基于TCN-FEDformer的碳排放权交易价格预测

Carbon Emission Allowance Trading Price Forecasting Based on TCN-FEDformer Model

摘要: 碳排放权交易价格的准确预测对于完善碳市场运行机制、引导企业减排决策至关重要。本文提出了一种TCN-FEDformer碳排放权交易价格预测模型。该模型将时序卷积网络(Temporal Convolutional Networks, TCN)与频率增强分解Transformer (Frequency Enhanced Decomposed Transformer, FEDformer)进行并行建模，并引入多层感知机(Multi-Layer Perceptron, MLP)对两路特征进行融合，从而实现对碳排放权交易价格的有效预测。基于湖北和广东碳排放权交易市场的实际数据，对构建的模型进行了系统的实证分析。实验结果表明，TCN-FEDformer模型在预测精度和稳定性方面均优于传统时序预测模型与Transformer架构的相关模型，Diebold-Mariano (DM)检验结果也进一步验证了该模型预测性能的提升在统计意义上显著。

Abstract: Accurate forecasting of carbon emission allowance trading prices is of critical importance for improving the operational efficiency of carbon markets and guiding corporate emission reduction decisions. This study proposes a TCN-FEDformer carbon emission allowance trading price prediction model. The model performs parallel modeling using Temporal Convolutional Networks (TCN) and the Frequency Enhanced Decomposed Transformer (FEDformer), and introduces a multilayer perceptron (MLP) to fuse the features from the two branches, thereby enabling effective prediction of carbon emission trading prices. Based on actual data from the Hubei and Guangdong carbon emission allowance trading markets, a systematic empirical analysis is conducted to evaluate the performance of the proposed model. The results demonstrate that the TCN-FEDformer model outperforms traditional time-series forecasting models and Transformer-based models in terms of both predictive accuracy and stability. Furthermore, the Diebold-Mariano (DM) test confirms that the improvement in forecasting performance is statistically significant.