Parallel Power Flow Algorithm For Power System Based On Multi-core CPU Acceleration

As the scale of modern power grids continues to increase,the computational complexity of power flow calculation also increases.The performance of traditional serial power flow calculation algorithm has been difficult to meet the actual production requirements,so parallel computing technology is needed to speed up the power flow calculation.Multi-core CPU is a mainstream parallel computing hardware;this paper studies using multi-core CPU to speed up the Newton-Raphson power flow calculation.Since the main parts of the Newton-Raphson power flow calculation are jacobian matrix generation and trend revise equations solving,this paper focuses on the acceleration strategy of generating jacobian matrix and solving trend revise equations.The specific research contents are as follows:Firstly,the principles of designing parallel algorithm are studied and refined,then excavating the parallelism of sparse jacobian matrix generation algorithm and sparse left-looking LU decomposition algorithm,and proposed parallel sparse jacobian matrix generation algorithm and parallel sparse left-looking LU decomposition algorithm.Afterwards,implementing parallel sparse jacobian matrix generation algorithm and parallel sparse left-looking LU decomposition algorithm on multi-core CPU based on Posix Pthread parallel algorithm library.The algorithms research shows that the larger the grid scale,the more obvious the acceleration effect.On the 2.0 GHZ six-core CPU,these two algorithms start obtain acceleration compared with serial sparse jacobian matrix generation algorithm and serial KLU algorithm when the dimension is respectively about two thousand and five thousand.When the dimension is 12664,the acceleration ratio is respectively 4.61 and 1.79.Additionally,compared and analyzed domino and stratification parallel strategies on multi-core CPU.As a result,LU decomposition algorithm using domino method obtains acceleration on the condition of the equations' dimension is larger than five hundred.And when the equations' dimension is 12664,the acceleration ratio is 1.31.Lastly,the performance difference of accelerating power flow calculation between multicore CPU and GPU is compared and analyzed.The GPU parallel algorithm on K20C GPU adopts the same parallelism excavating strategy as the parallel algorithm of multi-core CPU.In CUDA7.0 programming environment,the GPU parallel Jacobian matrix generation'performance is good,but parallel trend revise equations' solving algorithm has not obtain acceleration compared with serial KLU algorithm and its peformance is worse than multi-core CPU parallel algorithm on the condition of the dimension is relatively large.