Learning-based scheduling of industrial hybrid renewable energy systems

• Reinforcement Learning based scheduling of industrial hybrid energy systems is developed. Hybrid mechanistic/data-driven modeling approach for optimal management proposed. solar/wind/fuel cell generation system integrated with chlor-alkali process. Minimization production cost and environmental carbon emissions achieved. The propagation distributed renewable resources poses several challenges in the operation microgrids due to uncertainty. In traditional approaches, algorithm often depends on accurate forecasts uncertainties, which many cases add complexities problem. While data-driven algorithms have been demonstrated overcome aforementioned challenges, these methods work better only if action space finite. However, most real world problems continuous making unsuitable practical applications. To address issues, this paper proposes a purely policy-based called Proximal Policy Optimization (PPO) that adopts (RL) determine schedule mix (HRES), case study highly energy-intensive We implement PPO, an Advantage Actor-Critic RL method enables samples be trained multiple epochs mini-batch updates effectively distribute among HRES without having access dynamics system. combat increased grid dependency process achieve profitability, grid-connected network consisting solar photovoltaic panels, wind turbines, fuel cells considered deliver power Simulation results demonstrate dispatch available sources overall economic saving around 32.8% corresponding emission reduction 28.5% by adopting as compared connected This framework can generally used any taking advantages its model-free property.