A Two-Level Method for Solving Power Generating Unit Commitment Problems

This paper is concerned with the solution of unit commitment problems by means of a two-level evolutionary algorithm (EA), handling the encoded operational states of units. The EA is assisted by the augmented Lagrange relaxation method to compute the optimal units' loading, for each candidate configuration. Emphasis is laid on the reduction of the CPU cost of the proposed method and, for this purpose, a two- level EA has been devised. At the first-preparatory level, the time units (hours) are grouped, based on heuristics. This gives rise to a coarse-grained optimization problem with a reduced number of unknowns which can readily be solved using EAs, without taking into consideration constraints related to units' start-up and shutdown. The so-obtained, sub-optimal and, likely, infeasible solution is then processed by the second level for further refinement. At this level, instead of optimizing the unit commitment over the whole period of time, a small number of consecutive sub-periods are formed and the corresponding optimization sub-problems are solved iteratively. For each sub- problem, an EA coupled with chromosome repairing and cost function penalization to account for inconsistencies between the current sub-problem solution and those computed at the adjacent sub-intervals.