OLFWA: A Novel Fireworks Algorithm with New Explosion Operator and Two Stages Information Utilization

Abstract The Fireworks Algorithm (FWA) is a relatively new algorithm which has shown to be competitive with other intelligent optimization algorithms. In most of FWA variants, a simple random explosion is used to generate sparks, together with the guiding operator to study the promising search direction. However, this mechanism is inefficient to exploit the local landscape and to explore the promising search direction. In this paper, we firstly propose a novel explosion method based on orthogonal design (OD), which applies the orthogonal array to control the distribution of sparks and enables the sparks' information to be efficiently analysed. Then, a two-stage information utilization mechanism is designed. In the first stage, the orthogonal learning (OL) prediction operator is proposed to exploit the local optima based on the factor analysis of both objective function value and the ranking feature information. In the second stage, an enhanced guiding method is employed to determine the promising search direction. The new guiding operator not only uses an extra archive to collect the valuable information in the searching history, but also guides the guided point to a more reasonable position. Experimental results show that the proposed algorithm (OLFWA) achieves better performance than the state-of-the-art FWA variants, several famous OD-based algorithms and other promising algorithms. Furthermore, the effectiveness of each proposed strategy and the interactions among them are deeply analysed. Finally, it is proven that the proposed information utilization mechanism can be easily transplanted on other OD-based algorithms, bringing them a better optimization performance.