Path-extrapolation in Evolutionary Algorithms

Evolutionary Algorithms are used to solve challenging optimization problems across a variety of domains. While simple and robust they often do not effectively exploit information generated during the search which in turn degrades their efficiency and often results in a slow convergence. As such this paper presents a new algorithm to accelerate the EA convergence by monitoring the path traversed by its population over several generations. This information is then used to construct interpolating polynomials which predict the position of the centroid in the next generation and is then used to shift the population along that direction to accelerate convergence. An extensive numerical performance analysis shows the effectiveness of the proposed approach.