Comparison of Gaussian mixture reductions for probabilistic studies in power systems

This paper presents the comparison of three pair-merging methods to reduce the number of Gaussian mixture components used to model non-Gaussian Probabilistic Density Function (PDF) of random power system variables such as power demands, wind power outputs or other intermittent power sources. It also introduces a fine-tuning algorithm to improve the solution of the pair-merging methods to better approximate the original Gaussian mixture. A Gaussian mixture distribution with seven components is used to validate and demonstrate the algorithms.