Operations Research Seminar: Jean-Paul Watson

Stochastic Unit Commitment: Scenario Generation, Scalable Computation, & Experimental Results

Jean-Paul WATSON, Center for Computing Research, Sandia National Laboratories

The objective in stochastic unit commitment is to optimize day-ahead and intraday electricity generation schedules taking into account the uncertainty associated with both load and renewables production. The resulting large scale stochastic mixed-integer programming problems present serious computational challenges. We address these challenges using scenario-based decomposition techniques, in particular variants of progressive hedging, and modest parallel computing resources, achieving tractable run-times on moderate-scale instances. Our solver is embedded in a stochastic simulation environment, which is used to validate the model and to quantify cost savings relative to a standard deterministic unit commitment model. We describe experimental results on an ISO-NE test case, in addition to a smaller WECC-240 case. We also describe challenges and novel solutions to probabilistic scenario generation, required to represent the uncertainty associated with load and renewables production.