Directionally Driven Self-Regulating Particle Swarm Optimization algorithm

In this paper, an improved variant of the Self-Regulating Particle Swarm Optimization (SRPSO) algorithm is proposed that further enhances the performance of the basic SRPSO algorithm and is referred to as a Directionally Driven Self-Regulating Particle Swarm Optimization (DD-SRPSO) algorithm. In DD-SRPSO, we incorporate two new strategies, namely, a directional update strategy and a rotational invariant strategy. As in SRPSO, the best particle in DD-SRPSO uses the same self-regulated inertia weight update strategy. The poorly performing particles are grouped together to get directional updates from the group of elite particles. All the remaining particles are randomly selected to undergo either the SRPSO strategy of self-perception of the global search direction or the rotational invariant strategy to explore the rotation variance property of the search space. The performance of the DD-SRPSO algorithm is evaluated using the complex, shifted and rotated benchmark functions from CEC2013. These results are compared with seven current state-of-the-art PSO variants like Social Learning PSO (SLPSO), Comprehensive Learning PSO (CLPSO), SRPSO, etc. The results clearly indicate that the proposed learning strategies have significantly enhanced the performance of the basic SRPSO algorithm. The performance has also been compared with state-of-the-art evolutionary algorithms like Mean Variance Mapping Optimization (MVMO), Covariance Matrix Adaptation Evolution Strategy (CMA-ES) on the recently proposed numerically expensive optimzation CEC2015 benchmark functions whereby DD-SRPSO has provided competitive solutions. The results also indicate that the DD-SRPSO algorithm achieves a faster convergence and provides better solutions in a diverse set of problems with a 95% confidence level, thereby promising to be an effective optimization algorithm for real-world applications.