Cross Entropy Method Meets Local Search for Continuous Optimization Problems

The effectiveness of cross entropy (CE) method has been investigated on both combinatorial and continuous optimization problems, though it lacks exploitative search to refine solutions. Hybrid with local search (LS) method can greatly improve the performance of evolutionary algorithm. This paper proposes a parameter-less framework combining CE with LS method. Four LS methods are chosen and four combination algorithms are obtained after combining them with the CE method. We first study the performance of the four combinations on a set of twenty eight mathematical functions including both unimodal and multimodal functions. CE hybrid with Powell’s method (CE-Pow) is identified as the most effective algorithm. Then the CE-Pow algorithm is applied to resolve proportional, integral, and derivative (PID) controller design problem and Lennard-Jones potential problem. Its performance has been verified by comparing with four state of the art evolutionary algorithms. Experimental results show that CE-Pow significantly outperforms other benchmark algorithms.