![]() |
Fusion of Neural Networks, Fuzzy Systems and Genetic Algorithms: Industrial Applications
by Lakhmi C. Jain; N.M. Martin CRC Press, CRC Press LLC ISBN: 0849398045 Pub Date: 11/01/98 |
Previous | Table of Contents | Next |
The set of strings in a generation is referred as a population. The population size could change, increase or decrease, from generation to generation, but, in most applications, it is taken to be constant. The population size is a function of string size and it should be big enough to avoid the incest effect in the reproduction process. If diversity of strings is small, there are few chances to change (improve) species using crossover operator. The diversity could be improved in this case by intensifying mutation, but, on the other hand, by increasing mutation probability, searching tends to be random. Generation gap is referred as the fraction of a population, in interval (0, 1), which takes part in reproduction procedure creating a new generation.
In each GA application, one should define how to generate the initial population and how to stop algorithm running. The initial population could be generated at random, or as genetic material from some previous procedure.
The termination of GA could be done simply by counting if some prescribed number of steps is reached, or by testing, if a termination criterion is fulfilled. An autonomous stopping could be done by fitness function convergence testing or by homogeneity checking of an entire population. If fitness function reaches global or some of local optima, then all strings, because of preferring the best solution, tend to be equivalent. In this case, a high degree of homogeneity could stop the procedure, or adapt GA parameters in order to move searching of the solution to other areas of local optima as candidates for the global one. In this case, GA is considered to have adaptivity features.
A GA application to a specific problem includes a number of steps. Some of them are listed below:
Three examples discussing GA applications in telecommunication system design follow.
Call is a generic term related to the establishment, utilization, and the release of a connection between a calling user and the called user for the purpose of exchanging information. The call is also defined as an association between two or more users, or between a user and a network established by using network capabilities. Service is offered by a network to its users, in order to satisfy a specific telecommunication requirement.
Users initiate calls and services. Each call/service causes a number of processing requests. The call/service can be represented by a sequence of requests occurring in random intervals. The first request initiates the call, while further requests start different communication and information operations named call phases. On this level of abstraction, unsuccessful calls shorten the call because some phases are skipped. The services shorten, modify (some phases are replaced by others), or enlarge the call (new phases are inserted), so the number of requests varies from case to case. The mean number of requests depends on implemented set of services as well as on traffic and other conditions in the network.
The call/service decomposition on phases depends on the interaction with the environment; information from the environment (request input data) is needed to start a specific call phase, and/or information (result) is sent to the environment after completing a call phase. Different types of calls and services represent different types of independent, cooperating, or mutually excluding tasks to be run on the network.
It should be pointed out that, when discussing call and service handling, control plane of the telecommunication network must be taken into consideration. Considering the aspect of control, the system is reactive. After receiving a request, and reflecting defined time conditions, a response has to be produced; otherwise, a call/service will be lost.
When discussing the decomposition of calls and services, the following features have to be taken into account:
There are different levels of parallelism to be considered in order to organize an efficient processing system:
The first two levels are related to the processing capacity, i.e., they influence the number of calls/services to be handled. The third level can improve the call/service processing time and, only in this case, a minimum processing time can be obtained. A parallel system should include all three levels [2, 3].
Previous | Table of Contents | Next |