Automated objective transformation for improved MOGA-II workflows
When objectives differ significantly in magnitude or range of values within the criterion space, multi-objective optimization often struggles to generate well-distributed Pareto solutions. Such discrepancies between objectives can distort search behaviour, bias utility-based formulations, and limit systematic exploration of the criterion space. Objective transformation can mitigate these issues by systematically reshaping objective functions such that they contribute comparably to the search process. Our work is based on the transformation principles that have been studied in [1] with regard to the weighted sum utility function. We discover that also without using a utility function, the multi-objective search of genetic algorithms like MOGA-II can benefit from the objective transformation. Our work presents an automated objective transformation workflow within modeFRONTIER. The engineering problem we solve is a representative one from high-voltage circuit breaker design, chosen for its characteristic multi-objective tradeoffs.
In our workflow, the transformation is applied based on the initial sampling results. The resulting transformed objective landscape is then used for full optimization. By embedding this systematic transformation step in the workflow, we reduce manual parameter tuning and enhance reproducibility across diverse objective functions, ensuring that each objective influences the search for Pareto-optimal solutions in a balanced manner.
To quantify the effect on Pareto front quality, we combine nearest neighbour spacing metrics (mean and standard deviation) with hypervolume evaluation, providing complementary measures of distribution uniformity and convergence.
Our results will demonstrate the potential advantage of automated objective transformation for making design decisions based on criterion space exploration and exploitation, and systematic comparison of trade-offs. We expect that a broad class of optimization algorithms may benefit in effectiveness from this approach.
[1] Marler, R. T., & Arora, J. S. (2005).
Function‑transformation methods for multi‑objective optimization.
Engineering Optimization, 37(6), 551–570.
