A study of the effciency of different decomposition methods on the network-constraint unit commitment problem with stochastic d

Specialeforsvar: Magnus Rimer

Titel: - A study of the effciency of different decomposition methods on the network-constraint unit commitment problem with
stochastic demand

Abstract: In this theses we study we test how different methods preform on the Network constrained unit commitment problem with stochastic demand. The algorithms that we tested was Benders Decomposition, L-Shaped method and two variations of L-shaped method. These where tested against each other and against IBMs Optimizations engine Cplex. The way they where testes that each method was implemented in Java by using the callback function and was tested on the same instances of the problem. The result of this testing was that the amount of scenarios used where the variable that had the biggest impact on the different methods relative performance. For small amount of scenarios Cplex was preforming strictly better than any of the other 4 methods. But with higher number of scenarios the two variations of the L-shape method did as good as Cplex or better. This thesis illustrate how impactful the choice of solution-method can be.

Vejleder: Giovanni Pantuso
Censor:   Konstantin Pavlikov, SDU