Parallel Algorithm for Transient Stability Simulation Based on GMRES Method Preconditioned by Hybrid LU Decomposition
|更新时间:2025-12-08
|
Parallel Algorithm for Transient Stability Simulation Based on GMRES Method Preconditioned by Hybrid LU Decomposition
Vol. 43, Issue 10, Pages: 3419-3426(2017)
作者机构:
1. 广东电网有限责任公司
2. 广东电网有限责任公司电力科学研究院
3. 中国人民解放军国防信息学院
4. 武汉大学电气工程学院
作者简介:
基金信息:
DOI:
CLC:
Published Online:30 October 2017,
Published:2017
稿件说明:
移动端阅览
DOU Peng, MA Ming, LIU Dong, et al. Parallel Algorithm for Transient Stability Simulation Based on GMRES Method Preconditioned by Hybrid LU Decomposition[J]. 2017, 43(10): 3419-3426.
DOI:
DOU Peng, MA Ming, LIU Dong, et al. Parallel Algorithm for Transient Stability Simulation Based on GMRES Method Preconditioned by Hybrid LU Decomposition[J]. 2017, 43(10): 3419-3426.DOI:
Parallel Algorithm for Transient Stability Simulation Based on GMRES Method Preconditioned by Hybrid LU Decomposition
The commonly used preconditioning technique is incomplete LU(ILU) decomposition
which is not suitable for parallelization due to the inherent model of forward and back substitutions.Therefore
the parallel computation method of power system transient stability based on preconditioned generalized minimal residual algorithm(GMRES) method with hybrid LU(HLU) decomposition is derived.On the basis of the implicit simultaneous solution method
and combined with the block bordered diagonal form
the Jacobian matrix is preconditioned with hybrid LU decomposition by the technique of dividing and ruling
and then the GMRES method is used to solve the linear equations.Using two different scale systems
the test results show that the iterations of HLU decomposition preconditioning method can almost reduce to the half of ILU decomposition preconditioning method
and with graphics processing unit(GPU) parallel technology
the preconditioned GMRES method with hybrid LU decomposition can obtain more than 10 times speedup when the system scale is up to tens of thousands of nodes.Therefore
the proposed method has not only good convergence rate but also high spatial parallelism