A CNN-LSTM based deep learning model with high accuracy and robustness for carbon price forecasting: A case of Shenzhen's carbon market in China

Accurately predicting carbon trading prices using deep learning models can help enterprises understand the operational mechanisms and regulations of the carbon market. This is crucial for expanding the industries covered by the carbon market and ensuring its stable and healthy development. To ensure the accuracy and reliability of the predictions in practical applications, it is important to evaluate the model's robustness. In this paper, we built models with different parameters to predict carbon trading prices, and proposed models with high accuracy and robustness. The accuracy of the models was assessed using traditional survey indicators. The robustness of the CNN-LSTM model was compared to that of the LSTM model using Z-scores. The CNN-LSTM model with the best prediction performance was compared to a single LSTM model, resulting in a 9% reduction in MSE and a 0.0133 shortening of the Z-score range. Furthermore, the CNN-LSTM model achieved a level of accuracy comparable to other popular models such as CEEMDAN, Boosting, and GRU. It also demonstrated a training speed improvement of at least 40% compared to the aforementioned methods. These results suggest that the CNN-LSTM enhances model resilience. Moreover, the practicality of using Z-score to evaluate model robustness is confirmed.

Carbon price prediction based on multiple decomposition and XGBoost algorithm

Carbon trading is an effective way to limit global carbon dioxide emissions. The carbon pricing mechanisms play an essential role in the decision of the market participants and policymakers. This study proposes a carbon price prediction model, multi-decomposition-XGBOOST, which is based on sample entropy and a new multiple decomposition algorithm. The main steps of the proposed model are as follows: (1) decompose the price series into multiple intrinsic mode functions (IMFs) by using complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN); (2) decompose the IMF with the highest sample entropy by variational mode decomposition (VMD); (3) select and recombine some IMFs based on their sample entropy, and then perform another round of decomposition via CEEMDAN; (4) predict IMFs by XGBoost model and sum up the prediction results. The model has exhibited reliable predictive performance in both the highly fluctuating Beijing carbon market and the comparatively stable Hubei carbon market. The proposed model in Beijing carbon market achieves improvements of 30.437%, 44.543%, and 42.895% in RMSE, MAE, and MAPE, when compared to the single XGBoost models. Similarly, in Hubei carbon market, the RMSE, MAE, and MAPE based on multi-decomposition-XGBOOST model decreased by 28.504%, 39.356%, and 39.394%. The findings indicate that the proposed model has better predictive performance for both volatile and stable carbon prices.

Carbon trading and COVID-19: a hybrid machine learning approach for international carbon price forecasting

Accurate carbon price forecasting can better allocate carbon emissions and thus ensure a balance between economic development and potential climate impacts. In this paper, we propose a new two-stage framework based on processes of decomposition and re-estimation to forecast prices across international carbon markets. We focus on the Emissions Trading System (ETS) in the EU, as well as the five main pilot schemes in China, spanning the period from May 2014 to January 2022. In this way, the raw carbon prices are first separated into multiple sub-factors and then reconstructed into factors of 'trend' and 'period' with the use of Singular Spectrum Analysis (SSA). Once the subsequences have been thus decomposed, we further apply six machine learning and deep learning methods, allowing the data to be assembled and thus facilitating the prediction of the final carbon price values. We find that from amongst these machine learning models, the Support vector regression (SSA-SVR) and Least squares support vector regression (SSA-LSSVR) stand out in terms of performance for the prediction of carbon prices in both the European ETS and equivalent models in China. Another interesting finding to come out of our experiments is that the sophisticated algorithms are far from being the best performing models in the prediction of carbon prices. Even after accounting for the impacts of the COVID-19 pandemic and other macro-economic variables, as well as the prices of other energy sources, our framework still works effectively.

The role of online news sentiment in carbon price prediction of China's carbon markets

Carbon trading as a vital tool to reduce carbon dioxide emissions has developed rapidly in recent years. Reasonable prediction of the carbon price can improve the risk management in the carbon trading market and make healthy development of the carbon trading market. This paper aims to enhance the predictive performance of carbon price in the China's carbon markets, especially the China's national carbon market, by adding the online news sentiment index which is a kind of unconstructed data, to a deep learning model using traditionally constructed predictors innovatively. Long short-term memory (LSTM) network was applied as the primary model to predict carbon price and random forest as the additional experiment to validate the effectiveness of online news sentiment. The results in the China's national carbon market and Hubei pilot carbon market both proved that the model including the sentiment index performed better than the model does not, and the improvement was significant.

Carbon price prediction based on multi-factor MEEMD-LSTM model

China's national carbon market has already become the largest carbon market in the world. The prediction of carbon price is extremely important for policymakers and market participants. Therefore, the prediction of carbon price in China is of great significance. To achieve a better prediction effect, a multi-factor hybrid model combined with modified ensemble empirical mode decomposition (MEEMD) and long short-term memory (LSTM) neural network optimized by machine reasoning system on the basis of production rules is proposed in this paper. In addition to historical carbon price, other factors, such as energy, macroeconomy, environmental condition, temperature, exchange rate which affect carbon price fluctuation, are formed as multi-factor. The change characteristics of carbon price time series data and other associated factors are extracted in the carbon price prediction. The MEEMD is used to decompose data which is taken as potential input variables into LSTM neural network for prediction and the machine reasoning system based on production rules can automatically search and optimize the parameters of LSTM to further improve the prediction results. The experimental results demonstrate that the proposed method has better prediction effect, robustness and adaptability than the LSTM model without MEEMD decomposition, the single factor MEEMD-LSTM method and other benchmark models. Overall it seems that the proposed method is an advanced approach for predicting the non-stationary and non-linear carbon price time series.

A novel hybrid learning paradigm with feature extraction for carbon price prediction based on Bi-directional long short-term memory network optimized by an improved sparrow search algorithm

An efficient carbon trading market can effectively curb excessive carbon emissions and thus slow down the pace of global warming, which heightens the necessity of improving the accuracy of carbon price forecasting. In order to overcome the weakness of previous prediction model that always trained data in one-way neural networks and propagated the data sequentially, this paper proposes a novel hybrid learning paradigm WPD-ISSA-BiLSTM combining wavelet packet decomposition (WPD), improved sparrow search algorithm (ISSA), and Bi-directional long short-term memory network for deep feature exploration of carbon prices. Firstly, WPD decomposes and reconstructs the original carbon price series into several independent subseries. Then, the input features of the all subseries are filtered with random forest to select the best input features for the prediction model. Finally, a Bi-directional long short-term memory network optimized by the ISSA is employed to deeply delineate the intrinsic evolutionary trends of carbon prices, and the prediction results of all subseries are superimposed on each other to obtain the final carbon price prediction results. The actual carbon emission trading prices are collected as input to the model, and the experimental results show that the RMSE values of the proposed model are 0.2516 and 0.2962 under the mild and severe volatility scenarios, respectively. The proposed model has superiority and robustness compared to the comparison model and several existing models and better understands the intrinsic correlation between historical carbon price data. The results of this study can provide meaningful references for the carbon market development and emission reduction pathways.

Carbon price forecasting: a novel deep learning approach

Carbon emission trading market promotes carbon emission reduction effectively. Accurate carbon price forecasting is crucial for relevant policy makers and investors. However, due to the non-linearity, uncertainty, and complexity of carbon prices, the current predication models fail to predict carbon prices accurately. In this paper, an advanced deep neural network model named TCN-Seq2Seq is proposed to forecast carbon prices. The novelty of the proposed model focuses on the "sequence to sequence" layout to learn temporal data dependencies using only fully convolutional layers. Being provided with parallel training for fewer parameters, TCN-Seq2Seq forecasting model is more suitable for small carbon price dataset in few-shot learning way. Qualitatively and quantitatively, we find that the proposed framework consistently and significantly outperforms traditional statistical forecasting models and state-of-the-art deep learning prediction model with respect to predictive ability and robustness. Particularly, our proposed model achieves forecasting accuracy with the highest DA value (0.9697), the lowest MAPE value (0.0027), and the lowest RMSE value (0.0149), showing superior prediction performance compared with the traditional statistical forecasting models. The accuracy of carbon price forecasting gives insight to policy makers and carbon market investors.

An innovative random forest-based nonlinear ensemble paradigm of improved feature extraction and deep learning for carbon price forecasting

Carbon price is the basis of developing a low carbon economy. The accurate carbon price forecast can not only stimulate the actions of enterprises and families, but also encourage the study and development of low carbon technology. However, as the original carbon price series is non-stationary and nonlinear, traditional methods are less robust to predict it. In this study, an innovative nonlinear ensemble paradigm of improved feature extraction and deep learning algorithm is proposed for carbon price forecasting, which includes complete ensemble empirical mode decomposition (CEEMDAN), sample entropy (SE), long short-term memory (LSTM) and random forest (RF). As the core of the proposed model, LSTM enhanced from the recurrent neural network is utilized to establish appropriate prediction models by extracting memory features of the long and short term. Improved feature extraction, as assistant data preprocessing, represents its unique advantage for improving calculating efficiency and accuracy. Removing irrelevant features from original time series through CEEMDAN lets learning easier and it's even better for using SE to recombine similar-complexity modes. Furthermore, compared with simple linear ensemble learning, RF increases the generalization ability for robustness to achieve the final nonlinear output results. Two markets' real data of carbon trading in china are as the experiment cases to test the effectiveness of the above model. The final simulation results indicate that the proposed model performs better than the other four benchmark methods reflected by the smaller statistical errors. Overall, the developed approach provides an effective method for predicting carbon price.

Carbon price prediction based on modified wavelet least square support vector machine

It is widely believed that setting a sensible carbon price can contribute to the mitigation of global warming, so it is particularly major to raise the precision of carbon price prediction. As such it has important implications not only for beautifying the environment but also for promoting the benign development of the carbon trading market in China. However, consideration is given to the high non-determinacy and non-linearity of the carbon price series, a single model cannot meet the prediction accuracy anymore. Since this is the case, this paper puts forward a novel hybrid forecasting model, consisting of the ensemble empirical mode decomposition (EEMD), the linearly decreasing weight particle swarm optimization (LDWPSO), and the wavelet least square support vector machine (wLSSVM). Innovatively, wLSSVM is utilized in the field of carbon price prediction for the first time. Firstly, EEMD decomposes the raw carbon price into several stable sub-sequences and a residual. Then, the inputs of each sequence are determined by the partial auto-correlation function (PACF). Next, wLSSVM optimized by LDWPSO forecasts each sequence separately. Finally, the final prediction result is obtained by adding all prediction results. For the purpose of verifying the effectiveness and superiority of the EEMD-LDWPSO-wLSSVM model, a total of 12 models were built to compare their performance in three regions of Guangdong, Hubei, and Shanghai respectively from three evaluating indicators: MAPE, RMSE, and R2. All the predicted results showed that the model presented in this paper has the best forecasting performance among all the model combinations and can substantially improve the accuracy of carbon price prediction. Therefore, the model would be an increasingly extensive application in the field of carbon price prediction in the future.