Development and testing of improved statistical wind power forecasting methods.

Energy Time Series Forecasting

“…Thus, the estimation and maintenance of the involved models is very time consuming [112,138,239]. Thus, to maintain a high accuracy of the provided forecasts, it is neces-sary to automatically pick up those changes and to adapt the employed forecast models accordingly [11,90,128,165]. In addition, an EMS fulfills a number of different tasks each posing different requirements with respect to runtime and accuracy of the forecast calculation [171,172].…”

“…In particular, these changes target to enable quick, automatic, and autonomous reactions on changing consumption and production situations. Furthermore, providing accurate forecasts also means to continuously adapt the employed forecast models with respect to the most recent values to pick up potential changes in the time series developments and characteristics [11,92,165]. While previously the focus was clearly on improving the forecasting accuracy, recently the efficiency of calculating forecasts is gaining increasing importance [18,193,251].…”

“…In this book we explicitly focus on the very-short term and short-term horizon, meaning an intra-hour horizon up to some days ahead. Thus, forecast models are required to pick-up these changes and adapt with respect to changing time series characteristics [11,128,165]. Also it is common sense in the research domain dealing with energy-related forecasting, that the short-term horizon is the most important horizon and the most interesting field of research [128,251].…”

Dynamic Factor Graphs - A New Wind Power Forecasting Approach

“…ARGUS-PRIMA is used to test advanced statistical algorithms for short-term wind power forecasting. The platform, which consists of a set of statistical algorithms to generate wind power point and uncertainty forecasts, has been used for systematic testing and comparison of different computational learning algorithms [21]. For wind power point forecasting, ARGUS-PRIMA uses concepts from information theoretic learning (ITL) for training an ANN.…”

Large-scale probabilistic forecasting in energy systems using sparse Gaussian conditional random fields

“…[10]); or only predicting non-Gaussian marginal distributions for single output variables (e.g. [9]). Indeed, past work has explicitly highlighted the challenge of developing models that can capture joint distributions over future values.…”