A hybrid intelligent algorithm for optimum forecasting of CO2 emission in complex environments: the cases of Brazil, Canada, France, Japan, India, UK and US

Abstract

This study presents a hybrid meta-modeling algorithm for optimum carbon dioxide (CO2) emission estimation. It is composed of artificial neural network (ANN), fuzzy linear regression (FLR), and conventional regression (CR). Different FLR models are considered to cover the latest algorithms and viewpoints. ANN with different training algorithms and transfer functions is also applied to data sets. The proposed hybrid algorithms uses analysis of variance (ANOVA), and mean absolute percentage error (MAPE) to select between ANN, FLR or conventional regression for future CO2 emission estimation. The intelligent algorithm of this study is then applied to estimate CO2 emission in seven countries including India, Canada, Brazil, France, Japan, United Kingdom and United States. Different models are selected as preferred model for annual CO2 emission estimation in these countries. The preferred model for India, Brazil, United Kingdom and United States is selected as FLR whereas the preferred model for CO2 emission estimation in Japan, Canada and France is ANN. This shows how adopting the proposed hybrid algorithm could help in selecting the preferred model between FLR, ANN and CR in order to cover possible noise, complexity and ambiguity. This is the first study that utilizes a hybrid algorithm based on ANN, FLR and CR for accurate and optimum long term CO2 emission estimation.