UC Riverside

The rapid development of computing devices and artificial intelligence (AI) in recent decades have dramatically reshaped the ecosystem of critical infrastructure systems. Intelligent control and data-driven algorithms have received widespread interests due to their great potential in reducing the operating costs and improving the system efficiency and reliability. The increasing data collected from different sectors of infrastructure provide abundant resources for scientific studies propelled by machine learning and statistics. Nevertheless, successful design and applications of novel intelligent algorithms on infrastructure can be challenging due to the complex domain-specific contexts and constraints therein. The goal of this dissertation is to investigate and design innovative solutions to emerging problems in power systems, transportation, energy efficient buildings, and wildfire camera networks. For power transmission systems, we developed automatic event detection and identification algorithms based on real-world synchrophasor data. For power distribution systems, we investigated and characterized the spatio-temporal correlation between distribution feeders with statistical models. We analyzed and quantified the impacts of different socioeconomic and weather factors on residents' electricity consumption. For transportation systems, we initially formulated the electric vehicle (EV) routing for ride-hailing services as a mixed integer programming problem. This framework does not scale well to large amount of EVs. To address this issue, we developed a reinforcement learning based algorithm to operate an EV fleet, which is characterized by decentralized learning and centralized decision making. For energy efficient buildings, we designed an innovative change-point logistic regression model to provide an accurate forecast of building occupancy. A novel building HVAC control algorithm, which aims at reducing the energy consumption, was developed by embedding the occupancy prediction model into a model predictive control framework. \par For wildfire camera networks, we developed an efficient video smoke detection framework designed for embedded applications on local cameras. We also proposed an optimal wildfire camera placement strategy by maximizing the overall camera network benefits with limited budget.