Date of Award
Santa Clara : Santa Clara University, 2023.
Doctor of Philosophy (PhD)
Computer Science and Engineering
With the rapid development and deployment of distributed energy resources (DERs), Internet of Things (IoT) technologies, and innovative energy trading markets, residential buildings as the primary energy consumers show enormous potential in promoting energy sustainability. However, achieving residential energy sustainability faces various challenges. (1) Promoting consistent energy-efficient behaviors among all residents can be challenging due to variations in their awareness and commitment to energy-saving practices. (2) While training can cultivate energy-saving habits among residents, educating residents without energy-related knowledge to effectively coordinate and operate various electrical appliances and DERs remains challenging, especially with the widespread adoption of such equipment. (3) When residential buildings participate in transactive energy market (TEM) of a smart grid, the optimal management of energy generation, storage, consumption, and profitable energy trading presents greater challenges and demands for residents. (4) Although energy control systems that rely on occupancy information can assist in achieving automated energy consumption management in residential buildings, accurately predicting occupancy information using data collected from low-cost and non-intrusive sensors remains a challenge. (5) Some energy trading models have been proposed to coordinate different DERs and optimize trading strategies. These models have limitations in terms of residential building privacy, market scalability and stability, and optimization of energy management and trading.
To address above challenges, in this thesis, three efficient models are proposed to help residential buildings achieve energy sustainability by improving energy efficiency. (1) The proposed trust-based occupancy detection model can provide accurate occupancy status to energy control systems, thereby increasing the energy efficiency of residential buildings through reduced energy consumption. Specifically, instead of using expensive or intrusive detection sensors, this proposed model only adopts five low-cost and non-intrusive IoT sensors. Due to their non-intrusive nature, such sensors may be triggered so infrequently that takes several weeks s to collect sufficient training data. To address this challenge, the proposed model designs a human active sequence module to effectively preprocess the raw sensor data and a trustworthy sequence matching module to make accurate occupancy decisions based on limited training data. (2) The proposed deep reinforcement learning (DRL) based price model with the accurate occupancy information can increase the energy efficiency of residential buildings by managing the controllable loads and rewarding energy-efficient behaviors. Specifically, in this model, the residential buildings are integrated into a TEM, where residential buildings become market-interactive consumers who can bid the energy with preferred price. In fact, the traditional TEM is difficult to model explicitly due to its complex market clearing mechanism and random bidding behaviors of market participants. To address this challenge, a double uniform auction based TEM is designed to efficiently manage the energy trading and maximize perceived social welfare. More importantly, a DRL based price model is proposed to reduce the total energy payment while maintaining preferred comfort level by optimizing the energy bidding price and controlling the energy consumption of appliances. (3) Based on the DRL price model, a DRL based energy trading model is proposed to further improve energy efficiency of residential buildings by additional energy management (e.g., energy generation and storage). Specifically, in this model, the residential buildings are equipped with multiple and diverse DERs, which convert buildings into market-interactive prosumers. To maximize the economic benefits, optimize the energy consumption, ensure the scalability and privacy of the market environment, a DRL based energy trading model with distributed learning and execution is proposed to generate the optimal energy trading strategies (e.g., role, price, quantity). Comprehensive experimental results show that (1) the proposed trust-based occupancy detection model can accurately predict the occupancy status, especially when only limited training data is available, (2) the proposed DRL based price model and energy trading model can effectively balance the energy payment and comfort satisfaction for residential buildings, (3) the proposed DRL based energy trading model can efficiently manage the energy generation, storage, trading and consumption.
Jiang, Jun, "Enhancing Residential Energy Sustainability with Occupancy Detection and AI-based Trading" (2023). Engineering Ph.D. Theses. 47.
Available for download on Wednesday, July 16, 2025