Parallel computation for reservoir thermal simulation: An overlapping domain decomposition approach

In this dissertation, we are involved in parallel computing for the thermal simulation of multicomponent, multiphase fluid flow in petroleum reservoirs. We report the development and applications of such a simulator.; Unlike many efforts made to parallelize locally the solver of a linear equations system which affects the performance the most, this research takes a global parallelization strategy by decomposing the computational domain into smaller subdomains. This dissertation addresses the domain decomposition techniques and, based on the comparison, adopts an overlapping domain decomposition method. This global parallelization method hands over each subdomain to a single processor of the parallel computer to process. Communication is required when handling overlapping regions between subdomains. For this purpose, MPI (message passing interface) is used for data communication and communication control.; A physical and mathematical model is introduced for the reservoir thermal simulation. Numerical tests on two sets of industrial data of practical oilfields indicate that this model and the parallel implementation match the history data accurately. Therefore, we expect to use both the model and the parallel code to predict oil production and guide the design, implementation and real-time fine tuning of new well operating schemes.; A new adaptive mechanism to synchronize processes on different processors has been introduced, which not only ensures the computational accuracy but also improves the time performance.; To accelerate the convergence rate of iterative solution of the large linear equations systems derived from the discretization of governing equations of our physical and mathematical model in space and time, we adopt the ORTHOMIN method in conjunction with an incomplete LU factorization preconditioning technique. Important improvements have been made in both ORTHOMIN method and incomplete LU factorization in order to enhance time performance without affecting the convergence rate of iterative solution.; More importantly, the parallel implementation may serve as a working platform for any further research, for example, building and testing new physical and mathematical models, developing and testing new solver of pertinent linear equations system, etc.