Short-Term Forecasting in Smart Grid Environment using N-BEATS

Abstract Forecasting is increasingly becoming necessary in the modern Smart Grid to properly operate and control renewable integrated power systems. Forecasting accuracy plays a vital role in the adequate scheduling of conventional resources to allow the maximum utilization of renewable sources and provide proper operation of the system. Mainly, accurate short-term forecasting of load, electricity price, solar photovoltaic (PV), and wind turbine (WT) based renewable power generations are of central importance to manage remaining sources optimally to supply cost-effective power to the consumers and reduce carbon emissions. This paper proposes a Neural Basis Expansion Analysis Time Series Forecasting (N-BEATS) algorithm for highly accurate short-term forecasting of SG load, electricity price, PV and WT-based power generations. Data preprocessing, such as normalisation, outlier management, and new cyclic feature building, is employed in the proposed algorithm to enhance its performance. The developed N-BEATS algorithm performs excellently on dynamically fluctuating datasets and peak load prediction. The suitability and effectiveness of the N-BEATS algorithm are validated on four different datasets. Further, the superiority of the algorithm is demonstrated by comparing the performance of the algorithm with several well-explored state-of-the-art forecasting algorithms.