Abstract

In microgrids, Water-Energy-Food Nexus refers to the interconnected relationships between food production, energy systems, and water resources. Thus, modelling these interdependencies is important for reaching a proper management system design for this nexus. Among the methods available in the literature to implement energy and water management systems, reinforcement learning agents can be highlighted because of their capacity to control complex dynamic systems without requiring a model of the process. This work proposes two novel reward functions for reinforcement learning control, allowing the optimal management of resources in a Water-Energy-Food microgrid. The proposed control scheme follows a hierarchical two-layer approach, where one reinforcement learning agent is used for operating each layer. At the top layer, its corresponding agent maximises long-term requirements, such as crop yields, while the bottom layer’s agent manages short-term water and energy use. These agents are trained using three reinforcement learning algorithms to analyse their strengths and limitations. Simulation results confirm that the proposal fulfils the control objectives and the feasibility of using the proposed reward functions to manage the available resources optimally.