Abstract
In this project, we accelerate the computationally intensive Forward Backward Sweep (FBS) method utilizing a Graphics Processing Unit (GPU). FBS is a well-known method for power flow computation used in modeling and analysis of electric power distribution systems in North America that typically involves a very large number of iterative calculations. Since distribution systems are becoming increasingly complex, performing this iterative analysis with traditional modeling tools may take hours or sometimes days. Hence, there is a growing need for speeding up the runtime of power distribution system modeling tools by parallelizing their implementations. GPUs, which have massively parallel architectures, have been traditionally used for graphics-related tasks which have abundant parallelism. However, after the introduction of compute-specific GPU programming languages (e.g., CUDA), GPUs were also started to be intensively used for accelerating computationally intensive tasks (such as FBS) that have nothing to do with graphics but exhibit abundant parallelism. In this study, we use parallel computation patterns (i.e., segmented scan, reduction) and kernels on the GPU to accelerate FBS. To evaluate our approach, we perform our tests on binary power distribution trees that have number of nodes between 1K to 256K. Our results show that the parallel implementation brings up to 3.9x total speedup over the serial implementation. As expected, for the parts of the computation that entirely run on the GPU, larger speedups are achieved as the size of the distribution tree increases. We also provide a discussion on how the topology of the tree would affect the results.