Local Fitness Landscape from Paired Comparison-Based Memetic Search in Interactive Differential Evolution and Differential Evolution

We propose a triple comparison-based interactive differential evolution (IDE)algorithm and differential evolution (DE) algorithm. The comparison of target vector andtrial vector supports a local fitness landscape for IDE and DE algorithms to conduct amemetic search. In addition to the target vector and trial vector used in canonical IDEand DE algorithm frameworks, we conduct a memetic search around whichever vectorhas better fitness. We use a random number from a normal distribution generator or auniform distribution generator to perturb the vector, thereby generating a third vector. Bycomparing the target vector, the trial vector, and the third vector, we implement a triplecomparison mechanism in IDE and DE algorithms. A Gaussian mixture model is used asa pseudo-IDE user for evaluating the IDE and 25 benchmark functions from the CEC2005test suite are employed to evaluate the DE. We compare our proposals with canonicalIDE and triple comparison-based IDE implemented by opposite-based learning and applyseveral statistical tests to investigate the significance of our proposed algorithms. We alsocompare our proposals with several evaluation metrics, such as number of function calls,success rate and acceleration rate. Our proposed triple comparison-based IDE and DEalgorithms show significantly better optimization performance arising from the evaluationresults. We also investigate potential issues arising from our proposal and discuss someopen topics and future opportunities