Data-driven Demand Response Energy Management for Sustainable and Smart Manufacturing

Demand response energy management is a key enabler for enhancing energy efficiency, electricity demand responsiveness, and cost-effectiveness for modern manufacturing systems. Nevertheless, most current demand response management strategies for manufacturing systems are based on empirical knowledge, statistical analysis, and static optimization. The existing approaches are either biased or time-consuming, which are unsuitable for real-time manufacturing system control under dynamic demand response programs. A grand challenge in dynamic demand response management is making real-time decisions responding to time-varying electricity prices and manufacturing uncertainties. To tackle this challenge and advance state-of-the-art demand response management, in this dissertation, an analytical model-based scheduling approach is first developed for production scheduling under static demand response. Then, based on the analytical system model, an explainable deep reinforcement learning algorithm is proposed for real-time manufacturing system control under stochastic and dynamic demand response. In order to support the data-driven scheduling approaches, data acquisition and processing are necessary. The third topic in this thesis focuses on proposing a cost-effective industrial IoT planning algorithm to address the data acquisition problem on the manufacturing shop floor. Finally, a new evaluation metric is proposed for data processing and real-time price prediction for utilities. This thesis provides insight into the design of sustainable and smart manufacturing systems as well as the application of data-driven methods in demand response control toward sustainable manufacturing. The outcomes of this thesis can facilitate the implementation of demand response in the manufacturing sector and contribute positively to the reliability and stability of power systems.