摘要
碳排放的趋势分析与短期预测为制定有效减排计划提供了科学依据,然而受排放来源多且规律复杂的影响,经测算的碳排放量序列具有高噪声、非线性和非平稳的特性,其发展趋势难以被捕捉,对预测工作造成了阻碍。针对碳排放序列变化复杂难以被分析的问题,引入变分模态分解提取序列中不同频率的规律。随后,以时间尺度为视角对子模态进行了重构,用于对CO_(2)日排放内部规律的分析。最后,基于长短期记忆人工神经网络模型实现碳排放趋势的短期预测。通过中国CO_(2)日排放量实例验证,结果表明:新模型可有效提取复杂碳排放序列中不同时间尺度的变化趋势,有助于精准把握排放的发展趋势;相较于其他对比模型,新模型的MAPE和RMSE平均提高了15.65%和20.55%;与多变量预测模型相比较,一维时间序列预测模型受因素复杂性影响较小,预测性能更优。新模型实现了对碳排放发展趋势的有效捕捉及精确预测,可为科学决策提供定量依据。
The trend analysis and short-term forecasting of carbon emissions provide scientific basis for the formulation of effective emission reduction plans.However,effected by the multiple emission sources and complicated development regulars,the carbon emission time series owns the characteristics of high noise,nonlinear and non-stationary.Its development trend is difficult to be captured,which causes obstacles to the forecasting.To address those issues,variational mode decomposition is introduced to extract the regulars with different frequency.Subsequently,the decomposed modes are reconstructed from the perspective of time scale,which is beneficial to analyze the internal trend of carbon emission.Finally,the short-term forecasting is completed via long short-term memory model.According to the application of China’s daily carbon dioxide emissions,the results show that:The proposed model effectively extract the development trend with different time scales in the complicated carbon emission time series,assisting people to capture the development trend precisely.Moreover,compared with other comparison models,the MAPE and RMSE of the proposed model are improved by 15.65%and 20.55%.Compared with the multi-variable prediction model,the one-dimensional time series prediction model is less affected by the complexity of factors and has better prediction performance.The proposed model has realized effective capture and accurate forecasting for the development trend of carbon emissions,which can provide quantitative benchmark for scientific decision-making.
作者
李柚洁
杨萍
赵顺昱
王业林
Li Youjie;Yang Ping;Zhao Shunyu;Wang Yelin(Faculty of Management and Economics,Kunming University of Science and Technology,Kunming 650093,China)
出处
《环境工程》
CAS
CSCD
北大核心
2023年第S02期279-285,249,共8页
Environmental Engineering
关键词
变分模态分解
碳排放
LSTM
组合预测
趋势分析
variational mode decomposition
carbon emission
LSTM
hybrid forecasting
trend analysis
