Meteorological Benchmark Forecasts for Energy Management Systems

Novel energy management and scheduling algorithms that tightly integrate volatile renewable energy sources often depend on external forecasts such as numerical weather predictions. Since the quality of the forecasting inputs may widely affect the system performance, detailed assessments are needed. This work addresses the need for large-scale forecasting datasets by presenting methods to refine and generate artificial benchmarking inputs. A k-NN-based localization approach and a synthetization technique generating artificial PV forecasts for benchmarking purposes are herein developed and thoughtfully assessed. It is shown that the localization approach can successfully push the accuracy of coarse long-term reforecasting datasets into the range of state-of-the-art services. The artificial forecast generation method complements the work by providing highly controllable benchmarks. Hence, new prospects in assessing algorithms under test with respect to their forecast-quality requirements are provided.