Smart Grid Seminar

In this talk, we bridge the gap between learning and control of energy systems through two topics. Firstly, we show how to trade-off between model accuracy and control tractability faced by neural networks, by explicitly constructing networks that are convex with respect to their inputs. Then optimal controllers can be achieved via solving a convex model predictive control problem. We present the application of our approach to building HVAC control and distribution system voltage regulation. Secondly, we study how to model DC optimal power flow as a decoding problem and use machine learning and duality theory to achieve orders of magnitude faster computations compared to conventional optimization approaches.