A Hierarchical Approach to Probabilistic Wind Power Forecasting

Abstract

This paper describes a method to generate improved probabilistic wind farm power forecasts in a hierarchical framework with the incorporation of production data from individual wind turbines. Wind power forms a natural hierarchy as generated electricity is aggregated from the individual turbine, to farm, to the regional level and so on. To forecast the wind farm power generation, a layered approach is proposed whereby deterministic forecasts from the lower layer (turbine level) are used as input features to an upper-level (wind farm) probabilistic model. In a case study at a utility scale wind farm it is shown that improvements in probabilistic forecast skill (CRPS) of 1.24% and 2.39% are obtainable when compared to two very competitive benchmarks based on direct forecasting of the wind farm power using Gradient Boosting Trees and an Analog Ensemble, respectively.