A Universal Model for Harmonic Source Uncertainty Based on Gaussian Process Regression

The extensive integration of renewable energy sources and large-scale power electronic devices into the grid has made the analysis and modeling of harmonic source behavior increasingly challenging. Existing models primarily provide precise value estimations but struggle to adequately reflect the uncertain impacts of unconsidered physical factors on the characteristics of harmonic sources, thus having limited universality and adaptability. This paper suggests a comprehensive model for addressing these concerns by utilizing Gaussian Process Regression (GPR) to manage harmonic source uncertainty. Firstly, the linear dependency between harmonic currents and voltages of harmonic sources is embedded into the mean function of the Gaussian process. Secondly, an appropriate covariance function is selected to reflect the similarity of harmonic properties under different operating conditions of harmonic sources. Thirdly, model parameters are solved using the maximum likelihood method to perform interval predictions of harmonic currents. Finally, the feasibility of the proposed method for modeling single harmonic sources and complex multiharmonic source networks is verified through simulation data from a 12-pulse rectifier device and a complex network containing multiple harmonic sources. The results of the case studies demonstrate that the proposed method can capture the uncertain behavior of harmonic characteristics, with features such as high accuracy and strong universality.