Forecasting based on an ensemble Autoregressive Moving Average - Adaptive neuro - Fuzzy inference system – Neural network - Genetic Algorithm Framework

• Published in: Energy (Oxford) Vol. 197; p. 117159
• Main Authors: Prado, Francisco, Minutolo, Marcel C., Kristjanpoller, Werner
• Format: Journal Article
• Language: English
• Published: Oxford Elsevier Ltd 15.04.2020; Elsevier BV
• Subjects: Accuracy; Adaptive neuro fuzzy inference system; Adaptive systems; Algorithms; Artificial intelligence; Artificial neural network; Artificial neural networks; Autoregressive moving average; econometric models; Econometrics; Economic forecasting; energy; Energy consumption forecasting; Energy demand; Errors; Forecasting; Forecasting techniques; Fuzzy logic; Genetic algorithm; Genetic algorithms; Inference; Learning algorithms; Lower bounds; Machine learning; Mathematical models; Model accuracy; Neural networks; Optimization; Post-production processing; prediction; regression analysis; Support vector machines; Upper bounds; Artificial neural network; Adaptive neuro fuzzy inference system; Energy consumption forecasting; Genetic algorithm
• ISSN: 0360-5442; 1873-6785
• DOI: 10.1016/j.energy.2020.117159

This paper proposes a novel ensemble methodology comprising an auto regressive integrated moving average, artificial neural network, fuzzy inference system model, adaptive neuro fuzzy inference system, support vector regression, extreme machine learning, and genetic algorithm to forecast aggregated, long-term energy demand. After comparing the framework with several benchmark methods by the loss functions mean squared error and mean absolute percentage error, and applying a model confidence set this work suggests that the proposed method improves forecasting accuracy over previous approaches. The proposed approach resulted in a mean squared error decrease of 22.3% and mean absolute percentage error by 33.1% with respect to the best artificial intelligence and econometric models in a sample study. Post-processing optimization of the forecasting ensemble in this methodology improves prediction accuracy. The approach developed herein provides an addition to the field for how hybridized models and augmented forecasting accuracy can be improved. Continued improvements to forecasting techniques are extremely important especially in areas where there are upper bound constraints on supply and lower bound on minimum operation levels. •Econometric and artificial intelligence methods results in improved forecasting models.•Post processing optimization of the ensemble methods of forecasting improve prediction accuracy.•The framework results outperform benchmark models.•The accuracy of the primary energy consumption improved by 22.3%.