An Innovization-based Evolutionary Algorithm Framework for Large-Scale Practical Multi-Objective Optimization Problems

—Large-scale multi-objective optimization problems (LSMOPs) are challenging to solve for most optimization algorithms including standard and generic multi-objective evolution- ary algorithms (MOEAs) due to the ”curse of dimensionality.” Although there exist a few case studies involving customized evolutionary algorithms to solve large-scale problems, specific problem knowledge obtained through extensive experience of users or domain experts can facilitate a more efficient application. Innovization is the process of uncovering patterns that exist among good solutions. Any common pattern extracted from good solutions discovered during an optimization run can be used to modify candidate solutions in the form of a repair operator, but the key aspect is to strike a balance between the relevance of the pattern identified and the extent of their use in the repair operator, lest the learned patterns are properties of artificially-stuck solutions. Thus, the use of problem knowledge for the initial guidance and innovization-based repair of solutions during optimization in an online fashion can help find better solutions, faster. This paper proposes a customized MOEA frame- work which combines problem-specific knowledge and online innovization methods to solve two real-world LSMOPs – 879 and 1,479-variable truss structure design and 544-variable solid fuel rocket design. Four different repair operators are proposed that are suitable for uncovering monotonic relations involving multiple decision variables. The performance variations resulting from different combinations of initial user knowledge and repair operators have also been studied. The proposed approach is generic and provides a viable direction for handling large-scale multi-objective practical problems.