Interpretable uncertainty forecasting framework for robust configuration of energy storage in a virtual power plant

Accurate forecasting of load and renewable energy sources (RES) is crucial in the configuration of an energy system. However, uncertainties of these variables present a challenge when making decisions regarding the energy system. To address this challenge, this paper proposes an ensemble interpretable uncertainty forecasting framework called En-IENN, based on an Interpretable Neural Network (INN) and an improved Evidential Neural Network (IENN). The raw features are initially filtered by INN and subsequently utilized as inputs for IENN to forecast the uncertainty. INN clearly explains the reasons for feature selection, and ultimately, the evidential distribution is determined by minimizing an improved loss function of ENN. In comparison with other feature selection and uncertainty forecasting methods, En-IENN is demonstrated to be more interpretable and reliable. After that, utilizing the predicted load and RES, a battery capacity configuration in a virtual power plant is simulated to demonstrate the effectiveness of En-IENN. By comparing the configuration under different prediction intervals of loads and RES, the En-IENN is proven to be superior to other prediction methods.