IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 13, NO. 5, OCTOBER 2017 2587 An Accurate and Fast Converging Short-Term Load Forecasting Model for Industrial Applications in a Smart Grid Ashfaq Ahmad, Student Member, IEEE, Nadeem Javaid, Senior Member, IEEE, Mohsen Guizani, Fellow, IEEE, Nabil Alrajeh, and Zahoor Ali Khan, Senior Member, IEEE Abstract—Short-term load forecasting (STLF) models are very important for electric industry in the trade of energy. These models have many applications in the day-to-day operations of electric utilities such as energy generation planning, load switching, energy purchasing, infrastructure maintenance, and contract evaluation. A large variety of STLF models have been developed that trade off between forecast accuracy and convergence rate. This paper presents an accurate and fast converging STLF model for industrial applications in a smart grid. In order to improve the forecast accuracy, modifications are devised in two popular techniques: mutual information based feature selection; and enhanced differential evolution algorithm based error minimization. On the other hand, the conver- gence rate of the overall forecast strategy is enhanced by devising modifications in the heuristic algorithm and in the training process of the artificial neural network. Simulation results show that accuracy of the newly proposed forecast model is 99.5% with moderate execution time, i.e., we have decreased the average execution of the existing bilevel forecast strategy by 52.38%. Index Terms—Activation function, artificial neuron, dif- ferential evolution, fitness function, load forecast, mutual information (MI), short term, smart grid (SG), training. I. INTRODUCTION P RIOR and after advanced technology installation, indus- trial utilities seek as much return on investment as possible. On the other hand, customers seek for the least electricity con- sumption cost possible [1]. Thus, the nature of industrial utilities and consumers is greedy. The traditional grid was unable to sat- Manuscript received August 19, 2016; revised November 23, 2016; accepted December 6, 2016. Date of publication December 9, 2017; date of current version October 3, 2017. This work was supported by the International Scientific Partnership Program ISPP at King Saud Uni- versity through ISPP# 0053. Paper no. TII-16-0877.R1. (Corresponding author: Ashfaq Ahmad.) A. Ahmad and N. Javaid are with the COMSATS Institute of Information and Technology, Islamabad 44000, Pakistan (e-mail: ashfaqcomsats@ gmail.com; nadeemjavaidqau@gmail.com). M. Guizani is with the University of Idaho, Moscow, ID 83844 USA (e-mail: mguizani@uidaho.edu). N. Alrajeh is with CAMS, Department of Biomedical Technology, KSU, Riyadh 11633, Saudi Arabia (e-mail: nabil@ksu.edu.sa). Z. A. Khan is with CIS, Higher Colleges of Technology, Fujairah 4114, United Arab Emirates (e-mail: Zahoor.Khan@dal.ca). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TII.2016.2638322 isfy both parties at the same time to meet modern grid challenges such as reliability, robustness, etc. [2]. The new and smart con- cept of grid, the smart grid (SG), renovates the traditional grid by integrating the latest communication technologies [3]. In this way, the consumers are engaged in the trade of energy. The bidi- rectional communication or energy flow benefits the industrial utilities and the consumers [4]. More specifically, the consumers are no longer only consumers, instead they are “prosumers” who have the ability to access the electricity market as both sellers and buyers. At the same time, the smart industrial utilities have the ability to efficiently manage their resources [5], [6]. In order to optimize the performance of SG, especially its distribution part, a decision making entity is needed. Proper de- cision making leads to reduction in the electricity cost of end user(s) along with minimization of total power losses and alle- viation of peak to average ratio [19]. Keeping these objectives in mind, the current research in SGs mainly focuses on the optimization techniques of power scheduling [20]–[22]. How- ever, prior to scheduling, an accurate short-term load forecasting (STLF) model is needed to properly plan the ongoing grid oper- ations subject to efficient management of resources. High ran- domness and nonlinearity in history load curves make the STLF highly challenging. In the literature, many STLF models have been presented [1], [10], [11], [15], [16]; however, the accuracy of these models is either not satisfactory or their convergence rate is slow. For example, [12] uses a hybrid ANN-based ap- proach to increase the forecast accuracy, however, in doing so the complexity of the overall strategy is increased in terms of implementation, which causes its convergence rate to decrease. In another ANN-based work [10], the convergence rate is im- proved by paying the cost of increased forecast error. In order to solve the accuracy problem of [10], [11] integrates an optimizer with an ANN-based strategy in [10]. However, in doing so, the execution time of [11] increases. (Note: In Section IV, we have compared [11] with our proposed work. Results show that our proposed model takes 52.38% less time to execute than the work in [11]. The work in [12] adds an evolutionary algorithm based module to the work in [11]. This means that [12] will take more time to execute than [11]. That is why we have claimed the execution time of [12] as high.) The contribution in this paper is, first, to improve the rel- ative forecast accuracy of existing ANN-based STLF models, 1551-3203 © 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications standards/publications/rights/index.html for more information.