Efficient Distributed Parallel Algorithms Design And Implementation Of Inverse Problems On Cluser Of Computers

Along with the developments of science and technology, computing power of computing system becomes stronger and stronger, and the computing speed becomes faster and faster, however people's demands for high performance computing are increasing and infinite in some sense, so that in addition to enhancing the computing power of a processor, parallel processing is also an efficient way to enhance the computing power of a system. In the past, the parallel processing can only be performed on expensive supercomputers. As the cost of PCs and network descending, the distributed parallel computing concept is widely used in the parallel computing. This thesis focuses on determination of parameters in a two-dimension heat conduction equation on the MPI network parallel environment, by solving an inverse problem using the regularization method.Firstly this thesis introduce background and significances related with the subject, then describe basic theory of the parallel computing, the cluster concept, the genetic algorithms principle and the characteristic as well as the message passing system of MPI., after that, regularization method is introduced to solve inverse heat conduction problem, afterward, establish a parallel computing environment based on the MPI and Linux. According to parallel algorithm design principles of the network computing, and fully exploring the parallelizable characteristics of the genetic algorithms, an inverse heat conduction problem of the ceramic/metal combined material has been solved with a specially designed regularization method. Multiform noise disturbance exist in temperature survey process inevitable, all of noise disturbance will bring survey error, and this error will be magnified during the solving inverse heat conduction problem, hence the result is not useable. Regularization method is well known that introducing a regularity item in the heat conductivity equation could reduce solution sensitivity to measured temperature errors in solving the inverse problem;therefore it would accelerate the inverse problem solution. So a parallel genetic approach to solve the inverse problem byintroducing a regularization item in the equation has been explored in the thesis. The numerical results have been compared and analyzed. Finally, conclusions and suggestions for further research are given in the last part of the thesis. This thesis has six chapters.Chapter 1 introduces background and significances related with the subject and main works that have been done.Chapter 2 introduces development and classification of the parallel computers, discusses the basic theory of parallel computing and the theory of PGA.Chapter 3 introduces the MPI system;a detail configuration of MPI cluster system that this thesis has used is given.Chapter 4 introduces the regularization method theory. Its algorithm description is also given.Chapter 5 introduces the inverse problem model numerical solution and its regularization method, established the experimental environment of the inverse problem. The result is also analyzed and compared.Chapter 6 summarizes this thesis, and suggestions for future research are given.This subject is supported by National Natural Science Foundation of China (NSFC Grant No. 60173046).