Classification Techniques to Predict the Behavior of Electrical Energy Consumption

Abstract

In this paper, we studied the feasibility of developing a complete intelligent system for predicting the behavior of electrical energy consumption as electricity is gaining room as energy source. This work is presenting an overview to organize and visualize all the data using different representations according to the specific goal of classifying the consumption of electricity usage with respect to its behavior. It has been interesting to provide a method that allowed the visual representation for each algorithm used for classification of electrical data and satisfies all the purposes sought. This facilitates the comparison among electricity usage, permits to access easily to the individual electric consumption main characteristics and detect possible irregularities or problems. Electricity is gaining room as energy source, its share will keep increasing constantly in the following decades. In this close future, smart grids and smart meters deployment will benefit both the utility and the consumer. This work has classified the electricity usage with respect to the consumption according to the similarities of their electrical load profiles, using the proportion of energy usage per hour as a common framework using the classification (five algorithms in particular) and clustering theory. The objective behind this segmentation or classification is to be able to provide personalized recommendations to each group in order to reduce their energy consumption and the associated costs, fostering energy efficiency measures and improving the electricity usage engagement. The desired classification is obtained by an iterative process, based on computational classification calculation (using Weka for analysis and results generation) and finalized by a post-clustering and classification analysis applying visualization and statistical techniques to detect the outliers and reallocate them to a more appropriate classes. Five different classification techniques (Decision Tree, Support Vector Machine, Naïve Bayes, Random Forest and Hybrid), were tested and compared, giving similar outputs. The solution from the Hybrid is the one that better adapts to the classification sought, which is used as the base of the post-classification stage to obtain the final classification.