The increasing share of solar and wind energy production introduces challenges related to excess electricity generation and its negative prices, which affect the stability of power systems. Battery storage is emerging as a promising solution, but its optimal operation requires advanced control strategies. This paper compares two reinforcement learning methods for battery management: the tabular SARSA and the deep QNetwork (DQN) method. Both methods are trained on historical data from a photovoltaic plant and electricity market prices, with the objective of minimizing energy purchase costs. The results demonstrate that both methods significantly reduce the costs compared to the baseline without storage, with DQN outperforming SARSA due to its ability to capture complex patterns and handle continuous data input. The findings confirm that reinforcement learning can effectively support decision-making in energy storage management, contributing to improved economic efficiency and system stability while also highlighting some implementation challenges, such as training instability and sensitivity to hyperparameters.
|
