Research On Combined Model Of Carbon Trading Price Forecast Based On CEEMDAN

To deal with global warming,the right to emit carbon dioxide is regarded as a kind of commodity in the Kyoto Protocol for the first time.Carbon trading price study is of great significance in restraining global warming,helping the healthy development of carbon market,providing reference for the formulation of relevant policies by the state and the participation of enterprises.In this paper,the bibliometric method and knowledge graph are used to provide data support for the topic selection,and selecting carbon trading price influence factors from multiple dimensions such as structured and unstructured,domestic and foreign.Through experiments,a combined prediction model that integrates deep learning and machine learning—LSTM-LSTM-LGBM combined prediction model is proposed.The main research contents and conclusions are as follows:(1)In this paper,Cite Space is used to analyze the documents related to carbon trading in the core database of the WOS platform,and finds that many researchers have been keeping on investigating carbon trading,and it is in a period of rapid development,showing the characteristics of multidisciplinary cross-discipline.It is found that carbon trading price research is a hot topic and also a future research trend,which provide data support for topic selection.(2)In this paper,the closing price of Guangdong Carbon Emission Allowance(GDEA)is taken as the research object.First,13 possible structural influencing factors are selected from macroeconomics,energy prices,exchange rates,international carbon trading prices,etc.,and then searches the index from the Internet Selected 36 relevant influencing factors,and the 11 most influential factors were obtained through Lasso algorithm,including 6 structural factors and 5 network search index factors.(3)Adopt the decomposition-prediction-integration concept,use CEEMDAN to solve the problems of spectrum leakage and modal aliasing effects existing in mainstream EMD models,and then reconstruct the decomposed carbon trading into three components by Fine-toCoarse algorithm,which are high-frequency,low-frequency and trend items.Prophet,LSTM and LGBM are proposed to predict carbon trading price,and the optimal combination prediction model is obtained through random combination,which is LSTM-LSTM-LGBM model.The model that only considers structural influencing factors is compared and analyzed with the prediction accuracy of the model added to the web search index,and it is found that adding the web search index can improve the prediction accuracy.A comparative analysis of the single model and the combined forecasting model found that in the single forecasting model,LGBM,based on the decision tree idea is the best model,and the Prophet is the worst model in this experiment.The possible reason is that the periodicity of the fluctuation of carbon trading prices is not good.The LSTM-LSTM-LGBM combined prediction model has significantly better accuracy than a single prediction model.In conclusion,based on CEEMDAN decomposition algorithm,this paper proposes LSTMLSTM-LGBM combined prediction model,and takes Guangdong pilot carbon trading price as an example to carry out simulation experiment and analysis,which proves that the combined prediction model has a good prediction effect and can provide reference for carbon trading market price prediction.