Research on Optimizeation of Implementing Telephone Netwarks by Using Learning Theory

点击翻译按钮获取中文摘要

In 1992, by applying our previously presented learning control scheme to a telephone network we designed and presented a a routing scheme with aim at load equalization on the network. This scheme chooses a route which minimizeds a weighted blocking frequency from an appropriately determined set of routes. It was shown that this scheme achieves the load equalization in the sense that the blocking frequencies asymptotically become equal amoug the set of routes. According to computer simulation, it was ascertained that our scheme is superior in blocking count and average route length to the scheme using learning automata. However, it was found that the characteristic of load equalization greatly depends on determination of the set of routes.Viewing recent rapid development of functions of network observation and control, we can expect that higher performance routing than that considered in 1992 is realizable. From this, in 1993, we presented a learning routing scheme which achieves more strict load equalization based on information about call generation at all stations. This scheme seeks blocking equalization among links by adaptively assigning overflow calls from main sources generating congestion to vacant routes. According to detailed computer simulation, it was ascertained that our new scheme is most superior compared with our previous one and the state-dependent scheme under various conditions in the case that we determined the set of routes using the design method of a routing domain presented by other investigators.Significance of this research is that it was shown theoretically and practically that our schemes achieve load equalization which is not ensured by the present schemes. Our research results will be useful for development of available routing schemes in future.