A Simplified Approach to High Performance Computing in Power Flow Analysis

The power grid is a complex network interconnecting energy sources with loads. The power flow and state estimation problems model the power flows and voltages throughout the grid. They are currently solved using computer simulations computed with serial processing methods, on traditional CPUs. To reduce the simulation timelines, and provide a faster, “real-time” solution, the practicality of using the Graphics Processing Unit (GPU) will be investigated. High Performance Computing (HPC) technology is becoming more mature as it is being interweaved in many high-computational tasks related to the STEM industry. Currently, the knowledge barrier to entry is relatively low and therefore the concepts of HPC can be applied to power simulation software used in MATPOWER and MATLAB™ with minimal knowledge of GPU primitives coding in OpenGL and DirectX. MATLAB’s parallel computing toolbox has many built-in functions that will compute their operations directly on the GPU. This thesis will take this simplified approach to HPC and leverage this toolbox to determine if the power flow problem can be sped up by paralleizing MATPOWER’s algorithm, which is already optimized to run on the CPU. If speed up is achieved this technology can be used with MATPOWER’s already robust libraries and functions to help researchers in the areas Optimal Power Flow (OPF), State Estimation, and Economic Dispatch to name a few. The conclusion to this thesis will demonstrate that yes, a simplified approach to power flow can be executed using MATLAB’s parallel computing toolbox, but there are bottlenecks created by the limitations of MATLAB, which if addressed through future research, could unlock the full potential behind the GPU