Contents lists available at ScienceDirect Solar Energy journal homepage: www.elsevier.com/locate/solener Using artificial neural networks to assess HVAC related energy saving in retrofitted office buildings Chirag Deb a, ⁎ , Siew Eang Lee a , Mattheos Santamouris b a Department of Building, School of Design and Environment, National University of Singapore, Singapore 117566, Singapore b Faculty of Built Environment, University of New South Wales, Sydney, NSW 2052, Australia ARTICLE INFO Keywords: Artificial Neural Network (ANN) Energy saving Building retrofit Variable selection ABSTRACT This study aims to develop prediction models for HVAC related energy saving in office buildings. The data-driven modelling makes use of data gathered from several energy audit reports. These reports entail building and energy consumption data for 56 office buildings in Singapore. The two models are developed using Multiple Linear Regression (MLR) and Artificial Neural Network (ANN). The methodology to select the most appropriate input variables forms the essence of this study. This variable selection procedure involves 819,150 iterations, taking all possible combinations of the 14 input variables to determine the most accurate model. The dependent variable is taken as the change in energy use intensity (EUI, measured in kWh/m 2 .year) between pre- and post- retrofit conditions. The results show that the ANN model is more accurate with a mean absolute percentage error (MAPE) of 14.8%. The best combination of variables to achieve this comprises of gross floor area (GFA), air- conditioning energy consumption, operational hours and chiller plant efficiency. The information on these four variables, along with the prediction model can be used to predict HVAC related energy savings in office buildings to be retrofitted. 1. Introduction The challenges posed due to climate change have accelerated re- search in energy efficient buildings. Many studies now present ways to reduce cooling load in buildings by active as well as passive means (Ascione, 2017). A lot of research has also been carried out on exploring the effect of external environmental changes on building energy con- sumption (Wong et al., 2011; Pisello, 2017; Santamouris et al., 2001). There is also a growing awareness on not just net zero- but positive- energy buildings (Kolokotsa et al., 2011). Although renewable energy technologies have a promising outlook, it is important to simulta- neously advance research in energy efficient buildings and avoid wa- stage in energy consumption. Energy efficiency in buildings is one of the five measures to secure long term decarbonisation as per the International Energy Agency (IEA) 1 (IEA, 2015). Energy consumed in the building sector consists of residential and commercial end users and accounts for about 20% of the total delivered energy worldwide. According to the U.S. Energy In- formation Administration (EIA), energy consumption in the commercial building sector is projected to be the fastest growing, at a rate of 1.6%/ year (EIA, 2016). Due to this, governments around the world have embarked on various initiatives. A survey by the World Energy Council (WEC) shows that most countries have employed either voluntary or mandatory energy efficiency regulations for buildings (Council, 2004). This survey consisted of 63 countries that constitutes 83% of the global energy consumption. These regulations not only aim to achieve energy efficiency in new buildings but also in existing buildings by outlining retrofitting guidelines. This is because the number of existing buildings constitutes a large part of current and future building stock. Energy use forecasts also show that portion of energy consumed per capita by the commercial building sector is expected to increase while that of the residential building sector is expected to decrease (U.S. Department of Energy, 2011). This signifies the importance of realizing the potential of energy efficiency in the commercial building sector. Although there is a vast scope of retrofitting opportunities, the literature and current ret- rofit practices show that energy efficiency improvement projects have been conducted on an ad hoc basis without a systematic decision making process (Ruparathna et al., 2016; Hall, 2014). There have been many advances in research in building energy performance. Currently, there exists a wide variety of methodologies to identify energy conservation opportunities for retrofitting buildings. These range from a detailed analysis of an individual building to a https://doi.org/10.1016/j.solener.2018.01.075 Received 31 August 2017; Received in revised form 5 January 2018; Accepted 23 January 2018 ⁎ Corresponding author. E-mail address: chirag.deb@u.nus.edu (C. Deb). 1 ©OECD/IEA 2015 World Energy Outlook Special Report, IEA Publishing. License: [http://www.iea.org/t&c/termsandconditions/]. Solar Energy 163 (2018) 32–44 0038-092X/ © 2018 Elsevier Ltd. All rights reserved. T