Electric Energy Consumer Characterization, Classification and Demand Forecasting using Convolutional Neural Networks

This paper presents a procedure to support the retail and distribution companies on the extraction of knowledge from electricity consumption data. Our final objective is to forecast individual load profiles of consumers for a particular season of the year using the historical information gathered during the months of the previous season. The algorithm classifies consumers in one of the 7 clusters, with 70% accuracy in the best case, which gives retail companies an excellent point to start in tariff customization. We encode the power use (time series) in GAFs and MTFs to represent the images that are processed by the CNNs to perform classification tasks.