A Machine Learning-Assisted Clustering Engine to Enhance the Accuracy of Hourly Load Forecasting

Due to the growing integration of renewable generations into power networks and the operational challenges made by their volatility and variability, one major concern of power system and market operators to provide the security of the network and to efficiently run the day-ahead and the balancing markets is an accurate load prediction in a short-term horizon. So as to reach this goal, in this research, different data-clustering methods based on machine-learning algorithms are tested whereby a hybrid forecasting engine composed of a search technique called Symbiotic Organisms Search (SOS) and Levenberg-Marquardt learning algorithm is proposed, aiming to increase the precision of short-term forecasting. In this approach, the adjustable weighting coefficients of the Artificial Neural Network (ANN) can be automatically fine-tuned using SOS. The efficiency and applicability of the proposed engine are analyzed and illustrated by its implementation on the EUNITE competition test case and thereby relevant conclusions are drawn to suitably demonstrate the merits of the proposed method.