Abstract: Because of large price fluctuation in carbon trading market and its complicated influence on carbon price caused by some uncertainty factors, the short-term and the long-term optimal forecasting models are built to forecast the carbon price. Considering the periodicity and?trend of carbon price fluctuation, this paper conducts the short-and the long-term forecasts of China's carbon price based on empirical mode decomposition (EMD), genetic algorithm and back propagation neural net model (GA-BP), particle swarm optimization and least squares support vector machines model(PSO-LSSVM) and their combination models. Different macroeconomic factors affecting carbon price and carbon price time series are respectively employed as input variables into the combination models and two sub-models for forecasting in empirical analysis. The results show that, the EMD-GA-BP model performs better than the GA-BP model and the PSO-LSSVM model in the short-term forecasting, while the combination models (EMD-PSO-LSSVM) perform better than the sub-models in the long-term forecasting.

Key words: carbon price forecasting, empirical mode decomposition, genetic algorithm and back propagation neural net, particle swarm optimization and least squares support vector machines, macroeconomic factors

摘要： 由于碳交易市场价格的波动性大及相互影响关系的复杂性,本文试图构建碳价格长期和短期的最优预测模型。考虑到碳交易价格波动的趋势性和周期性特点,基于经验模态分解算法(EMD)、遗传算法(GA)—神经网络(BP)模型、粒子群算法(PSO)—最小二乘支持向量机(LSSVM)模型及由它们构建的组合预测模型,对中国碳市场交易价格进行短期预测和长期预测。实证分析中将影响碳交易价格的不同宏观经济因素和碳价格时间序列因素做为输入变量,分别代入组合模型进行预测。研究结果表明,在短期预测中,EMD-GA-BP模型预测效果优于GA-BP模型和PSO-LSSVM模型;而在长期预测中,组合模型EMD-PSO-LSSVM模型预测效果优于只考虑碳价格波动趋势性或周期性预测效果。

关键词: 碳价格预测, 经验模态分解算法, 遗传算法—神经网络, 粒子群算法-最小二乘支持向量机, 宏观经济因素