Predicting Household power consumption: Using Gradient Boosting and Deep Quantile Regression Model

Abstract

The rapid development of technology, cities, and the introduction of IoT has caused high fluctuations in energy consumption. Therefore, efficient energy management and forecasting energy consumption for buildings are important in decision-making for effective energy-saving and keeping the world a better place. Over the years researchers have tried to propose machine learning approaches to forecast and monitor abnormal electricity consumption, however, there is still a need for more new approaches to forecasting energy consumption correctly and precisely. This paper attempts to address the global issue of efficient electricity management through a data science approach. The paper proposes a quantile learning approach for predicting household electricity consumption. We propose a deep quantile regression model and a gradient boosting model to forecast power consumption. In addition, we propose an approach to detect abnormal patterns in energy consumption data using estimated consumption intervals. We tested our approach on a publicly available univariate time series dataset collected from a house located in Sceaux.