COLLABORATIVE RESEARCH: Data-Driven Risk-Averse Models and Algorithms for Power Generation Scheduling with Renewable Energy Integration

点击翻译按钮获取中文摘要

Renewable energy has been increasingly penetrating into the power grid system during the past years due to its contribution toward cleaner and lower-polluting American energy. Meanwhile, however, its intermittent nature brings challenges to power system operators. One challenging problem is how to derive a cost-effective and reliable power generation scheduling for thermal units in a short time to accommodate renewable generation uncertainties. The other outstanding question is how the data collected by the renewable facilities and intelligent devices can be transformed into valuable information and actionable insights in the decision-making process. To help address these challenges, this project aims to explore innovative data-driven optimization models and develop corresponding intelligent algorithms, as well as the implementation of the algorithms in high-performance computing facilities, to achieve cost-effective and robust daily power system operations. If successful, the proposed innovative approaches can be implemented in the industry in a short time and help improve current operations practices. The results of research outcomes will be incorporated into course works, which will train students to utilize cutting edge data-driven optimization methods to solve upfront power system problems with renewable energy integration. Educational activities also include outreach to K-12 students to promote science and engineering and to under-represented minorities in all aspects of this research effort. The proposed creative approach integrates statistical and optimization methods to derive innovative decision-making under uncertainty models for optimal power flow and unit commitment problems incorporating demand response and renewable energy. It provides one of the first studies on data-driven optimization addressing distributional ambiguity for power system operations. Starting from a given set of historical data, a confidence set for the true unknown distribution is constructed and accordingly data-driven risk-averse optimization models are developed for both system operators and market participants. Besides ensuring system robustness, the advantage of this approach is that the conservatism of the proposed model is adjustable based on the amount of historical data and eventually vanishes as the size of historical data goes to infinity. Also, the proposed advanced techniques in strengthening the formulation by exploring the problem structure and decomposition algorithms implementable at high-performance computing facilities can help improve the computational efficiency to solve the derived models. Finally, integration of innovative data-driven optimization models and development of efficient algorithms will enrich the tool set and advance the cutting edge technology to solve power generation scheduling problems under uncertainty.