Time series prediction using artiﬁcial wavelet neural network and multi-resolution analysis: Application to wind speed data Boubacar Doucoure, Kodjo Agbossou, Alben Cardenas * Department of Electrical and Computer Engineering - Hydrogen Research Institute, University of Quebec at Trois Rivi eres (UQTR), Trois Rivi eres, Quebec, Canada article info Article history: Received 23 December 2014 Received in revised form 22 December 2015 Accepted 2 February 2016 Available online xxx Keywords: Wind speed forecasting Adaptive wavelet neural network Multi-resolution analysis abstract The aim of this work is to develop a prediction method for renewable energy sources in order to achieve an intelligent management of a microgrid system and to promote the utilization of renewable energy in grid connected and isolated power systems. The proposed method is based on the multi-resolution analysis of the time-series by means of Wavelet decomposition and artiﬁcial neural networks. The analysis of predictability of each component of the input data using the Hurst coefﬁcient is also proposed. In this context, using the information of predictability, it is possible to eliminate some components, having low predictability potential, without a negative effect on the accuracy of the prediction and reducing the computational complexity of the algorithm. In the evaluated case, it was possible to reduce the resources needed to implement the algorithm of about 29% by eliminating the two (of seven) components with lower Hurst coefﬁcient. This complexity reduction has not impacted the performance of the prediction algorithm. © 2016 Elsevier Ltd. All rights reserved. 1. Introduction The renewable energy sources (RES) are emerging as one of the best alternatives for sustainable electricity generation. The transi- tion of the traditional energy systems towards renewable sources is required to reduce green-house gas emissions, and consequently to decelerate the global warming [1,2]. Different types of RES tech- nologies are nowadays available for electricity generation. The wind turbines and the photo-voltaic modules are proﬁtable alter- natives for areas with high electricity costs, and are promoted by governments to reinforce the future smart grid. In fact, these sources have experienced strong energy market growth in the past few years. The optimal integration of renewable sources, as they are intermittent, exhibit fundamental challenges including the energy storage, the power conversion and the prediction of the available power [3,4]. The performance and efﬁciency of the renewable energy sys- tems are thoroughly related to the power ﬂow management be- tween all components of the system. The complete energy storage system for renewable sources is not still technologically and economically well developed. However, the overproduction and the demand management could be better performed if the RES power proﬁle is known in advance. Nonetheless, the electricity production from wind energy has some particularities derived from the intermittent behavior of wind speed, e.g. the production proﬁle can not be adjusted satisfactorily to the one of the load demand; a balance of production and demand is then required which can be achieved using power reserves from other energy sources and/or a storage system support [5,6]. Scientiﬁc and industrial research ef- forts are unceasing to develop more accurate and reliable fore- casting tools to mitigate the problem of irregularity in the RES power production [7]. The aim of those efforts seeks specially short term forecast and their applications on wind power management; they would enable e.g. the scheduling of the energy requirements for a given period (planning and delivering), solving more accu- rately and safely the micro-grid constraints and the schedule of maintenance [8e10]. Proﬁles of the available power of wind and solar sources depend on the geographic location, the season (or time of the year), the time of the day and other physical parameters. Wind speed modeling using time series is usually used to analyze wind proﬁles to obtain the predicted values. In the context of local energy management, the study of time series by means of a predictability analysis can be very helpful. Predictability analysis of time series * Corresponding author. E-mail address: Alben.cardenasgonzalez@uqtr.ca (A. Cardenas). Contents lists available at ScienceDirect Renewable Energy journal homepage: www.elsevier.com/locate/renene http://dx.doi.org/10.1016/j.renene.2016.02.003 0960-1481/© 2016 Elsevier Ltd. All rights reserved. Renewable Energy 92 (2016) 202e211