Research On Prediction Of Carbon Emission Trading Price Based On Deep Learning And Decomposition-ensemble

Excessive emissions of greenhouse gases have caused global warming to increase and natural disasters to occur frequently.In response to climate change,countries around the world have started to take measures to limit carbon emissions in order to curb global warming,and as a tool to effectively reduce carbon emissions,carbon emission rights and carbon emission trading markets have emerged in this process.China is also gradually promoting the construction of the carbon emission trading market.In the process of establishing and developing carbon emission trading market,it is very important to accurately grasp the trend of carbon emission price.However,due to the lack of maturity of China’s carbon emission trading market,carbon prices fluctuate drastically and it is difficult to control the trend of carbon prices.It is necessary to find models and methods that can capture the characteristics of carbon price series and predict carbon prices accurately.In the theoretical analysis section,we review the research literature and related materials on carbon emission price,and introduce the theories on carbon emission price,forecasting methods and carbon emission trading.In the empirical analysis section,we analyze the construction of China’s carbon emission trading market and select the representative market of Guangdong among the eight pilot markets for empirical analysis.After obtaining the daily closing price series of carbon emission rights in Guangdong market,deep learning methods including Recurrent Neural Network(RNN)and Long Short-Term Memory(LSTM)are chosen to model and predict the carbon price series,and Empirical Mode Decomposition(EMD)and Variational Mode Decomposition(VMD)are added to deal with the non-stationary carbon price series based on the idea of decompositionensemble,the original series is decomposed into different subsequences and then predicted separately,then the final prediction results are obtained by summing the predicted values of each subsequence.This paper then changes the ensemble method of subsequences based on the decomposition of carbon price series using VMD,and uses deep learning methods to integrate the prediction results of subsequences,and constructs four hybrid prediction models using this idea.After comparing with the other ten models,it is found that the proposed hybrid prediction method has some advantages in prediction accuracy and can predict the carbon price series more accurately.And the VMD-RNNRNN model performs best on the carbon price data of Guangdong pilot carbon emission trading market among the four hybrid models.The results show that the proposed hybrid prediction method can predict the carbon price more accurately,enrich the methods of carbon price prediction and provide a new idea for decomposition-ensemble model.