关键词: 禁忌搜索算法, 参数优化, 邻域, 多维正态分布, 收敛速度

Abstract: There are many shortages for the optimization of multidimensional continuous function, such as the low convergence speed and the poor accuracy. In order to improve the global optimization performance of multidimensional continuous function, this paper proposes the parameters optimization of the Modified Staged Continuous Tabu Search algorithm（MSCTS） based on multidimensional normal distribution. MSCTS is based on Staged Continuous Tabu Search algorithm（SCTS）, which includes three stages and has different starting points, selection range of neighborhoods and searching step size at different stages. To get the best accuracy and high convergence speed, the multidimensional normal distribution function is used to give the direction of neighborhoods selection. The results of experiments show that the reduction ratio of solution space in second stage has great impact on the accuracy and convergence speed, and the best reduction ratio is 0.5. At the same time, to improve the accuracy, the step size is shrinking at the third stage. The results of experiments show the best shrinking ratio is 0.1. Compared with SCTS algorithm, especially to solve multidimensional continuous function, MSCTS algorithm, after parameters optimization, can obtain nearly 5 times higher convergence speed and raises 10 orders of magnitude on the accuracy. Therefore, MSCTS algorithm is more suitable for solving multidimensional continuous function optimization.

Key words: Tabu search algorithm, parameter optimization, neighborhoods, multidimensional normal distribution, convergence speed