The paper presents a new approach to the large-scale unit-commitment problem. To reduce computation time and to satisfy the minimum up/down-time constraint easily, a group of units having analogous characteristics is clustered. Then, this 'clustered compress' problem is solved by means of a genetic algorithm. Besides, problem-oriented powerful tools such as relaxed-pruned ELD, intelligent mutation, shift operator etc. make the proposed approach more effective with respect to both cost and execution time. The proposed algorithm is tested using the reported problem data set. Simulation results for systems of up to 100-unit are compared with previous reported results. Numerical results show an improvement in the solution cost compared with the results obtained from a genetic algorithm with standard operations.