Solving a real-life, large-scale energy management problem

This paper introduces a three-phase hybrid heuristic for a large-scale energy management and maintenance scheduling problem. The problem is to schedule maintenance periods and refueling amounts for nuclear power plants with a time horizon of up to five years, and handling a number of scenarios for future demand and prices. The goal is to minimize the expected total production cost. The first phase of the heuristic solves a constraint programming model of a simplified version of the problem, the second performs a local search, and the third handles overproduction in a greedy fashion. This work was initiated in the context of the ROADEF/EURO Challenge 2010. In the concluding phase of the competition, our team ranked second in the junior category and sixth overall. After correcting a small implementation bug in the program that was submitted for final evaluation, our solver ranks first in the overall results from the competition.