Enhancing hybrid renewable energy performance through deep Q-learning networks improved by fuzzy reward control

Abstract

In a stand-alone system, the use of renewable energies, load changes, and interruptions to transmission lines can cause voltage drops, impacting its reliability. A way to offset a change in the nature of hybrid renewable energy immediately is to utilize energy storage without needing to turn on other plants. Photovoltaic panels, a wind turbine, and a wallbox unit (responsible for providing the vehicle’s electrical need) are the components of the proposed system; in addition to being a power source, batteries also serve as a storage unit. Taking advantage of deep learning, particularly convolutional neural networks, and this new system will take advantage of recent advances in machine learning. By employing algorithms for deep Q-learning, the agent learns from the data of the various elements of the system to create the optimal policy for enhancing performance. To increase the learning efficiency, the reward function is implemented using a fuzzy Mamdani system. Our proposed experimental results shows that the new system with fuzzy reward using deep Q-learning networks (DQN) keeps the battery and the wallbox unit optimally charged and less discharged. Moreover confirms the economic advantages of the proposed approach performs better approximate to +25% Moreover, it has dynamic response capabilities and is more efficient over the existing optimization approach using deep learning without fuzzy logic.