Scenario reduction techniques and solution stability for stochastic unit commitment problems

In power systems rife with uncertainty, stochastic unit commitment (SUC) models may be used to properly size and allocate operational reserves, in order to ensure a reliable and cost-efficient operation of the power system. The performance of SUC-based unit commitment schedules is however fully dependent on the scenario sets used to describe the uncertainty at hand. Dedicated scenario generation & reduction techniques (SGT & SRT) have been developed to generate and select scenario sets that capture the uncertain parameter, e.g. wind power, and yield a cost-optimal unit commitment (UC) schedule in reasonable computing times. Probability-distance based SRTs are by far the most used. In an extensive numerical study, we analyze the performance of so-called cost functions used in these SRTs. In addition, we propose a new cost function, which allows selecting a well-balanced subset of scenarios, resulting in a tractable SUC model and a cost-optimal UC schedule.