Global optimization: an auxiliary cost function approach

An efficient and practical solution to a class of global function optimization is proposed. The algorithm consists of a stochastic search of initial guesses and a gradient-based solution-finding algorithm. The key idea is to introduce an auxiliary cost function that can indicate whether the gradient-based solution-finding process goes toward a global minimum of the cost function and that helps us to prevent the process from going to local minima. Simulation examples are used to show the mechanism, power, and restrictions of the approach.