Performance Analysis of Mixed-Integer Conic and Mixed-Integer Linear Unit Commitment Models

Computational tractability and scalability are general concerns of Unit Commitment (UC) formulations given the inherent non-convex nature of the problem. Mixed-integer linear programming (MILP) version of UC is very common in modern Energy Management Systems. Lately, mixed-integer second order cone programming (MISOCP) versions of UC are also gaining research attention. To this end, this paper presents a comparative analysis of MILP and MISOCP based UC formulations with and without network constraints. Extensive numerical simulations are performed to investigate accuracy and scalability of MISOCP and MILP UC formulations with several test cases up to 1,000 generating units. Results show that the MISOCP UC model is generally superior to its MILP counterpart in terms of costs. On the computational time, MISOCP performed superior compared to MILP for large power systems. However, for small systems, MILP UC model performed very similar to MISOCP version in terms of the computational time.