Nonlinear Slover Parallelization For Reservoir Simulation Based On Clusters

Reservoir simulation is an effective way to analyze the complicated fluidunderground in petroleum engineering. As many oilfields have entered thelater phase of exploitation, the requirement of efficient large scale refinedreservoir simulation make the workload of computation in reservoirsimulation increase enormously. On the other hand, clusters have becomethe mainstream of parallel systems and have been used widely. So it'snecessary to research the parallel algorithms and develop parallel softwarefor reservoir simulation based on clusters. Focusing on the parallelalgorithms for reservoir simulation and the implementation techniques onclusters, this thesis has three achievements as follows:(1) The thesis researched the parallel algorithm for non-linear equationsrising from reservoir simulation and proposed a hybrid quasi-Newtonalgorithm. Based on the domain decomposition method and thequasi-Newton method, as well as according to the specified problem, hybridquasi-Newton algorithm took several measures to accelerate the convergencerate of quasi-Newton method, such as accelerating pressure on border grids,estimating the initial approximate pressure, and overlapped additive Schwarzmethod.(2)This thesis designed and adopted several techniques in theparallelization. It optimized the data structures and the module of processcommunication. Flexible 2-Dimension domain decomposition on horizontaldirection can be accomplished by using a domain information file. The datafield can be divided into each sub-domain according a serial of indice.(3) On the basis of the widely used software called SIMBESTII, thisthesis designed and implemented the parallel software for homogeneous blackoil problems, named TH_PSIM2 v1.0. The parallel software built thefoundation for further optimization.The parallel algorithm and software are tested on several numericalmodels on Itanium 2 clusters connecting with 1G Ethernet. The resultsshow that hybrid quasi-Newton algorithm and the optimization techniques caneffectively accelerate the convergence rate of the algorithm. To simulate ahomogeneous black oil reservoir with more than 1 000 000 grids in 10 years,the total running time of the parallel software is 1.08 hours. This canbasically meet the practical requirement in reservoir engineering.