Main content area

A novel method for carbon dioxide emission forecasting based on improved Gaussian processes regression

Fang, Debin, Zhang, Xiaoling, Yu, Qian, Jin, Trenton Chen, Tian, Luan
Journal of cleaner production 2018 v.173 pp. 143-150
algorithms, carbon, carbon dioxide, covariance, emissions, global warming, humans, immigration, issues and policy, methodology, neural networks, normal distribution, prediction, regression analysis, China, Japan, United States
The fact that global warming will bring impact on immigration, agriculture and also generate human conflicts is becoming a focus in climate change topic and the forecasting of carbon dioxide emission has been attracting much attention. In this paper, we proposed an improved Gaussian processes regression method for carbon dioxide emission forecasting based on a modified PSO algorithm which can efficiently optimize the hyper parameters of covariance function in the Gaussian processes regression. Also we tested our improved PSO-GPR method with the total carbon dioxide emissions data of U.S., China and Japan in 1980–2012, and compared the prediction precision of our method with original GPR and BP Neural Networks by the data of U.S., China and Japan. The performance of our improved Gaussian processes regression method enhanced the prediction accuracy of original GPR method and is superior to other traditional forecasting method like BP Neural Networks. Furthermore, we applied PSO-GPR method in generating a prediction total carbon dioxide emissions for 2013 to 2020 and found out that China's total carbon dioxide emission will still increasing but finally at a decreasing rate and U.S. and Japan will have a good control on their amount of carbon emission in the near future. Finally, policy implications about carbon dioxide emission reduction were proposed.