Harmonic Current Magnitude and Phase Angle Forecasting to Support Harmonic Power Market Planning

The massive proliferation of distributed energy resources (DERs) is changing the opportunities and challenges encountered in the power grid. Some technical barriers here include managing energy sources with high variability, loss of inertia or the occurrence of voltage disturbances. However, the growth of grid-connected inverter-based resources (IBRs) can also help address other vulnerabilities such as the presence of harmonic distortion. Therefore, apart from participating in the frequency and voltage regulation, IBRs can also exploit their capability of compensating current harmonics in the emerging harmonic power market (HPM). Nevertheless, Distribution System Operators' (DSOs) planning under this market becomes unrealistic without using forecasting tools. While deep learning (DL) approaches have been applied in the field of power quality (PQ), significantly less attention has been devoted to PQ time series forecasting, particularly in the case of current harmonics. Unlike all research to date, the present paper uses DL architectures to model multivariate time series (MTS) of the magnitude and phase angle of current harmonics computed according to the IEC 61000-4-30 and 61000-4-7 standards and the prevailing phasor methodology using a 10-minute interval. Measurements were performed at the point of connection (PoC) of the main data center building at the University of Cordoba, in southern Spain. Finally, absolute and relative error metrics were employed to assess the performance of the models, showing that the bidirectional long-short-term memory (BiLSTM) achieves the best results. The prediction error was also characterized to provide a confidence interval to drive optimization problems considering the magnitude and phase angle uncertainty.