Abstract:

The general dynamic canonical descent(DCD) algorithm is able to avoid being trapped by local optimization with a limited speed, and does not assume the cost function to be differentiable if firstly given a appropriate space interval. The iterative format of the algorithm is discussed. The mutative DCD algorithm is based on the general DCD algorithm; the two algorithms are compared in numerical experiments and engineering application. The recults show that the mutative DCD algorithm can converge to the global best point with a few objective function evaluations; and a tradeoff between time and space complexities and the capabilities to locate the global optimum up to a certain precision with a good numerical stability and credible results are obtained.

Key words: mutative parameters, global optimization, numerical stability, objective function evaluation, iterative format, back-analysis