Abstract: Using the information of known sample points to construct simplex gradient and interpolation function, we proposed a local search algorithm based on the simplex gradient. The algorithm combined the effective sample point set Ω with a hybrid selection strategy, and improved the multistart clustering global optimization algorithm (GLOBAL). Experimental results show that the new algorithm has great improvement in efficiency and stability, and can effectively deal with the problem of high evaluation of the original algorithm for the “narrow valley” function.

Key words: derivativefree; single link clustering, simplex gradient,  global algorithm