Forecast Error Modeling for Microgrid Operation Considering Correlation among Distributed Generators Using Gaussian Process Regression

Abstract

Small-scale and self-sustainable grids known as microgrid (MG) are a key element for future power systems with a high penetration of renewable generators. The uncertainty of renewable power generation in systems such as photovoltaic (PV) and wind turbine needs to be compensated with balancing devices. The balancing devices can be an Energy Storage System (ESS) or conventional thermal fossil-fuel generators with enhanced flexibility. The uncertainty of renewable generators can be statistically analyzed for cost-effective assessment and operation of those compensating devices. Multiple uncertainty factors can be investigated and modeled as a joint probability distribution function (PDF) considering the temporal correlation among themselves. Several uncertainty factors with different marginal distributions and scales can be integrated as multivariate probability distribution by transforming them into normal distribution using rank correlation. As the number of uncertainty factors considered in a microgrid increases, it leads to much more complexity to in defining the conditional probability distribution generated from a joint PDF. In this paper, a method to model the distribution of net-load forecast error is proposed considering the correlation among uncertainty factors. A data-driven Gaussian process regression is introduced to train and validate conditional PDF among uncertainty factors, which are transformed into normal distribution without losing intrinsic marginal distribution. The conditional density function based on the proposed method has better suitability to estimate distribution of netload error. The conditional density function based on the proposed method shows better suitability for estimation of net load error distribution.