A Data-driven Approach to Probabilistic Harmonic Current Prediction for Residential Loads

With the widespread use of nonlinear power electronics in residential households, the harmonic content caused by them is increasing. To effectively counter the increasingly negative impact of residential harmonics, it is necessary to predict the harmonic currents of residential loads. Since residential loads are highly random, their harmonic current data have probabilistic fluctuations. For this reason, this paper proposes a probabilistic time-series harmonic current prediction method based on a neural network point prediction method. In this paper, multiple neural network methods are used to predict the deterministic components of harmonic currents, and the best prediction method is selected for each case according to the prediction error, while the statistical probabilistic prediction method is then used to predict the stochastic harmonic currents. Finally, the accuracy and practicality of the proposed probabilistic harmonic current prediction method are verified based on the measured data.