A Hybrid Multiobjective Optimization Approach for Dynamic Problems: Evolutionary Algorithm Using Hypervolume Indicator

Abstract

Real world problems are often dynamic and involve multiple objectives and/or constraints varying over time. Thus, dynamic multiobjective optimization algorithms are required to continuously track the moving of Pareto Front (PF) after any change. Recently, evolutionary algorithms have successfully solved dynamic single objective problems. However, few works have focused on multiobjective case through the common “detector technique” to detect/predict change in the fitness landscape, which is sometimes hard or impossible. In this paper, we propose a hybrid approach to tackle dynamic multiobjective optimization problems with undetectable changes. In this hybrid approach, a new local search and novel techniques of population selection and maintain diversity are combined within multiobjective evolutionary algorithm. Our proposed approach (Dynamic HV-MOEA) uses a hypervolume indicator as a performance metric in the local search and population selection techniques in order to accelerate convergence speed towards the Pareto Front. The population diversity is maintained through an efficient mutation strategy. The performances of our hybrid approach are assessed on various benchmark problems. Experimental results show the efficiency and the outperformance of Dynamic HV-MOEA in tracking pareto front and maintaining diversity in comparison with existing dynamic algorithms.