LSTM-based Network for Current Harmonic Distortion Time Series Forecasting

In recent years, machine learning (ML) and deep learning (DL) algorithms have successfully contributed to the power system transformation into the present smart grid (SG). Most of their potential has been deployed in both electric loads and renewable energy sources (RES) production time series forecasting as a crucial support for power system planning. In contrast, much less attention has been paid to forecasting other time series of significant relevance for the power system, as is the case of the time series of power quality (PQ) indices. This situation strongly contrasts with the growing concern of the research community about the large number of PQ issues reported within the electric power system due to multiple causes (i.e., an increasing number of grid-connected RES and the proliferation of nonlinear loads) and with the importance of keeping these PQ indices under the limits reported by international standards to preserve the power system's stability. This research provides an exhaustive overview of the progress the research community has made over the use of DL in forecasting time series of PQ indices, showing that this discipline is far from being mature. Given this scenario, a long short-term memory (LSTM) network is proposed to forecast the steady state of several PQ indices time series which evaluate the current distortion at the point of common coupling (PCC) of a dwelling. The main contributions of this paper include the use of higher data granularity (l-minute of aggregation interval), the time series forecasting for both components of each current harmonic (real and imaginary components) to facilitate the harmonic mitigation task and the integration of the proposed model as a part of an internet of things (IoT) infrastructure to enable the online PQ indices time series forecasting and control over the energy resources.