Parallel GPU Implementation for Fast Generating System Adequacy Assessment via Sequential Monte Carlo Simulation

The Sequential Monte Carlo Simulation (SMCS) is a powerful and flexible method commonly used for generating system adequacy assessment. By sampling outage events in sequence and their respective duration, this method can easily incorporate time-dependent issues such as renewable power production, the capacity of hydro units, scheduled maintenance, complex correlated load models, etc, and is the only method that provides probability distributions for the reliability indexes. Despite these advantages, the SMCS method requires considerably more simulation time than the Non-sequential Monte Carlo Simulation approach to provide accurate estimates for the reliability indexes. In an attempt to reduce the simulation time, the SMCS method has been implemented in parallel using a Graphics Processing Unit (GPU) to take advantage of the fast calculations provided by these computing platforms. Two parallelization strategies are proposed: Strategy A, which creates and evaluates yearly samples in a completely parallel approach and while the estimates of the reliability indexes are computed in the CPU; and Strategy B, which consists on concurrently sampling the outage events for the generating units while the state evaluation and the index estimation stages are executed in serial. Simulation results for the IEEE RTS 79, IEEE RTS 96, and the new IEEE RTS GMLC test systems, show that both implementations lead to a significant acceleration of the SMCS method while keeping all its advantages. In addition, it was observed that Strategy B results in less simulation time than Strategy A for generation system adequacy assessment.