Parallel Power Flow on Graphics Processing Units for Concurrent Evaluation of Many Networks

Optimal power flow based on parallel metaheuristics for graphics processing units

“…The PF software module used in the proposed solver is parallelized using the approach we have previously published in [35]. Unlike other implementations on GPU which are limited to using dense matrices such as in [36][37][38], the method we published in [35] uses only sparse matrices and as a result reduces the order of complexity of the calculations from O(n 2 ) to O(n) where n is the number of buses in the network.…”

“…Unlike other implementations on GPU which are limited to using dense matrices such as in [36][37][38], the method we published in [35] uses only sparse matrices and as a result reduces the order of complexity of the calculations from O(n 2 ) to O(n) where n is the number of buses in the network. This reduction of complexity results is a significant performance improvement.…”

“…This reduction of complexity results is a significant performance improvement. As an example, the PF solver using sparse matrices published in [35] takes 5.63 s on the CPU and only 0.32 s on the GPU to analyze concurrently 500 instances of the IEEE 300-bus test network. For the sake of comparison, the PF solver using dense matrices published in [38] takes 2.69 s on the GPU to analyze a single instance of the same network.…”

“…However, these cannot be used here as they are limited to the solution of the linear system and do not consider the other operations involved in the PF analysis such as building the admittance matrix, building the Jacobian matrix or computing the bus complex power injection. Finally, the approach we selected based on [35] also has the advantage of exploiting a much high level of parallelism compared to [36][37][38][39][40][41] as it is specially designed for the concurrent evaluation of a large number of networks which is critical to the development of a fast OPF solver as we just explained it.…”

Exploiting Coarse-Grained Parallelism Using Cloud Computing in Massive Power Flow Computation

“…Analysis of a power system requires many calculations related to the matrix and a number of studies have focused on parallel processing of matrix calculations [18,19]. In addition, graphics processing units (GPUs) have been used for fast power flow calculations, as an alternative to other machine cores for parallel processing [20][21][22][23]. Detailed explanations of the parallel processing issues affecting early research are provided in Section 3.…”