Differential Evolution with Two Subpopulations

In this paper, differential evolution with two subpopulations is proposed for balancing exploration and exploitation capabilities. The first population is responsible for exploring over the search space to find good regions using only its own subpopulation. The second subpopulation is responsible for exploiting good regions. The exploitation-oriented sub-population is permitted to make use of the whole population to select best solution candidates to generate offspring. Hence, this heterogeneous one-way information transfer allows the exploration subpopulation to maintain diversity even when exploitation group converges. This is an efficient realization of population based algorithm enabling simultaneous use of highly exploitative and explorative characteristics simultaneously. Hence, this approach can be an effective substitute for memetic algorithms in the real-parameter optimization domain. The performance of the algorithm is evaluated using the shifted and rotated benchmark problems. To verify the performance of the proposed algorithm, it is also applied to solve the unit commitment problem by considering 10 and 20 unit power systems over 24 h scheduling period.