Please cite this article in press as: C. Tong, et al., An efficient deep model for day-ahead electricity load forecasting with stacked denoising auto-encoders, J. Parallel Distrib. Comput. (2017), http://dx.doi.org/10.1016/j.jpdc.2017.06.007. J. Parallel Distrib. Comput. ( ) – Contents lists available at ScienceDirect J. Parallel Distrib. Comput. journal homepage: www.elsevier.com/locate/jpdc An efficient deep model for day-ahead electricity load forecasting with stacked denoising auto-encoders Chao Tong a , Jun Li a , Chao Lang a , Fanxin Kong b , Jianwei Niu a , Joel J.P.C. Rodrigues c , d, e, f , * a School of Computer Science and Engineering, Beihang University, Beijing, 100191, China b School of Computer Science, McGill University, Montreal, Canada, H3A2T5 c National Institute of Telecommunications (Inatel), 37540-000 Santa Rita do Sapucaí - MG, Brazil d Instituto de Telecomunicações, 6201-001 Covilhã, Portugal e University of Fortaleza (UNIFOR), 60811-905 Fortaleza-CE, Brazil f ITMO University, 191002 St. Petersburg, Russia highlights • We propose a deep learning based model for forecasting day-ahead electricity load. • It uses history load data, weather and season parameters. • It uses multiple stacked denoising auto-encoders to extract features. • The refined features and a season parameter are fed into a SVR model for training. article info Article history: Received 2 March 2017 Received in revised form 5 May 2017 Accepted 9 June 2017 Available online xxxx Keywords: Deep learning Multi-modal Stacked denoising auto-encoders Feature extraction Support vector regression abstract In real word it is quite meaningful to forecast the day-ahead electricity load for an area, which is beneficial to reduction of electricity waste and rational arrangement of electric generator units. The deployment of various sensors strongly pushes this forecasting research into a ‘‘big data’’ era for a huge amount of information has been accumulated. Meanwhile the prosperous development of deep learning (DL) theory provides powerful tools to handle massive data and often outperforms conventional machine learning methods in many traditional fields. Inspired by these, we propose a deep learning based model which firstly refines features by stacked denoising auto-encoders (SDAs) from history electricity load data and related temperature parameters, subsequently trains a support vector regression (SVR) model to forecast the day-ahead total electricity load. The most significant contribution of this heterogeneous deep model is that the abstract features extracted by SADs from original electricity load data are proven to describe and forecast the load tendency more accurately with lower errors. We evaluate this proposed model by comparing with plain SVR and artificial neural networks (ANNs) models, and the experimental results validate its performance improvements. © 2017 Elsevier Inc. All rights reserved. 1. Introduction 1.1. Background and motivation Smart Grids (SGs) [17,18,29] (as shown in Fig. 1) are new-type electrical grids which provide a promising scheme for more utility electricity delivery, aiming at reaching higher reliability and using electrical resources more economically and rationally, based on computer remote control and automation with advanced tech- niques of communication and sensing. These systems are made * Corresponding author at: National Institute of Telecommunications (Inatel), 37540-000 Santa Rita do Sapucaí - MG, Brazil. E-mail address: joeljr@ieee.org (J.J.P.C. Rodrigues). possible by two-way communication technology which has been used for decades in other industries. Recent years, both theory de- velopment and practical applications of SGs have gradually made major strides in the real life. Sensor technique [15,23] is one of the applications which make grids ‘‘smart’’. Massive sensors with dif- ferent functions are deployed and return abundant heterogeneous data describing the actual situation of grids in detail, and then some controls and adjustments can be done to fit the dynamic changes in the real environment. Thus the occurrence of SGs strongly pushes electrical grids research into a ‘‘big data’’ era due to a huge amount of information collected by sensors. One of the most common tasks in electrical grids is forecasting the day-ahead electricity load for an area, e.g., a city or a state. http://dx.doi.org/10.1016/j.jpdc.2017.06.007 0743-7315/© 2017 Elsevier Inc. All rights reserved.