Battery storage systems (BSS) are critical for maintaining stability and flexibility in renewable energy microgrids. However, the generation and load profiles introduce challenges for optimal energy management. This study presents a deep reinforcement learning (DRL) approach to real-time battery scheduling in a grid-connected microgrid. The control problem is modeled as a Markov decision process (MDP), and a Deep Q-Network (DQN) agent is trained to output optimal charging and discharging actions based on system states. Results indicate that the proposed DQN-based EMS achieves superior performance in balancing economic cost, energy self-sufficiency, and battery health compared to conventional rule-based strategy. This study contributes to advancing reinforcement learning-based approaches for real-time battery scheduling in microgrid energy systems.

The study introduces a novel Radial Basis Function Neural Network-based Super-Twisting Sliding Mode Collective Blade Pitch Control (RBFNN-STSM-CBPC), designed specifically for semi-submersible platform-based Floating Offshore Wind Turbines (FOWTs) operating above rated speed (Region III). The proposed composite controller is developed using a refined nonlinear Control-Oriented Model, including lumped unmodeled dynamics and external disturbances. To our knowledge, this is the first time that a neural network STSM-CPBC approach is designed for this application. The RBFNN operates as an adaptive observer for the lumped disturbance, enhancing the robustness and performance of the standard STSM-CBPC for the same gains. Its adaptive law, formulated through the Lyapunov method, ensures stability and convergence by adjusting the adaptive weight. Simulation results demonstrate the superiority of the RBFNN-STSM-CBPC over the standard STSM-CBPC method in regulating rotor speed and mitigating platform motion.

The floating offshore wind turbine (FOWT) technology has great energy potential, however, minimizing the movement of the structure, under the combined effect of wind and waves, while ensuring maximum power extraction on all operating ranges remains a challenge. This paper proposes the design of a deep reinforcement learning (DRL) controller for FOWTs in the operating area III. To our knowledge, this is the first time that DRL-based control approach is used for this application. The proposed DRL controller is based on trust region policy optimization (TRPO) algorithm, composed of two neural networks, the actor and the critic networks, for the learning of the optimal control law. Simulation results and comparison study are provided to validate the proposed DRL controller for the 5-MW baseline ITI Barge wind turbine model on OpenFAST.