Accurate and time-efficient negative binomial linear model for electric load forecasting in IoE

Abstract

Accurate and efficient model predictive control (MPC) is essential for Internet of energy (IoE) to enable active real-time control, decentralized demand-supply balance, and dynamic energy management. The MPC consists of short-term electric load forecasting, whose accuracy is affected by the load characteristics, such as overdispersion, autocorrelation, and seasonal patterns. The forecasting efficiency depends on the computational time that is required to produce accurate results and is affected by the IoE data volume. Although several fundamental short-term forecasting models have been proposed, more accurate and efficient models are needed for IoE. Therefore, we propose a novel forecasting temporal negative binomial linear model (NBLM) that handles overdispersion and captures nonlinearity of electric load. We also classify the load into low, moderate, and high intraday seasons to increase the forecast accuracy by modeling the autocorrelation in each season, separately. The temporal NBLM was evaluated using real-world data from Jericho city, and its results were compared to other forecasting models. The temporal NBLM is found more accurate than the other models as the mean absolute percentage error (MAPE) is reduced by 29% compared to the ARMA model. In addition, the proposed model is more efficient as its running time is reduced by 63% in the training phase and by 87% in the forecast phase compared to the Holt-Winter model. This increase in accuracy and efficiency makes the proposed model applicable for load forecasting in IoE contexts where data volume is large and load is highly fluctuated, is over dispersed, is autocorrelated, and follows seasonal patterns.