由於全球暖化所帶來的問題，許多國家已開始實施碳排放限制，這導致碳交易市場的出現，並對精確的碳價預測需求日益增加。在時間序列預測領域中，存在著多種方法來提高預測準確性。由於多種影響因素，預測碳價具有一定的困難性。這些因素包括政治和經濟條件、技術進步、演變中的氣候政策、波動的化石燃料成本、可再生能源替代品的可用性以及氣候政策在減少碳排放方面的有效性。此外，由於碳價機制如配額交易系統和碳稅在許多國家仍屬相對新的政策，因此缺乏準確預測所需的歷史數據。我們的方法著重於有效地整合新聞資訊，通過引入政府政策、市場供需、經濟狀況、能源價格、氣候變化事件和投資者情緒等多個影響因素的有意義的新聞資訊，以增強基於Transformer的時間序列模型的預測能力。為了實現這一目標，我們使用ChatGPT將新聞數據轉化為基於不同觀點的多個強度指標，這些指標有助於預測價格上升。這為模型提供了更多具有資訊量的新聞輸入。此外，我們提出了NTEformer，這是一種新的方式，相較於最廣泛使用的早期連接方法，將文本數據與時間序列模型相結合，並引入了News-Trend Extractor，旨在更好地利用新聞資訊。我們在模型的解碼器中增加了一個News-Trend Extractor，專門用於從新聞數據中學習。通過一系列的實驗，我們將NTEformer與其他結合時間序列數據和文本的方法進行了比較，證明了其優異的性能。

通過有效整合新聞資訊並充分利用新聞趨勢提取器的專業知識，NTEformer模型在預測碳交易價格方面提供了增強的能力，相比不使用新聞資訊的方法，將整體誤差減少了顯著的28％。全面的分析和實驗驗證了我們方法的有效性，展示了其優於其他融合策略的能力。

Due to the problems caused by global warming, many countries have initiated carbon emission restrictions, leading to the emergence of carbon trading markets and a growing demand for accurate carbon price predictions. In the realm of time series forecasting, various methods exist to improve prediction accuracy. Predicting carbon pricing is difficult due to several influential factors. These factors encompass political and economic conditions, technological advancements, evolving climate policies, fluctuating fossil fuel costs, the availability of renewable energy alternatives, and the effectiveness of climate policies in reducing carbon emissions. Furthermore, limited historical data exists for accurate predictions, as carbon pricing mechanisms like cap-and-trade systems and carbon taxes are relatively new policies in many countries. Our approach focuses on effectively incorporating news information to enhance the predictive capabilities of the Transformer-based time series model by introducing meaningful news information based on several influential factors, including government policies, market supply and demand, economic conditions, energy prices, climate change events, and investor sentiment. To achieve this, we employ ChatGPT to transform news data into multiple strength indicators based on different perspectives that contribute to price increases. This provides us with more informative news inputs for the model. Furthermore, we propose NTEformer, a novel way, compared to the most widely used early concatenation method, to combine text data into time series model with News-Trend Extractor, designed to better leverage news information. We augment the decoder of the model with a News-Trend Extractor, specifically designed to learn from news data. Through a series of experiments, we compare NTEformer with other methods that combine time series data and text, demonstrating its superior performance.

By effectively integrating news information and leveraging the expertise of the News-Trend Extractor, the NTEformer model offers enhanced capabilities in predicting carbon trading prices, reducing the overall error by a remarkable 28% compared to the method without using news information. The comprehensive analysis and experiments validate the effectiveness of our approach, showcasing its ability to outperform other fusion strategy.

[1] D. Salinas, V. Flunkert, and J. Gasthaus, “Deepar: Probabilistic forecasting with autoregressive recurrent networks,” Feb. 2019.
[2] G.Lai,W.-C.Chang,Y.Yang,andH.Liu,“Modelinglong-andshort-termtemporal patterns with deep neural networks,” Apr. 2018.
[3] A. Borovykh, S. Bohte, and C. W. Oosterlee, “Conditional time series forecasting with convolutional neural networks,” Sep. 2018.
[4] R. Sen, H.-F. Yu, and I. Dhillon, “Think globally, act locally: A deep neural network approach to high-dimensional time series forecasting,” Oct. 2019.
[5] H. Wu, J. Xu, J. Wang, and M. Long, “Autoformer: Decomposition transformers with auto-correlation for long-term series forecasting,” Jan. 2022.
[6] H. Zhou, S. Zhang, J. Peng, S. Zhang, J. Li, H. Xiong, and W. Zhang, “Informer: Beyond efficient transformer for long sequence time-series forecasting,” Mar. 2021.
[7] G. Woo, C. Liu, D. Sahoo, A. Kumar, and S. Hoi, “Etsformer: Exponential smooth- ing transformers for time-series forecasting,” Jun. 2022.
[8] T. Zhou, Z. Ma, Q. Wen, X. Wang, L. Sun, and R. Jin, “Fedformer: Frequency enhanced decomposed transformer for long-term series forecasting,” Jun. 2022.
[9] Y. Liu, H. Wu, J. Wang, and M. Long, “Non-stationary transformers: Exploring the stationarity in time series forecasting,” Oct. 2022.
[10] D. Du, B. Su, and Z. Wei, “Preformer: Predictive transformer with multi-scale segment-wise correlations for long-term time series forecasting,” Feb. 2022.
[11] Z.Liu,Y.Li,andH.Liu,“Fuzzytime-seriespredictionmodelbasedontextfeatures and network features,” Neural Computing and Applications, Mar. 2021.
[12] N. Kanungsukkasem and T. Leelanupab, “Financial latent dirichlet allocation (finlda): Feature extraction in text and data mining for financial time series pre- diction,” IEEE Access, vol. 7, pp. 71 645–71 664, 2019.
[13] K. Keswani, I. Das, B. Shrivastava, A. Gupta, and R. Katarya, “Lda based model for mining textual features from financial news articles,” in 2020 2nd International Conference on Advances in Computing, Communication Control and Networking (ICACCCN). Greater Noida, India: IEEE, Dec. 2020, pp. 43–48.
[14] P.Xu,X.Zhu,andD.A.Clifton,“Multimodallearningwithtransformers:Asurvey,” Jun. 2022.
[15] A. Lopez-Lira and Y. Tang, “Can chatgpt forecast stock price movements? return predictability and large language models,” Rochester, NY, Apr. 2023.
[16] Y. Hao, C. Tian, and C. Wu, “Modelling of carbon price in two real carbon trading markets,” Journal of Cleaner Production, vol. 244, p. 118556, 2020.
[17] L.-T. Zhao, J. Miao, S. Qu, and X.-H. Chen, “A multi-factor integrated model for carbon price forecasting: Market interaction promoting carbon emission reduction,” Science of The Total Environment, vol. 796, p. 149110, 2021.
[18] M. Yahşi, E. Çanakoğlu, and S. Ağralı, “Carbon price forecasting models based on big data analytics,” Carbon Management, vol. 10, no. 2, pp. 175–187, 2019.
[19] B. Box, G. Jenkins, G. Reinsel, and G. Ljung, Time Series Analysis: Forecasting and Control, Jan. 2016, vol. 68.
[20] A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin, “Attention is all you need,” Dec. 2017.
[21] S. Li, X. Jin, Y. Xuan, X. Zhou, W. Chen, Y.-X. Wang, and X. Yan, “Enhancing the locality and breaking the memory bottleneck of transformer on time series forecast- ing,” Jan. 2020.
[22] Y. Yang and J. Lu, “A fusion transformer for multivariable time series forecasting: The mooney viscosity prediction case,” Entropy, vol. 24, no. 4, p. 528, Apr. 2022.
[23] D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” Jan. 2017.