A Nonparametric Bayesian Approach for Probabilistic Representation of Power Uncertainties

This paper develops a nonparameteric Bayesian approach for the probabilistic representation of power system uncertainties involved with wind, solar and load power. The developed approach based on Dirichlet process mixture model (DPMM) analytically formulates the probability distributions of power uncertainties without prior knowledge of the number of mixture components. This provides a great improvement in probabilistic representation of power uncertainties as the proposed model can accommodate the ever growing power data. A computationally efficient VBI method is exploited to estimate the parameters involved with DPMM. Moreover, a novel truncated DPMM is designed to fit the special truncation feature of wind power distributions. The performance of proposed probabilistic representation approach for power uncertainties on real datasets of wind, solar and load power are validated and illustrated in the numerical simulations.