A Deep Neural Network as a Strategy for Optimal Sizing and Location of Reactive Compensation Considering Power Consumption Uncertainties

This research proposes a methodology for the optimal location and sizing of reactive compensation in an electrical transmission system through deep neural network (DNN) by considering smallest cost compensation. An power (EPS) is subjected to unexpected increases loads which are physically translated as increment users EPS. phenomenon decreases voltage profiles whole also EPS’s reliability. One strategy face this problem compensation; however, finding not easy task. Different algorithms techniques such genetic non-linear programming have been used find solution problem; these generally need big processing time usually considerable. That being stated, paper’s aims improve profile under scenarios PQ load randomly connected any busbar system. The will be found DNN capable relatively small time. tested three case studies, IEEE 14, 30 118 systems. In each systems, brute force algorithm (BFA) implemented connecting composed 80% active 20% (which varies from 1 MW 100 MW) every busbar, scenario, 10 Mvar 300 Mvar) busbar. Then flows generated selecting scenario closest 90% original profiles, selected overcompensation would increase cost) avoided. Through BFA, trained 70% data training other 30% test data. Finally, achieving 100% accuracy (in all studies when compared with BFA) objective deviation has difference 3.18%, 7.43% 0% respectively (when BFA). With methodology, it possible increment, almost no (with trained, takes seconds solution).