Future Generation Computer Systems 97 (2019) 180–193 Contents lists available at ScienceDirect Future Generation Computer Systems journal homepage: www.elsevier.com/locate/fgcs Energy production predication via Internet of Thing based machine learning system Pedro P. Rebouças Filho a,b , Samuel L. Gomes a , Navar M. Mendonça e Nascimento a,b , Cláudio M.S. Medeiros a , Fatma Outay c , Victor Hugo C. de Albuquerque d,* a Laboratório de Processamento de Imagens, Sinais e Computação Aplicada, Instituto Federal de Educação, Ciencia e Tecnologia do Ceará, Fortaleza, CE, Brazil b Graduate Program on Teleinformatics Engineering, Fortaleza, CE, Brazil c College of Technological Innovation, Zayed University, United Arab Emirates d Programa de Pós-Graduação em Informática Aplicada, Universidade de Fortaleza, Fortaleza, Ceará, Brazil highlights • Estimation of the electric power production of a wind turbine. • IoT-based machine learning to predict energy production. • Real wind and power data generated in aerogenerators installed in a wind farm in Ceará State, Brazil. • To obtain the power curve using logistic regression, integrated with Recursive Neural Network to forecast wind speeds. article info Article history: Received 12 November 2018 Received in revised form 11 January 2019 Accepted 13 January 2019 Available online 1 March 2019 Keywords: Wind power Modeling Power curve Time series Nonlinear Autoregressive abstract Wind energy is an interesting source of alternative energy to complement the Brazilian energy matrix. However, one of the great challenges lies in managing this resource, due to its uncertainty behavior. This study addresses the estimation of the electric power generation of a wind turbine, so that this energy can be used efficiently and sustainable. Real wind and power data generated in set of wind turbines installed in a wind farm in Ceará State, Brazil, were used to obtain the power curve from a wind turbine using logistic regression, integrated with Nonlinear Autoregressive neural networks to forecast wind speeds. In our system the average error in power generation estimate is of 29 W for 5 days ahead forecast. We decreased the error in the manufacturer’s power curve in 63%, with a logics regression approach, providing a 2.7 times more accurate estimate. The results have a large potential impact for the wind farm managers since it could drive not only the operation and maintenance but management level of energy sells. © 2019 Elsevier B.V. All rights reserved. 1. Introduction Lee and Lee [1] categorize the Internet of Thing (IoT) as (i) a monitoring and control tool for the automation of systems to track variables and calculate their performance in real time from anywhere. This is particularity interesting for technologies that require advanced monitoring and control, such as electrical grids. (ii) Big data and business analyses are another categories, and are based on systems that can generate large amounts of data, which may come from sensors. This data might be used to find * Corresponding author. E-mail addresses: pedrosarf@ifce.edu.br (P.P. Rebouças Filho), samuelluz@lapisco.ifce.edu.br (S.L. Gomes), navar@lapisco.ifce.edu.br (N.M.M. e Nascimento), claudiosa@ifce.edu.br (C.M.S. Medeiros), fatma.outay@zu.ac.ae (F. Outay), victor.albuquerque@unifor.br (V.H.C. de Albuquerque). relationships between different information systems, in order to improve business issues and strategies. The last category, (iii) Information sharing and collaboration, concern the tracking of information based on predefined thresholds, which might share data with other devices and services. For example, having in- formation as soon as possible in a supply chain service could guarantee optimization of this service. However, a real system is not exclusive to one of the categories above, it has, in fact, features from each one of these categories, such as the various monitoring tools developed under these basis [2–4]. Various researchers have studied the benefits of inserting IoT devices in energy industries. Faheem and Gungor [5] presented a work that gathers wireless sensor networks and applying al- gorithms to optimize smart grid integrations. This field of study lines up with another area, the pattern recognition, which con- sists of various kinds of data-driven methods that aim to find https://doi.org/10.1016/j.future.2019.01.020 0167-739X/© 2019 Elsevier B.V. All rights reserved.