Unit Commitment is a mix-integer optimization problem and has long been intractable for power system operators. The binary status and power generation of online units need to be determined simultaneously, while the system constraints are required to maintain at the same time. This paper proposes two binary teaching-learning based optimization methods to solve the unit commitment problem. The new methods are employed to solve different scales of 10–100 unit commitment problems with different spinning reserves. Numerical results show that the proposed new algorithms have superb performance in multiple cases studies.