Investigation of Optimum Reserve Capacity Requirement in Ancillary Services with Extreme Learning Machine

Abstract

In advanced power systems, reserves are used to balance load mismatches in the face of unexpected events and to meet unforeseen contingencies. This reserve used in the units is defined as the spinning reserve (SR), which is realized with the loading and unloading instructions within the scope of the Ancillary Services (AS). As the contribution of solar and wind turbines increases in installed power, it requires larger amounts of SR and creates a significant increase in production costs. This article uses the Extreme Learning Machine (ELM) method to estimate SR capacity in the day-ahead and intraday market. In this study, experiments were conducted to determine the effectiveness of ELM-SR in estimating the SR parameter, by randomly dividing the data set into 60–40% training-test sets and using 1–100 neurons in the latent layer. The results obtained with the ELM were also compared with the results observed with the ANN, LR, and SVM methods, which determine the optimum reserve amount of the method used. Thus, even in the face of increased production in renewable energy in the power system with a properly planned reserve with ELM, a significant economic gain is achieved with the allocation of the optimum reserve.