologies that provide forecast uncertainty. In the IEA Wind Task 36 ”Wind Power Fore- casting” 1 interview analysis it was found that many people had difficulties distinguishing: 1) forecast error spread 2) confidence interval 3) forecast uncertainty 4) forecast interval The forecast error spread is defined as the historically observed deviation of a forecast to its corresponding ob- servation at a specific time. It can also refer to an average error provided by an error metric, e.g. variance or standard deviation. One of the common misunderstandings is that a confidence interval is showing the uncertainty of a forecast. This is not the case. By adding and subtracting for example one standard deviation to the deterministic forecast of wind speed and converting it to wind power, such intervals rep- resent a measure of the deviation to climatology and do not represent current or geographically distributed uncertainty. The forecast uncertainty on the other hand is defined as a possible range of forecast values in the future. In meteorology this range is defined by the uncertainty of the atmospheric development in the future and represented in ensemble forecasts by applying perturbations to inital and boundary conditions and expressing model physics differences. When represented in forecast intervals the so- determined uncertainty band represents forecast uncertainty containing the respective probability of the real value being contained in the range of forecasted values, which will only be observed in the future. 1 see http://ieawindforecasting.dk *WEPROG ApS, Assens, Denmark, Email: com@weprog.com, juj@weprog.com 16th Int'l Wind Integration Workshop | Berlin, Germany | 25-27 October, 2017