多情景视角下中国能源消费和碳达峰路径

准确预测能源消费及碳排放量对于科学有序落实我国"2030年前碳达峰,2060年前碳中和"目标有重要现实意义.提出了一种融合位置扰动和模拟退火的改进粒子群算法（IPSO）优化基于两层"分解-集成"策略的预测方法：首先利用趋势分解（TD）将原始能源消费时序分解成趋势项和非趋势项,继而使用经验模态分解（EMD）将非趋势项分解成若干本征模态函数（IMFs）和一个残差项,然后对上述趋势项、IMFs和残差项分别建模预测,利用IPSO优化多元线性回归模型（MLR）预测趋势项,采用长短期记忆神经网络（LSTM）预测非趋势项的本征模态函数IMFs和残差子序列,最后通过相加集成求取最终能源消费预测值.实证分析表明,基于TD-EMD两层"分解-集成"策略的IPSO-MLR-LSTM模型融合了TD、EMD、IPSO和LSTM的优点,更全面地捕捉了趋势项和非趋势项演化规律,提升了预测性能,将其应用于能源消费领域是可行且有效的.最后,预测了在不同能源消费结构、经济增长、人口数量、能源效率和人均生活能源消费水平情景下的我国2021~2035年能源消费和碳排放量,并给出相关政策建议.

China's Energy Consumption and Carbon Peak Path Under Different Scenarios

Accurately predicting energy consumption and carbon emission is important for China to make energy and carbon emission policy formulation more scientific and to achieve the goal of carbon peak before 2030 and carbon neutrality before 2060. Since energy demand is affected by numerous complex factors, it is hard to capture the dynamically developing rules of energy consumption comprehensively. Therefore, a novel two-layer decomposition-ensemble forecasting approach that was optimized by an improved particle swarm optimization algorithm based on simulation anneal and position disturbance strategy (IPSO) was proposed. Firstly, trend decomposition (TD) was utilized to break energy consumption time series down into a trend and a non-trend subseries. Then, empirical mode decomposition (EMD) was adopted to break the non-trend subseries down into several intrinsic mode functions (IMFs) and a residuum subseries. Subsequently, the aforementioned trend subseries, intrinsic mode functions, and residuum series were respectively modeled for prediction. The trend subseries was predicted using the multivariate linear regression model (MLR), which was optimized using IPSO. Both IMFs and residuum series were predicted using long short-term memory (LSTM). Finally, the final prediction of energy consumption was obtained by integrating the forecasting results of these subseries. According to China''s energy consumption empirical analysis, the proposed IPSO-MLR-LSTM forecasting model based on the two-layer decomposition-ensemble approach using TD-EMD combined the advantages of TD, EMD, IPSO, and LSTM, which could comprehensively extract the developing rules of energy consumption by implementing a deeper decomposition strategy. Therefore, it is feasible and effective to apply the proposed forecasting model for energy consumption prediction. Finally, the energy consumption and carbon emissions of China under different energy consumption structure, economic growth, population, energy efficiency, and household energy consumption per capita scenarios in 2021-2035 were predicted. Then, some relevant policies and suggestions were put forward based on the forecasting results.