Accelerated parallel WLS state estimation for large-scale power systems on GPU

Abstract

Owing to the growing size and complexity of power networks, online monitoring of the power system state is a challenging computational problem. State estimation is paramount for reliable operation of large-scale power systems. Even with modern multi-core processors, the large number of computations in state estimation are still a burden on time and are highly memory intensive. In this paper the idea of using massively parallel graphic processing units (GPUs) for weighted least squares (WLS) based state estimation is introduced and executed. The GPU is especially designed for processing large data sets. A data-parallel implementation of the WLS method is proposed. The speed of the GPU-based state estimation for several test systems is compared with a sequential CPU-based program. The simulation results show a speed-up of 38 for a 4992-bus system.