The operation of power systems with high shares of wind and solar energy requires reliable tools to facilitate a secure grid and market integration. For over 15 years many firms and research institutes around the world have been developing and operating forecast systems to capture the weather-dependent variability in the power production of a huge number of decentralized wind turbines and photovoltaic (PV) systems. Today, there are forecast systems especially developed to predict the power production of single wind and solar units and variously sized portfolios, as well as within local transformer stations and sub-grids, distribution and transmission grids, and entire countries. In the literature, hundreds of ways to improve forecast quality are described and most of them are already implemented within the process chain of forecast providers.
Many grid operators and power traders are already experts in handling and interpreting wind power and PV forecast time series. In most cases, they establish automatic systems to process the forecasts within their operational management and trading strategies. However, most of these forecasts are deterministic, i.e. they are tuned for the best average performance in terms of an expected value. The forecast error, i.e. the actual deviation between forecasted and measured value at one point in time, indicates the quality of deterministic forecasts. Improvements in the forecast quality in terms of average errors have already leveled off. In contrast, the forecast uncertainty has not. This refers to a range of possible values in the future and hence provides additional information to quantify the uncertainty or warn of alternative weather conditions affecting power system operation.
It is worth noting that uncertainty forecasting in general is no new research area. Financial mathematicians have for years applied probabilistic approaches based on marginal probability distributions and scenario forecasts to optimize their revenues. Probabilistic products like rainfall probability or weather warning systems are also state-of-the-art in meteorology. Due to their inherent weather-dependence, uncertainty forecasting of wind and solar power are often based on probabilistic weather forecasts and methods from meteorology. Weather forecasts from a physically based ensemble prediction system (EPS) together with advanced calibration methods are often used to generate probabilistic wind and solar power forecast scenarios. The key to this approach is the set of numerical weather forecasts that make up the EPS, produced by perturbing the initial or boundary conditions or the result from different parameterization schemes of one weather model. The EPS is configured to represent the physical uncertainty of the weather ahead of time rather than uncertainty as a function of past experience. In practice, this means that the EPS is “event-driven,” i.e. it can produce outliers and even catch extreme, even 50-year events.
Aside from a physically based EPS, statistical algorithms can also generate power uncertainty forecasts. Statistical or machine learning algorithms with an adequate loss function can be fit to a single historical weather forecast and associated wind and/or solar power data, and then generate uncertainty power forecasts from live weather forecasts. This is very distinct from physically based EPS methods, because even long time series of historic data contain too few extreme events to teach the learning algorithms to forecast very rare outlier and extreme events.
Even though there are many different methods to generate a reliable uncertainty forecast of wind and solar power production, they are not currently sufficiently established in the energy economy. We have however observed a clear trend towards probabilistic power forecasts in power systems with shares of weather dependent energy sources exceeding about 30% of gross annual energy supply. In such systems, there are times when renewable energy sources cover nearly the complete power consumption. During these times, the renewables must also provide system services in terms of active and reactive power for frequency and voltage stability issues. Uncertainty forecasts are here often applied for the dynamic dimensioning of the required reserve power and the provision of reserve power by wind farms and PV systems.
Due to electric grids not having been constructed to transport huge amount of decentrally generated wind and PV power, the amount of redispatch and RES curtailment needed for grid stability has increased in many power systems with increasing RES installations. In response, proactive concepts for flexibility like demand-side management, storage (incl. power-to-x), and the establishment of local flexibility markets are currently under investigation in different countries. Two aspects are noteworthy here. Firstly, we are sure that the usage of uncertainty forecasts and probabilistic methods in general are key drivers not only of the secure operation of flexible resources, but also of new business models that make the provision of flexibility much more attractive. Secondly, redispatch and curtailment of RES, as well as the proactive use of flexible resources, affect the RES grid feed-in. If these grid-security-related actions are unscheduled, the forecast must then anticipate them to calculate the resulting power flows and network operating resources. In some countries, one can already observe that the measured grid feed-in is much smaller than the expected feed-in based on the present weather situation. The grid-security-related actions are an additional source of uncertainty besides the weather, resulting directly from the grid operators’ control actions.
A paradigm shift is required to facilitate even larger shares of renewable energy sources. This regards the way forecasts have been used and evaluated so far and the implementation of uncertainty forecasting and new ways of handling uncertainties inherent to renewable power sources. Future challenges can be tackled this way and more flexible, secure, and economical power systems can be further developed. Uncertainty forecasting is a key element to developing this next generation power system. It is here to stay.
Dr. Jan Dobschinski
Head of Forecasting in Energy Systems Group
Fraunhofer Institute for Energy Economics and Energy System Technology IEE