UC Riverside

Renewable energy is an environment-friendly and economically attractive source of electricity generation. However, substantial grid integration of renewable energy is challenging as the power generation from renewables is weather-dependent, highly intermittent, and uncontrollable. To address these challenges, we exploited machine learning and data analytic techniques to develop frameworks and algorithms for integrating renewables into the grid.

Distribution grid planning, control, and optimization require accurate estimation of solar photovoltaic generation and electric load in the system. Most small residential solar PV systems are installed behind-the-meter making only the net load readings available to the utilities. We developed an unsupervised framework for estimating solar PV generation of individual customers by disaggregating the net load readings. Next, we developed an unsupervised framework for joint disaggregation of the net load readings of a group of customers. Our algorithms synergistically combined a physical PV system performance model for individual solar PV generation estimation with a statistical model for load estimation.

High solar PV penetration in the distribution grids gives rise to frequent voltage fluctuations due to the intermittent nature of solar PV production. The slow operating conventional voltage regulating devices, therefore, need to be supplemented with fast operating real and reactive power control of smart inverters. Complete and accurate information about distribution network topology and line parameters needed for traditional model-based Volt-Var optimization methods is often unavailable. To tackle these challenges, we developed a two timescale Volt-Var control framework with model-based slow timescale control and a reinforcement learning-based fast timescale smart inverter control. The proposed framework does not rely on any distribution network secondary feeder information but requires primary feeder information. Next, we proposed a completely model-free reinforcement learning-based two timescale Volt-Var control framework that does not rely on any distribution network primary or secondary feeder topology or parameter information.

Natural and anthropogenic aerosols have a great influence on meteorological variables which in turn impact the reservoir inflow and ultimately hydropower generation. We developed a comprehensive framework to quantify the impact of aerosols on reservoir inflow by integrating the physical Weather Research and Forecasting Model with chemistry (WRF-Chem) and a statistical dynamic regression model. We quantified the impact of aerosols on hydropower generation and revenue by incorporating the hydropower operation optimization toolbox into the framework.

Lastly, we developed a data-driven framework for the predictive maintenance of distribution transformers to increase the reliability of the distribution system. We utilized readily available data such as the transformers' specification, loading, location, and weather.