Flower Pollination Algorithm (FPA) has increasingly attracted researchers' attention in the computational intelligence field. This is due to its simplicity and efficiency in searching for global optimality of many optimization problems. However, there is a possibility to enhance its search performance further. This paper aspires to develop a new FPA variant that aims to improve the convergence rate and solution quality, which will be called modified global FPA (mgFPA). The mgFPA is designed to better utilize features of existing solutions through extracting its characteristics, and direct the exploration process towards specific search areas. Several continuous optimization problems were used to investigate the positive impact of the proposed algorithm. The eligibility of mgFPA was also validated on real optimization problems, where it trains artificial neural networks to perform pattern classification. Computational results show that the proposed algorithm provides satisfactory performance in terms of finding better solutions compared to six state-of-the-art optimization algorithms that had been used for benchmarking.
Keywords: Artificial neural networks; Computational intelligent; Exploration; Flower Pollination Algorithm; Optimization problems.