Data-Driven Distributionally Robust Control of Energy Storage Systems

Energy storage is an important resource that can balance fluctuations in energy generation from renewable energy sources, such as wind, to increase their penetration. Many existing storage control methods require perfect information about probability distribution of uncertainties. In practice, however, the distribution of renewable energy production is difficult to reliably estimate. To resolve this challenge, we develop a new storage operation method, based on the theory of distributionally robust stochastic control, which has the following advantages. First, our controller is robust against errors in the distribution of uncertainties such as power generated from a wind farm. Second, the proposed method is effective even with a small number of data samples. Third, the construction of our controller is computationally tractable due to the proposed duality-based dynamic programming method that converts infinite-dimensional minimax optimization problems into semi-infinite programs. The performance of the proposed method is demonstrated using data about energy production levels at wind farms in the PennsylvaniaJersey-Maryland interconnection (PJM) area.