A stochastic approximation approach to load shedding inpower networks

A system comprising a utility company serving a set of electricity end-users is considered. The utility company can purchase energy from the wholesale market. It is also connected to a renewable energy production facility, from which it can harvest energy at no cost, and also to a battery for energy storage. Ahead of a scheduling horizon, the utility purchases energy based on forecasted demand and renewable energy production. During online operation, if the renewable energy is not adequate, real-time decisions with respect to user load shedding, energy procurement, and battery charging or discharging need to be made. The problem is cast in a stochastic approximation framework, and is solved online via a dual stochastic subgradient method with low per-slot complexity.