Day-ahead forecasting approach for energy consumption of an office building using support vector machines

This paper presents a Support Vector Machine (SVM) based approach for energy consumption forecasting. The proposed approach includes the combination of both the historic log of past consumption data and the history of contextual information. By combining variables that influence the electrical energy consumption, such as the temperature, luminosity, seasonality, with the log of consumption data, it is possible for the proposed method by find patterns and correlations between the different sources of data and therefore improves the forecasting performance. A case study based on real data from a pilot microgrid located at the GECAD campus in the Polytechnic of Porto is presented. Data from the pilot buildings are used, and the results are compared to those achieved by several states of the art forecasting approaches. Results show that the proposed method can reach lower forecasting errors than the other considered methods.

I. Introduction

Worldwide policies are placing the consumer as the central piece of future power and energy systems [1]. Consumption flexibility is seen as the most promising solution to enable the system to cope with the increasing penetration of renewable energy sources. While the massive investment in energy efficiency measures contributes to lowering the overall energy consumption, thus being crucial to pave the path towards the dream of a fully renewable energy based energy system; the uncertainty and variability of renewable based generation bring additional challenges that need to be addressed [2]. Being able to take advantage from the consumers’ flexibility potential is, in this way, essential to enable guaranteeing the balance between (variable) generation and consumption [3].