Auto-tuning fuzzy granulation for evolutionary optimization

Much of the computational complexity in employing evolutionary algorithms as optimization tool is due to the fitness function evaluation that may either not exist or be computationally very expensive. With the proposed approach, the expensive fitness evaluation step is replaced by an approximate model. An intelligent guided technique via an adaptive fuzzy similarity analysis for fitness granulation is used to decide on use of expensive function evaluation and dynamically adapt the predicted model. In order to avoid tuning parameters in this approach, a fuzzy supervisor as auto-tuning algorithm is employed with three inputs. The proposed method is then applied to three traditional optimization benchmarks with four different choices for the dimensionality of the search apace. Effect of number of granules on rate of convergence is also studied. In comparison with standard application of evolutionary algorithms, statistical analysis confirms that the proposed approach demonstrates an ability to reduce the computational complexity of the design problem without sacrificing performance. Furthermore, the auto-tuning of the fuzzy supervisory removes the need for exact parameter determination.