Artificial neural network-based distribution substation and feeder load forecast

Abstract

A methodology for estimating future demand values at both distribution substation and primary feeder levels is described in this paper. The software implementation of the proposed methodology is already running in a 138/11.9-kV, 3×40-MVA distribution substation. Results obtained with this implementation are very encouraging, even when using as little historical data as 3 months. Forecast error is also very low when a demand curve substantially different from the ones presented to the artificial neural network in its training phase are used in the processing mode. A separate module for dealing with load transfers between primary feeders during contingencies is currently in its final stages of development. (5 pages)