The Impact of Forecast Characteristics on the Forecast Value for the Dispatchable Feeder

Transforming the energy system to decentralised, renewable energy sources requires measures to balance their fluctuating nature and stabilise the energy system. One such measure is a dispatchable feeder, which combines inflexible prosumption with a flexible energy storage system. The energy storage system’s management is formulated as a stochastic optimisation problem that requires energy time series forecasts as input. These forecasts can significantly influence the performance of the dispatchable feeder: the forecasts have a so-called forecast value for the dispatchable feeder, which is not directly reflected by error-based forecast quality metrics. Therefore, we analyse how the considered forecast value for the dispatchable feeder is related to the considered forecast quality and influenced by forecasts with different characteristics. Furthermore, we examine the impact of problem-specific parameters such as the data and the battery capacity. To this means, we create forecasts with different characteristics using neural networks with varying loss functions and perform the analysis using a data set with 300 buildings. The results of our analysis show that the relation between the considered forecast quality and forecast value for the dispatchable feeder is non-monotonic. Furthermore, we show that the forecast characteristics influence the forecast value differently depending on the data and the battery capacity.