A GPU-Based Model for Composite Power System Reliability Evaluation and Sensitivity Analysis

Decarbonizing the grid by incorporating more renewable energy sources has made power system reliability more critical. Although Monte Carlo simulation is a traditional method for power system reliability assessment, the computational costs limit its utility as power grid complexity increases. As a result, this study introduces a Graphics Processing Unit (GPU) based Convolution Neural Network (CNN) model to a non-sequential Monte Carlo Simulation (MCS) model to improve computation efficiency while maintaining high evaluation accuracy during reliability evaluation. Furthermore, this study explores the viability of the proposed model for sensitivity analysis of power system components on the overall system reliability condition. The proposed model is tested using the IEEE Reliability Test System 79 and is compared to the baseline Monte Carlo simulation approach. Results show that the model provides more than 80% time savings during reliability analysis while maintaining good evaluation accuracy for sensitivity assessments.