Convergence analysis on an improved Mind Evolutionary Algorithm

As one of new type evolutionary computing methods, Mind Evolutionary Algorithm (MEA) processes evolutionary operations by using `similartax´ and `dissimilation´ operator. To improve the algorithmic convergence especially in the late evolutionary time, “information entropy” of populations is built to estimate the state of the evolution in a new self-adaptive MEA based on information entropy is proposed. In the self-adaptive strategy, population entropy computing module is built to estimate the region including global optimal solution. In this way, the exploring of the algorithm is more purposeful and sufficiently and the performance is improved. Then the variance of population during the evolution from the view of the functional analysis is studied. Interval sheath theorem is used to prove the global convergence of the new algorithm.