Abstract
Agricultural operations in rural and semi-rural areas often have unpredictable energy supplies, high operational expenses, and reliance on fossil-fuel generators. This research develops and evaluates a hybrid renewable energy system (HRES) optimisation and reliability framework that includes combined heat and power (CHP) units, photovoltaic (PV) panels, wind turbines, battery storage, and grid connection. The framework uses Proximal Policy Optimisation (PPO), a reinforcement learning algorithm that can handle continuous action spaces and adapt to stochastic renewable availability, and compares its performance to deterministic optimisation using Advanced Integrated Multidimensional Modelling Software (AIMMS).
The HRES components were simulated under operational, technical, and economic constraints, and a multi-objective formulation was developed to minimize operational costs, improve reliability, and increase renewable penetration. A multiobjective model was used to measure reliability, including probabilistic indices such as Loss of Load Probability (LOLP), Loss of Load Expectation (LOLE), Expected Energy Not Supplied (EENS), and cost of Energy Not Supplied (CENS). Five case study configurations, ranging from CHP-only to fully integrated photovoltaic, wind turbine, battery storage, and grid systems, were simulated using accurate South African meteorological and agricultural load profiles.
The reinforcement-learning-based dispatch achieved operational cost reductions of 26-31% relative to AIMMS across all case studies, 20-35% improvements in reliability measures, and renewable utilisation rates surpassing 70%. The finding highlights the potential to improve sustainability, reduce dependence on fossil fuels, and increase energy security in agricultural microgrids. This is the first study to compare reinforcement learning with deterministic optimisation for agricultural applications, resulting in a reproducible approach for sustainable rural electrification and carbon reduction in the agricultural sector.