UC Davis

District Energy systems have been used for centuries as an efficient way of generating and distributing heating and cooling utilities used for space conditioning and industrial process requirements. The centralization of equipment increases their efficiency and facilitates the use of demand-response (load shifting) strategies, allowed by the presence of energy storage. More recently, there has been a push towards the usage of low-carbon technologies, especially electricity-driven equipment such as chillers and heat pumps, and renewable technologies that harness solar and geothermal energy. In this work, we investigate optimization-based methodologies for performing the operation and design of district energy systems, with a focus on low-carbon (electric and renewable) technologies and demand-responsive operation. First, we consider the problem of equipment dispatching in a District Cooling plant with large-scale electrical centrifugal chillers and thermal energy storage, under day-ahead electricity prices. A closed-loop scheduling formulation is proposed and implemented in the real-plant. Next, we investigate the impact of participating in load-curtailment (also referred to as load-reduction) programs, simultaneously with the load-shifting program considered previously. We propose a novel formulation that accounts for both the bidding and implementation phases of the load-curtailment, and a framework that integrates the solution of the preceding problem (either load-shifting or -curtailment) into the following one as constraints. In the following section, we switch our attention to data-based methods for optimal operation. We employ a recently proposed Deep Reinforcement Learning method that learns the optimal demand-responsive policy by interacting directly with the environment. In the first subsection, we employ the Soft Actor-Critic (SAC) method for the optimal operation of District Cooling plants with only continuous actions. The following study focuses on the incorporation of hybrid (discrete-continuous) actions into the SAC framework. The main challenge lies in the back propagation of the gradient through the discrete stochastic nodes (i.e., sampling of the discrete probability distributions). Thus, a comprehensive framework is proposed that considers various ways to represent the stochastic discrete decisions. Finally, we consider a model-based approach, this time focusing on the integrated design and operation of District Heating systems using low-carbon technologies. This study uses optimization to perform techno-economic analysis or site-screening, i.e., identify the potential of each technology considering location-specific conditions.