An Improved Cross-Entropy Method Applied to Inverse Problems

An improved cross-entropy method for global optimizations of inverse problems with continuous variables is proposed. To enhance the convergence speed, improvements on both algorithm development and iterative process are introduced. To monitor and guide the searching process, the design space is divided into subdomains and three indicators are assigned for each subdomain in order to evaluate its performances. To balance exploitation and exploration searches, the whole iterative process is divided a diversification and an intensification phase. In the diversification phase, a novel mechanism is introduced to increase the sampling diversity to avoid the solution being trapped onto a local optimum; in the intensification phase, the strategy of shifting away from the worst subdomains equips the algorithm with enhanced convergence rates. The proposed method is applied to a mathematical function and the TEAM Workshop problem 22. Comparisons with its counterparts are made to demonstrate the effectiveness of the proposed work.