A Research On Parallel Computing Of Optimal Power Flow In Cuda Platform

Parallel computing is an effective tool to solve large-scale problems, analyse mass data, process information efficiently. Parallel computing break down the traditional methods, make a full use of the computer resources, and solve large-scale computing problems more quickly. Nowadays, parallel computing of multicore CPU (Central Processing Unit) of has got a good application. In recent years, with the quickly development of GPU (GPU) and CUDA (Compute Unified Device Architecture), a good platform has been provided for parallel computing.Security Constrained Optimal Power Flow (SCOPF) considered a set of expected faults (N-1) security constraints based on the Power Flow. Therefore, the model of SCOPF need to expand the equality constraints and the safety and quality inequality constraints after faults based on the feasible Optimal Power Flow (OPF) model. In SCOPF, with the expansion of the system scale and the increase of expected faults, require improving the accuracy of the solution, but it can’t be realized by traditional serial computational. This thesis use two methods to mapping SCOPF to the GPU for parallel computing:parallel computing by blocks and LDLT decomposition based on the supernodal multifrontal method. To a certain extent, it can improve the computational speed of large-scale SCOPF problems, and solve the large-scale SCOPF problems effectively.This thesis proposed parallel computing algorithm of SCOPF based on CUDA, which was tested in Graphics Processing Unit GTX460and Graphics Processing Unit Tasla C2050of the IMAX server. It sovled large-scale systems SCOPF problem successfully, and it achieved a certain degree of speedup rate and efficiency. But the speedup was far from the number of CUDA cores in the GPU, it also gived the detailed analysis of the reason in this thesis. Parallel computing based on CUDA platform for large-scale problems will have a broad prospect of application. It applied to the analysis and calculations of power system still need a great deal of research.