几种优化算法在浅海匹配场反演中的性能比较-声学技术.pdf

24 1 Vo.l 24, No. 1 2005 3 TechnicalA coustics M ar. , 2005 邹士新, 杨坤德, 马远良 (, , 710072) : , , , , , : ; ; ; : 0422. 2 : A : 1000-3 30( 2005) 01-0004-0 Performance comparison of optim ization algorithms in matched field inversion ZOU Shi-xin, YANG Kun-de, MA Yuan-liang ( Collag e of M arine, N or thwestern P o ly technical University, X i. A n, 710072, Ch ina) Abstract: Optmi ization efficiencies and mechan isms of smi u lated annealing, genetic algorithm, d ifferen- tial evolution and dow nhill smi plex d ifferential evolution are compared and analyzed. Smi u lated annealing and genetic algorithm use a direct random process to search the param eter space for an optmi al solution. They include the ab ility to avoid localm in mi a, but as no grad ient inform ation is used, searches are rela- tively inefficient. D ifferential evolution searches the parameter space by using d istance and azmi uth be- tw een individuals of a population, although initial searches are effective, the search speed decreases quickly because differential information betw een the individuals gradually van ishes. Local dow nhill smi - p lex and global d ifferential evolution methods are developed separately, and comb ined to produce a hy- brid downh ill smi plex d ifferential evolution algorithm. The hybrid algorithm is sensitive to grad ients of the object function and search of the parameter space is effective. These algorithm s are applied to matched field inversionw ith synthetic data. Optmi alparameter values, final values of object function and inversion tmi e are presented and compared. K ey words: