Applying load profiles propagation to machine learning based electrical energy forecasting

Abstract

Electrical energy generation represents an economical and environmental challenge requiring an optimal control of the production process. An accurate modeling of the electrical energy is essential to develop efficient forecasting systems. To achieve this goal, this paper introduces an innovative load forecasting approach using Load Profiles (LPs). First, the power consumption in Algeria is analyzed to detect the different factors affecting the demand. Then, the fluctuation of the seasonal data is applied through hourly temperature profiles. The LP-based forecasting is performed using three levels (annual, weekly and daily) LP-propagation. Short-term and mid-term load forecasting models were developed using multiple Artificial Intelligence techniques. Among them, a two-dimensional Convolutional Neural Network (CNN) used here for the first time in load forecasting. The resulting prediction accuracies of both the Artificial Intelligence (AI)-based and LP-based models were considerably high, producing (MAPE= 0.80%, RMSE= 75.57MW, Willmott's Index (WI) = 0.99) for the two-dimensional CNN. Compared to (MAPE= 3.86%, RMSE= 372.68 MW, WI= 0.95) for the LP-propagation technique.