Enhancing State-of-the-Art Multi-Objective Optimization Algorithms by Applying Domain Specific Operators

Abstract

To solve dynamic multi-optimization problems, optimization algorithms are required to converge quickly in response to changes in the environment without reducing the diversity of the found solutions. Most Multi-Objective Evolutionary Algorithms (MOEAs) are designed to solve static multi-objective optimization problems where the environment does not change dynamically. For that reason, the requirement for convergence in static optimization problems is not as time-critical as for dynamic optimization problems. Most MOEAs use generic variables and operators that scale to static multi-objective optimization problems. Problems emerge when the algorithms can not converge fast enough, due to scalability issues introduced by using too generic operators. This paper presents an evolutionary algorithm CONTROLEUM-GA that uses domain specific variables and operators to solve a real dynamic greenhouse climate control problem. The domain specific operators only encode existing knowledge about the environment. A comprehensive comparative study is provided to evaluate the results of applying the CONTROLEUM-GA compared to NSGAII, ∊-NSGAII and ∊-MOEA. Experimental results demonstrate clear improvements in convergence time without compromising the quality of the found solutions compared to other state-of-art algorithms.