I. Introduction
Electrical load forecasting has played a key role in power system operation and management as the basis of various power system analyses [1]. In particular, short-term load forecasting (STLF), which commonly implies from hour-ahead to week-ahead forecasts, is used in many power grid applications to ensure reliable operations of power grids [2], [3], [4]. For instance, day-ahead demand forecasts are utilized in load smoothing and peak shaving [5]. The sub-hourly electrical load forecasts are also considered as the key information required to optimize energy storage systems for frequency response [6], [7]. In addition, economic dispatch optimization and scheduling are investigated along with electrical load forecasting for multi-microgrids and decentralized co-generation plants [8], [9]. Furthermore, day-ahead grid load forecast is used as the input feature for the prediction on the next-day market price of electricity [10]. Accordingly, accurate STLF facilitates reliable and efficient power system operations through supporting power grid applications. Nevertheless, STLF remains a challenge owing to the high dimensionality and volatility of the electrical load data as a time series.