UC Riverside

With the increasing complexity of building infrastructure, building management systems have been widely used for managing various types of energy loads and optimizing building energy efficiency. In large commercial buildings, 50% of energy consumption is from HVAC (heating, ventilation, and air conditioning) for maintaining a comfortable environment throughout the day. The rapidly growing EV (electric vehicle) charging demand has distinct operating patterns and may cause spikes in power demand. In addition, majority of datacenters are collocated within mixed-use buildings, where they often share some common infrastructures and energy supplies with other building operations. To maximize building energy efficiency, i.e., effectively lower peak power demand and reduce overall electricity cost, this dissertation develops novel model predictive control based algorithms to co-schedule HVAC, EV charging and datacenter workload with heterogeneous power supplies (e.g., power grid, solar energy and battery storage) in a holistic framework. Furthermore, to avoid the costly and time-consuming thermal dynamics model design for building control and to further explore more effective building management scheme, we also present deep reinforcement learning based algorithms to intelligently learn the effective control strategies through direct interactions with real buildings, without relying on any inflexible models. The experiment results demonstrate the effectiveness of our data-driven approaches in improving building energy efficiency, while maintaining the desired temperature for building occupants, and meeting the deadlines for EV charging and datacenter workload.

To further exploit the significant energy scheduling flexibility provided by smart buildings, in addition to intelligently managing buildings' energy loads, it is also essential to coordinate the behavior of large number of buildings for enhancing the efficiency and stability of the entire power system. With the advent of advanced metering infrastructure, various energy demand from smart buildings can be coordinated together with power plants across the power grid. Most of the previous works focus on developing price-based demand response (DR) strategies, in which the buildings can only passively react to the market signals (e.g., real-time electricity price) from utilities. To improve power system efficiency and facilitate customers' engagement level in electricity market, we present a proactive building demand participation framework to integrate the operation of smart buildings into the electricity market economic dispatch. Our new DR scheme enables building customers to proactively express multi-level energy demand preferences to smart grid operators instead of passively following the load reduction instructions. The experiment results demonstrate that our proactive DR scheme can achieve significant cost reduction for both power generation and building operation, and is more robust to various malicious cyber attacks compared with passive DR strategies.