Robust Policies for Storage Used to Offset Renewable Variance

We describe a class of multistage optimization models used to plan storage operation in order to offset deviations from renewable output forecasts. These are OPF-like models where policy computation takes place at time zero, and at time t = 1, 2,… T storage operation is revised using a linear policy. We assume that renewable output deviations are estimated from measurements; errors in forecast and in measurement are handled using robust constraint modeling. The constraints we consider are line limits, and, especially, battery capacity limits. A notable detail is that due to differing charging and discharging battery efficiencies, battery operation is described using a nonconvex model (as has been noted by other authors). However, our cutting-plane algorithm is able to efficiently separate from this set.
In this talk we focus on several issues: (1) scalability of the algorithm to grids with thousands of buses and lines, (2) impact of the robust data model, and (3) impact of using "attribution" storage operation policies, whereby a given storage unit is used to respond to a selected subset of renewable locations only.
Daniel Bienstock is Professor at the IEOR and APAM departments at Columbia University. His work focuses on high-performance optimization algorithms, nonconvex optimization, and applications of optimization to topics of critical interest, such as the safe operation of power grids. He is an INFORMS fellow.