Study On Transient Simulation Of Gas Pipeline Network And Its Application Based On CPU+GPU Heterogeneous Computing

In recent years,gas has an increasing proportion in energy structure of our country,and the gas pipelines are now in the high tide of construction.Besides,there is a significant trend that most of gas pipeline networks will be connected as a whole system,so it is obvious that the scale of each independent gas pipeline network will be much larger in the near future.As an important tool for technology design and operation analysis of gas pipeline network,numerical simulation will play a significant role in the future network system.Currently,most commercial pipeline simulation softwares are still facing low efficiency for large pipeline network simulation.In this paper,we design a parallel algorithm and develop a program for pipeline network simulation based on CPU+GPU heterogeneous computing,which is a rising acceleration technology in HPC(High Performance Computing).Compared with famous commercial software SPS,our program get a equivalent accuracy and a desire speedup ratio up to 57.57 x.In our study,we firstly develop the DIMENS(Decoupled Implicit Method for Efficient Network Simulation)method based on the features of topological structure of pipeline network and linear equation system.DIMENS realizes a course level computation parallelization of each component by decoupling them with each other,and some other parallel methods are introduced in fine level parallelization to make a further acceleration.This two-level parallel method is easy to realize and fully exploits parallelism of the problem.Next,according to the CPU+GPU heterogeneous platform features,we study the parallel methods concretely for solving block tridiagonal equations of pipeline pre-solution and linear system of network connection equations.By comparing the efficiency of RD and PCR algorithm for solving block tridiagonal equations on GPU,and discussing the parallel model of PCG on GPU,we choose PCR+PCG(cu SPARSE)as the method to implement fine level parallelization.The method we propose realize rational use of computational resources and can make the best of GPU computing power.Finally,we apply our program in a real pipeline network and test its accuracy and stability by validating results under steady and transient working condition,and the result is great.Then,we make 5 large pipeline network experiments to find out the rule of the acceleration.The result shows a speedup ratio up to 57.57 obtained compared with commercial software SPS.Our study provides a more efficiency solution for future regional and even continental gas pipeline network simulation,and it also provides an important reference for pipeline construction and operation.