Rolling-horizon optimization integrated with recurrent neural network-driven forecasting for residential battery energy storage operations

Abstract

In recent years, the installation of battery energy storage (BES) integrated with solar photovoltaic (PV) panels in residential houses has been rapidly accelerated tied to the high penetration of renewable energy resources, and it is expected to play a significant role as distributed energy resources (DER) in smart grids and microgrid operations. To directly offset electricity demand by efficiently utilizing renewable energy, it is crucial to optimize BES operations in conjunction with forecasting uncertain renewable energy, electricity demand, and prices. Therefore, there is an urgent need to develop proper methodologies to optimize BES operations in residential settings to realize efficient and sustainable smart grid operations under the high penetration of renewable energy. Given this context, this study intends to develop a rolling-horizon optimization model integrated with a recurrent neural network-driven forecasting that is uniquely designed to interactively forecast uncertainty and optimize BES operations in an iterative fashion. Comprehensive numerical experiments have been conducted based on data collected from residential houses to evaluate the performance of the proposed optimization model. The results demonstrate that the proposed optimization model can be practically applied for optimizing residential BES operations while efficiently utilizing solar power in response to time-varying and uncertain electricity demand and prices. Moreover, because the length of the look-ahead period used for both forecasting and optimization should be appropriately set for the proposed model, experiments have been designed and conducted to investigate the impact of the length of a look-ahead period on overall performance.