Research

This project aims to develop a publicly available dataset of wind power with uncertainty quantification for the current and future onshore and offshore wind plants across the continental US within the domain of the Wind Toolkit. The objectives include: (1) developing 5 minute, 2 km super-resolved grid cell meteorological and power datasets; (2) developing a database of plant-level power time series; (3) quantifying plant-level power uncertainty and identifying the uncertainty sources and driving factors; (4) assessing physical accuracy of super resolved meteorological data, power data, and uncertainty quantification by comparison with observations (e.g. WFIP2) and historical actuals; (5) integrating the developed datasets into existing WETO datasets such NREL WIND toolkit and DOE Wind Data Dashboard database; and (6) engaging with industry stakeholders for data dissemination and feedback.

In this research, we propose a co-occurrence semantic segmentation theory based on cognitive scene understanding, therefore, solving challenging computer vision issues, including label imbalance and boundary ambiguity. The co-occurrence semantic segmentation theory is based on the conjecture that some relevant elements in images tend to occur and be located in a predictable manner, which is a conditional/marginal probability problem. The proposed co-occurrence semantic segmentation will be verified by BTM-PV array detection through satellite aerial image processing.

This project aims at generating open-source datasets consisting of one year of time-coincident load, wind, and solar actuals and probabilistic forecasts for regions similar to ERCOT, NYISO, SPP, and MISO. These datasets serve as an uncertainty quantification basis for the ARPA-E Performance-Based Energy Resource Feedback, Optimization, and Risk Management (PERFORM) program. The project utilizes several resource assessment, resource modeling, and machine/deep learning techniques/datasets, including the the National Solar Radiation Database (NSRDB), the Weather Research and Forecasting Model, the European Centre for Medium-Range Weather Forecasts (ECMWF), the Renewable Energy Potential (reV) Model, the Bayesian Model Averaging, and the Machine Learning-based Multi-Model (M3).

This project will bring probabilistic solar forecasts into ERCOT’s real-time operation environment through automated reserve and dispatch tools that increase economic efficiency and improve system reliability. The adaptive reserves aim to reduce overall reserve levels by 25% while maintaining or improving system reliability. The risk- parity dispatch will automate the use of probabilistic forecast information in a 5-minute dispatch window. The situational awareness tool will present forecast uncertainty information that is relevant, timely, and allows for better decision making.

This project developed an open-source situational awareness and decision support platform called WindView that provides grid operators with knowledge on the state and performance of wind power on their system. In this project, we developed a wind power forecaster, Machine Learning-based Multi-Model (M3), which provides both deterministic and probabilistic wind power forecasts and has been integrated into WindView.

This project addressed the challenges of collecting decentralized information in power systems, by developing an innovative dynamic big data driven nonintrusive disaggregate forecasting methodology based on deep machine learning. Detailed information is provided by the developed framework at the different levels to help the electricity users, utilities, and policy makers for better power system management.

The large amount of historical weather, outages, smart meter data collected by Oncor provides an opportunity for new analysis and forecasting of power outage and load. This project developed an data-driven analytical framework to produce data manipulation processes and algorithms enabling premise-based load forecasting for Oncor, which has been integrated in Oncor's Cloud Computing Server and Enterprise Data Warehouse.