融合CEEMDAN与CNN-GRU-Attentions的复合模型在碳价格预测中的应用研究

Research on the Application of a Composite Model Integrating CEEMDAN with CNN-GRU-Attentions for Carbon Price Forecasting

摘要: 为了提升碳价格预测的精确性，本研究引入了一种集成模型，结合了自适应噪声完备集合经验模态分解(CEEMDAN)、卷积神经网络(CNN)、改进长短期记忆网络(GRU)和注意力机制(Attentions)。该模型利用GRU的更新门简化了LSTM结构，降低了模型的参数量和复杂度。通过CNN的特征提取能力和GRU对时间序列的捕捉，模型能够模拟碳价格在时间和空间上的依赖性。注意力机制的加入进一步提升了模型对关键历史特征的识别能力。实验结果表明，该CNN-GRU-Attentions模型在拟合度、MAE、MAPE和RMSE等指标上均优于传统模型，具有较高的预测精度和实际应用潜力。

Abstract: To enhance the accuracy of carbon price forecasting, this study introduces an integrated model that combines Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN), Convolutional Neural Networks (CNN), Gated Recurrent Units (GRU), and Attentions mechanisms. The model utilizes the GRU’s update gate to simplify the Long Short-Term Memory (LSTM) structure, reducing the model’s parameter count and complexity. By leveraging the feature extraction capabilities of CNNs and GRU’s capture of time series, the model can simulate the temporal and spatial dependencies of carbon prices. The incorporation of the Attention mechanism further enhances the model’s ability to identify key historical features. Experimental results demonstrate that the CNN-GRU-Attention model outperforms traditional models in terms of fit, Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Root Mean Square Error (RMSE), indicating its higher predictive accuracy and practical application potential.