Using fuzzy logic to tune an evolutionary algorithm for dynamic optimization of chemical processes

Dynamic optimization of chemical processes can be carried out with evolutionary algorithms that involve many parameters. These parameters need to be given appropriate values for the algorithms to perform efficiently. This paper proposes parameter setting methods based on factorial experimentation and fuzzy logic, aimed at balancing convergence speed, robustness (consistent performance for each problem) and versatility (applicability to many different problems). The methods were tested on an existing dynamic optimisation method with at least nine tuneable parameters. The test problem set turned out to be quite demanding due to one particular problem behaving in opposite direction to the rest with respect to the most influential factor, population size. It is probable that no single tuning would be possible that will satisfy all problems. However, for the other problems, the Fuzzy Logic tuning method proposed in this paper proves to be a very promising approach.

► An evolutionary optimization method or dynamic optimization was tuned by a new method to balance speed, robustness and versatility.
► We carried out a factorial experiment where each reproduction parameter was turned on or off.
► From the experimental results, we derived optimal parameter values for the evolutionary algorithm by using fuzzy logic.
► The approach is suitable for quick tuning of algorithms with large number of parameters, where more rigorous methods may be too time consuming.