Scalable, approximate dynamic programming algorithms for high-dimensional storage portfolios

点击翻译按钮获取中文摘要

AbstractThe objective of this research is to develop methods to design and control heterogeneous portfolios of energy storage devices for the power grid. The approach of this research is to use approximate dynamic programming to develop optimal control policies, that will then be used to understand the economic value of different technologies in the context of a complete power grid. Intellectual meritThe intellectual merit of the project is the development of scalable algorithmic technologies for solving high-dimensional stochastic optimization problems arising in energy storage. We propose to use the framework of approximate dynamic programming coupled with tools from machine learning and convex optimization. We exploit convexity which makes it possible to construct effective approximations that scale to handle large numbers of storage devices. Broader impactsThe broader impacts of the research will be: 1) The research will guide the design of storage devices so that they meet the specific needs of the power grid in the presence of large supplies of intermittent energy such as wind and solar. 2) Renewable energy, coupled with appropriately designed storage, should dramatically reduce the need for coal. 3) The research, including the approximate dynamic programming models and algorithms, will be made available using a special website with datasets, software, published research and working papers, and downloadable presentations. The results will be integrated in courses at Princeton University, and presented at conferences and workshops to a broad community spanning energy systems and economics, as well as the algorithmic communities.