Parallel genetic algorithms with schema migration

Several main factors that affect the performances of several kinds of classic parallel genetic algorithms are analyzed in this paper to find some changes for improvements. On the basis of previous researches, authors take the decrease of the communication costs as the key to this problem and present a new migration scheme based on schema theory. This schema migration scheme (SMS) uses the transmission mechanism of information in network for reference. Through the methods of module identification, it distills and compresses the good genetic information from one sub-population and then spreads proportionately these schemes into another sub-population. The validity of this new migration scheme is discussed according to schema theory. The formalized measurements of communication costs decrease and the analysis on better scalability of the algorithms by use of SMS are proposed.