Forecasting the State-of-Charge of Batteries in Micro-Grid Systems

Micro-grid (MG) systems are increasingly deployed on buildings in order to enhance energy efficiency while maintaining occupants' comfort. A MG system includes renewable energy sources (RES) for energy production, storage devices for storing energy excess, and control strategies for orchestrating all components and for improving the efficiency of the system. However, forecasting methods are required to develop predictive control strategies in order to tackle the uncertainty and intermittency of energy production and consumption. The State-of-Charge (SoC) of batteries is one of the main parameters used in predictive control algorithms, because it indicates how much energy is stored and how long a MG system can be relying on the deployed storage devises. Several methods have been developed for SoC estimation, but little work has been dedicated for SoC forecasting in MG systems. In this paper, we introduce a predictive approach for SoC forecasting using the Long Short-Term Memory (LSTM) algorithm. Extensive experiments have been conducted and results are provided and show the accuracy of this algorithm for different forecasting horizons.