Daily Electrical Load Profile Forecasting to Peak Load Pricing Using Artificial Neural Network

Abstract

This paper presents a method of characterizing Load Distribution Networks for peak load pricing, using load profiles sampling from consumer units, commercial distribution database, and climate variables. It is considered rate subgroup, consumer class, and temperature as exogenous variables. The temperature data considered in the model are directly related to load destined for cooling and heating. Modeling is supported by Artificial Neural Networks methodology with Multi-Layer Perceptron architecture and Back-Propagation training algorithm. In a real case study, load profiles in the Brazilian electrical system, from September 2013 to August 2014, are compared with clustering models traditionally used in load profile characterization to peak-load pricing. The model provides a forecast error equivalent to 5.46% in the distribution sector, lower than the forecast error of 23.04% for the clustering model and load typologies.