DYNAMIC MULTIOBJECTIVE OPTIMIZATION USING EVOLUTIONARY ALGORITHMS

Abstract

Multiobjective Optimization involves the optimization of two or more conflicting objectives simultaneously. There is no single solution to such problems, but multiple trade-off solutions. Evolutionary Algorithms are a good candidate to solve such problems as they can obtain multiple solutions in a single run. When the optimal solutions change with time, it results in a Dynamic Multiobjective Optimization problem. Many real-world problems involve multiple objectives which maybe conflicting, are dynamic in nature and affected by constraints. In this thesis, the issues of dynamicity and presence of constraints are addressed by providing some possible solutions. Firstly, a Dynamic Multiobjective Evolutionary Algorithm based on MOEA/D-DE (Multiobjective Evolutionary Algorithm based on Decomposition with Differential Evolution) using Kalman Filter predictions in decision space is proposed to solve DMOPs. Secondly, MOEA/D-DE assisted by a non-linear prediction method using Support Vector Regression predictions is also explored. Only a handful of algorithms have been proposed to solve constrained dynamic optimization problems. To address this issue, the Kalman Filter based prediction mechanism is combined with an adaptive threshold based constraint handling method to ensure solution feasibility while simultaneously tracking the time varying solutions.