Research On Parallel Genetic Algorithms Based On MPI

Genetic Algorithms (GAs) are effective tools in solving problems of theoretical computing and engineering optimization. But to some practical multi-variable optimization problems which have large search space, the GAs cannot find the approving results in acceptable time, though they are considered to be able to reach the global optimum theoretically. It can be solved in two ways, one is using parallel methodology;the other is reducing the search space. The thesis studies on how to accelerate GAs in solving problems with large search space, and the main contents are listed below:1. Two parallel genetic algorithms (PGAs) models, master-slave model and coarse grain model, are thoroughly discussed, and the implementations of the two models using Massage Passing Interface (MPI) are given. The two implementations are tested using standard genetic algorithms test functions. And a PGAs generator is developed based on the code refactoring of the two implementations. It can generate the PGAs codes according to the user's requirements.2. An improved genetic algorithm is introduced. This algorithm uses variable step-size algorithm to yield feasible solution and reduces search space at the same time. The theoretical analysis and experiments show that this method can enhance the search ability of genetic algorithms in solving problems with large search space. An application of the algorithm to solve the one dimension single-humped function optimization is included, and the application of this algorithm to multi-dimension problems is also discussed.3. The optimization of tokamak plasma equilibrium shape, which is considered as a problem with large search space, is solved using master-slave PGAs. Three practical applications of the program are given, which show that the program can solve the time-consuming problem in computing positions and currents of the poloidal field coils.