Controlador de demanda de energia utilizando inteligência computacional

Abstract

With the current global situation of resource scarcity, the economy of electric energy has become relevant. Thus, the Universidade Federal de Lavras, to avoid exceeding the demand for electric energy purchased, acquired a demand controlling system. The functioning of the demand controller left much to be desired, especially concerning the priority defined for shutting down the least essential charges. Because of this, this work aimed at improving the performance of the demand controller, intelligently and dynamically optimizing the priority of shutting down the charges. Thus, knowledge on computational intelligence was used to create an automated system that, allied to the demand controller, can obtain better performance. Understanding that the demand controller work by shutting down the charges in order not to exceed the demand, and that the charges to be shut down are air-conditioners, a charge priority classification system was developed, in addition to a electric energy prediction system for the next 15 minutes. The methodology employed was based on Artificial Neural Networks for developing two computational systems to work in parallel. When analyzing the simulated results, we verified that, for the charge predictor, the mean and deviation for the EMQ, in the testing phase, was of 0.00006701 ± 0.00000262720; the R 2 training coefficient was of 0.9634 ± 0.00289; and the test R 2 was of 0.987 ± 0.00152. For the classification, the value obtained for the EMQ in the training phase was of 0.0014436. We also verified that the Kappa hit index of the general sequence was of 0.8239. In conclusion, the demand controller was optimized for the process of shutting down charges.