| Summary: | The initialization stage of a Soft Computing (SC) algorithm is vital as it affects the success rate of algorithms in solving multi-peak global optimization problems. The individuals in an initial population, which are known as search agents, are often generated randomly using pseudo-random number generator (PRNG) due to unavailability of prior information on the location of global peak (GP). The random nature of the generated search agents causes uneven distribution of the initial population over the search space (SS), which may lead the search towards unpromising regions from the very beginning. This paper proposes a new deterministic initialization method (DIM) for SC algorithms where search agents are evenly fixed in the SS by using a simple deterministic formulation. The performance of the proposed DIM is then compared to the conventional PRNG and more recent quasi-random number generator (QRNG). An optimization case study is carried out using two popular SC algorithms which are the Particle Swarm Algorithm (PSO) and the Evolutionary Programming (EP), and three relatively new SC algorithms which are the Whale Optimization Algorithm (WOA), the Elephant Herding Optimization (EHO), and the Butterfly Optimization Algorithm (BOA). The optimization is done on various one-dimensional (1D) benchmark functions, as well as practical problems such as partial shading condition (PSC). Simulation results show that the proposed DIM successfully improved the performance of each SC algorithm under study in solving almost all tested functions with 99% success rate compared to 88.7% and 80.2% for the QRNG and PRNG approaches respectively. Furthermore, the WOA is the most reliable and robust among the five SC techniques under study with a success rate of 93.8% when all initialization methods are employed. Copyright © 2020 Institute of Advanced Engineering and Science. All rights reserved.
|
|---|