Scholarlink

<p>Carbon price volatility is a critical indicator of the carbon trading market, characterized by its phase-wise variation. However, conventional forecasting methods frequently struggle to address the substantial noise and non-linear dynamics inherent in these sequences. To address these challenges, this study proposes a novel hybrid model, CEEMDAN-At-LSTM, which combines complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and an attention-enhanced long short-term memory network (LSTM). The model decomposes complex volatility sequences and captures hidden temporal features within each subsequence. Forecasting performance was evaluated using metrics including MAE, MSE, HMAE, and HMSE, and further validated through the model confidence set (MCS) test to ensure robustness. Experiments were conducted on datasets from the Hubei Carbon Emissions Trading Center and the EU ETS, representing emerging and developed markets, respectively. CEEMDAN-At-LSTM achieved at least a 38.87% reduction in forecasting error compared to baseline models, demonstrating its effectiveness in managing nonlinearity and noise. This approach provides a reliable framework for carbon market volatility forecasting, supporting informed policymaking and the advancement of a low-carbon economy.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Carbon price volatility prediction using a hybrid CEEMDAN-attention-LSTM approach

  • 期刊论文

  • 发表日期：

    2025年6月25日

查看全文
  • Junxuan Yao,
  • Heping Wang,
  • Suzhen Mao

摘要

Carbon price volatility is a critical indicator of the carbon trading market, characterized by its phase-wise variation. However, conventional forecasting methods frequently struggle to address the substantial noise and non-linear dynamics inherent in these sequences. To address these challenges, this study proposes a novel hybrid model, CEEMDAN-At-LSTM, which combines complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and an attention-enhanced long short-term memory network (LSTM). The model decomposes complex volatility sequences and captures hidden temporal features within each subsequence. Forecasting performance was evaluated using metrics including MAE, MSE, HMAE, and HMSE, and further validated through the model confidence set (MCS) test to ensure robustness. Experiments were conducted on datasets from the Hubei Carbon Emissions Trading Center and the EU ETS, representing emerging and developed markets, respectively. CEEMDAN-At-LSTM achieved at least a 38.87% reduction in forecasting error compared to baseline models, demonstrating its effectiveness in managing nonlinearity and noise. This approach provides a reliable framework for carbon market volatility forecasting, supporting informed policymaking and the advancement of a low-carbon economy.