A New Filled Function Method for the Non-convex Global Optimization

Abstract

The conundrum of the non-convex global optimization is that there are multiple local minima which are not global optimal solution, and conventional algorithms drop into local optimum easily. Filled function method is an available way which generally invokes an auxiliary function to move successively from a local minimizer to another better one. Definition-based on filled function, a new filled function with easy- adjustment parameter, simple form and non- exponential is proposed, and then prove the filled function can maintain the padding properties. Moreover, a local search method with stochastic and uniformity strategy is designed to strengthen the local search. Based on the above, a new filled function algorithm is presented. The numerical results indicate the proposed algorithm is feasible and effective, it follows that the stochastic and uniform strategy design is valid, further, the analysis and comparison of numerical experiments manifest high-efficiency, good stability and easy-realization.