Local learning strategy

To improve the performance of the proposed algorithm, especially self-learning capability of elite particle in the swarm, a local research strategy based on differential learning is designed to explore the areas with sparse solutions in search space. In this strategy, first a solution with big crowding distance in the archive is selected as a base vector, notified X _best, in differential learning. Second, two random solutions from the archive, notified X _n1 and X _n2, are set as differential vectors. Then, a new solution is generated by adding the difference between X _n1 and X _n2 to the base vector X _best:

This loop is implemented repeatedly until generating N′ new solutions. Finally, the N′ new solutions are saved into the archive. The parameter F is a scale factor that amplifies the difference between the two vectors. This paper sets F to be a random value within [0.1, 0.9] in order to improve the diversity of new solutions. Since the base vector often locates to good promising area, the local research strategy is competent for exploiting the area including sparse solutions.