Reinforcement learning algorithms for exploiting flexibility within a Net-Zero Energy Factory

Designing Net-Zero Energy Factories is a significant step towards sustainable industrial operations. Within industrial systems, the identification and exploitation of flexibility is a challenging task. This paper presents a unique approach to energy management, leveraging reinforcement learning to operate a manufacturing system (carpentry) as a Net-Zero Energy Factory. The plant operates on custom orders, allowing for flexibility in machinery operation, including options to postpone or anticipate tasks. In addition to real-time energy consumption data from various machines and energy production data from a 126 kW Photovoltaic (PV) system installed on the plant's roof, the plant also has an integrated battery system. This additional feature enhances the plant's energy management flexibility, enabling it to store excess energy or supply it when needed. We introduce an algorithm that uses these various data sources to suggest whether to continue, postpone, or adjust current energy consumption, as well as managing battery storage and discharge based on the energy demands and production. The study also compares three reinforcement learning algorithms, providing a comprehensive evaluation of our approach. The results highlight the potential of reinforcement learning in managing energy efficiently, operating the plant towards Net-Zero Energy Factory. This work contributes to the understanding of renewable energy usage optimization in the manufacturing sector, particularly in harnessing battery systems for energy storage, offering a promising pathway to sustainable manufacturing and energy management.