Effects of binary variables in mixed integer linear programming based unit commitment in large-scale electricity markets

Abstract

Mixed integer linear programming is one of the main approaches used to solve unit commitment problems. Due to the computational complexity of unit commitment problems, several researches remark the benefits of using less binary variables or relaxing them for the branch-and-cut algorithm. However, integrality constraints relaxation seems to be case dependent because there are many instances where applying it may not improve the computational burden. In addition, there is a lack of extensive numerical experiments evaluating the effects of the relaxation of binary variables in mixed integer linear programming based unit commitment. Therefore, the primary purpose of this work is to analyze the effects of binary variables and compare different relaxations, supported by extensive computational experiments. To accomplish this objective, two power systems are used for the numerical tests: the IEEE118 test system and a very large scale real system. The results suggest that a direct link between the relaxation of binary variables and computational burden cannot be easily assured in the general case. Therefore, relaxing binary variables should not be used as a general rule-of-practice to improve computational burden, at least, until each particular model is tested under different load scenarios and formulations to quantify the final effects of binary variables on the specific UC implementation. The secondary aim of this work is to give some preliminary insight into the reasons that could be supporting the binary relaxation in some UC instances.