A New Formulation for Planning Electric Power Generation Schedules*
Guglielmo Lulli
Lancaster University
July 14, Friday 13:40
EA-409
The Unit Commitment (UC) problem, which is the problem of planning electric power generation schedules, has attracted intense research for many decades. Lagrangian Relaxation was the prominent methodology to solve this class of problems, but more recently (roughly this millennium) direct MIP methods have become more popular. In this talk, I will present a new formulation for the UC problem, which is based on a new class of decision variables called state transition variables. This formulation establishes a completely new benchmark, cutting in half the computational time required by the state-of-the-art model of Morales-Espana et al. (IEEE Trans. on PS, 2013) to solve realistic instances (from the US Federal Energy Regulatory Commission) with about a thousand generators over a week.
The computational success of the proposed formulation derives from the following properties: i) the proposed formulation naturally incorporates many of the valid inequalities (cutting planes) in the literature without the need for any pre-processing, and ii) it induces a network structure. These features not only enhance the interpretability of valid inequalities, but in addition, the network structure helps attain a very high percentage of integer decisions when the LP relaxation is solved by the MIP solver.
*joint work with S. Atakan and S. Sen (USC)
Brief bio of the speaker
Guglielmo Lulli is senior lecturer in Network Analytics at Lancaster University. His professional expertise, research and teaching interests focus on deterministic and stochastic optimisation particularly as applied to transportation and logistics operations, air traffic flow management and energy. He received the 2013 Best Paper in Transportation Science and Logistics Prize of the Transportation and Logistics Society of INFORMS and he was recipient of the Fulbright Fellowship at M.I.T.
