EUA Price Prediction Based On PCA-TVF-EMD-MPA-LSTM Model

Since the industrial revolution,with the increasing emission of greenhouse gases,it has triggered a serious global warming problem.How to take effective measures to control it reasonably is crucial.In the past few years,carbon trading has become an effective measure to combat global climate change,and the global carbon market represented by the EU ETS has been quite effective in reducing emissions.Therefore,it is very important to forecast the EU carbon price.Currently,EUA futures are the most liquid and traded carbon credit product in the EU carbon market,accounting for more than 85%of the total EUA trading volume.Therefore,EUA futures price is more representative than EUA spot price,so this paper selects EUA futures price(referred to as"EUA price")to represent the carbon price of EU ETS market for prediction.From the analysis of the theory of supply and demand,a total of 24 influencing factors are selected from five aspects:supply,macroeconomy,international exchange rate,energy price,climate environment,and electricity market.Three mainstream neural network models for temporal forecasting,BP neural network model,extreme learning machine model(ELM),and long and short term memory model(LSTM),are used to forecast EUA prices respectively,and the main forecasting model of this paper is selected by comparison.For the first time,time-varying filter-based empirical modal decomposition(TVF-EMD)is introduced to decompose EUA price data,and principal component analysis(PCA)is used to reduce the dimensionality of EUA price influencing factors.Then,this paper innovatively introduces the advanced biological optimization algorithm,Marine Predator Algorithm(MPA),to optimize the parameters of the LSTM model.Finally,the hybrid prediction model of PCA-TVF-EMD-MPA-LSTM is used to forecast the EUA price.To verify the robustness of this hybrid model,this paper also forecasts the carbon price of Chongqing city.The empirical results show that.(1)The LSTM model has the lowest mean absolute percentage error(MAPE)among BP,ELM,and LSTM for EUA price prediction and performs better,reflecting the advantage of the LSTM model in the processing of time-series data.(2)The prediction accuracy of LSTM models can be improved by using signal decomposition methods to process the prediction data.Compared with the LSTM model,the TVF-EMD-LSTM model reduces the MAPE error by 44%and improves the directional prediction accuracy by 13%,reflecting the advantages of the TVF-EMD method in signal decomposition processing.(3)The use of PCA dimensionality reduction for impact factor indicators can improve the accuracy of the prediction model.Compared with the TVF-EMD-LSTM model,the PCA-TVF-EMD-LSTM model using the PCA dimensionality reduction method reduces the MAPE error by 43%and improves the directional prediction accuracy by about 10%,reflecting the advantages of the PCA dimensionality reduction method in data processing.(4)The hyperparameter optimization of the prediction model using MPA optimization algorithm can significantly improve the accuracy of the prediction model.In this paper,we innovatively introduce the MPA algorithm to optimize the LSTM model and construct a PCA-TVF-EMD-MPA-LSTM hybrid model to forecast EUA prices.Compared with the PCA-TVF-EMD-LSTM model,the PCA-TVF-EMD-MPA-LSTM hybrid model reduces the MAPE error by 39%and improves the directional prediction accuracy by 11%.(5)The PCA-TVF-EMD-MPA-LSTM hybrid model constructed in this paper is more robust and still applicable to the prediction of carbon emission rights prices in the Chinese carbon emission rights market,and the introduction of MPA optimization is still effective in improving the prediction model.Compared with the PCA-TVF-EMD-LSTM model,the PCA-TVF-EMD-MPA-LSTM model constructed in this paper significantly reduces the MAPE error by about 25%and improves the accuracy of the directional prediction by about 13%for the prediction of carbon emission rights price in Chongqing.