Parallelization Research Of Thermal Recovery Simulation Based On PETSc

Numerical reservoir simulation is one of the effective tools for complex fluid analysis in the oil industry. Because there is only once exploitation for hydrocarbon-bearing pool in the real world, the exploitation plan choosing is extremely important for the engineer. Using numerical reservoir simulation technology could repeat simulate the process of different exploitation plan. It also could provide fast and accurate method for dynamic analysis and comparative result for different exploitation plan.Currently, China's old oilfields have come into the high water-cut stage, requiring fine reservoir simulation to analyze the distribution of the remaining oils. So, the parallelization of thermal recovery numerical simulation is extremely important. Our country starts earlier in the research of numerical reservoir simulation. In the 90th, with the development of research work aboard, China began to research the parallelization software of numerical reservoir simulation. This paper uses the PETSc toolkit to design and implement the parallel software of thermal recovery simulation which is based on serial software. Main tasks include:(1) Because of using domain decomposition method, the parallel program realizes load-balance based on active blocks which could improve the balance of each processor. On the basis of this, the parallel program realizes automate domain decomposition in order to improve the practicality of parallel program.(2) PETSc toolkit is used to design and implement parallel solver for sparse linear equation. This paper compares various solutions and finds the suitable one for the linear equation in thermal recovery simulation.(3) The parallel program also optimizes the data communication. Then it realizes the parallel reading of well group data to improve the efficiency of parallel software.(4) Based on serial software and using MPI message interface, this paper designs and implements the parallel software of the thermal recovery simulation. The parallel program as exposition in this paper is tested on several reservoir numerical models. Results show that the parallel program could acquire higher speedup. Compared with the serial program, our implementation can gain 9.42times speedup using 16 cores over a large-scale reservoir test case a2.dat (96*127*10) .