Research On The Parallel Algorithms Of Discharge Simulation In SF₆/N₂ Gas Mixture Based On PC Cluster

SF6 gas is widely used in electric power industry for gas-insulated equipments as an important quenching and insulated medium, due to its excellent chemical and physical characteristics. However, SF6 is one of the six greenhouse gas listed in the Kyoto Protocol and its application should be limited. One of the proposed solutions to this problem is to replace SF6 by SF6 gas mixtures. As an ideal substitute gas, SF6/N2 gas mixture has good prospects for development with its insulation characteristics close to the pure SF6 gas. Thus, the research of simulating the insulator surface discharge in SF6/N2 gas'mixtures has become a pressing need that should be resolved for scientists and the electricity sectors.In the research, most of the researchers use motion equations of plasma particles to simulate the discharge progress in the SF6/N2 gas mixtures. In the simulation program, we track the motion process of all the charged particles under the action of external electric field. Therefore, in a large number of particles, we have to calculate the interaction and the resulting movement between two particles within each time step, producing very impressive calculation expense. This leads to almost endless demand of computer capacity and speed.To solve large-scale problems under limited computer resource, designing parallel discharge simulation algorithms in a cluster of PCs is a low-cost and high efficiency method. Consequently, parallelization approaches based on PC cluster were used in our simulation to-reduce the execution expense:(i) splitting by the charged particles using MPI, (ii) splitting by static physical domain using MPI, (iii) splitting by both domain and charged particles using hybrid MPI and OpenMP. However, all the three algorithms have the bottlenecks which affect the parallelization efficiency seriously, such as load-imbalance. Thus, a load-balance module based on the hybrid-level algorithm is added to obtain much better parallelization efficiency in the forth method.In the end, this paper establishes a Linux cluster composed of 10 dual-core PC computers to test all the parallelization algorithms in 10%-90% SF6/N2 mixture gas, not only compares the difference of time-consuming among parallel algorithms and the sequential algorithm, but also compares the performance among the four parallel algorithms. The simulation results show that our proposed parallelization algorithms have good feasibility and superiority to reduce the execution time.