为增强综合能源系统负荷精细化分解水平,充分利用误差信息以进一步提升预测性能,提出一种基于聚合混合模态分解和时序卷积神经网络(temporal convolutional network,TCN)的综合能源系统负荷修正预测框架。首先,采用改进完全集合经验模...为增强综合能源系统负荷精细化分解水平,充分利用误差信息以进一步提升预测性能,提出一种基于聚合混合模态分解和时序卷积神经网络(temporal convolutional network,TCN)的综合能源系统负荷修正预测框架。首先,采用改进完全集合经验模态分解对电、冷和热负荷初步分解处理,随后利用变分模态分解对具有强复杂性的子序列进一步分解。然后,依据最大信息系数(maximum information coefficient,MIC)分析多元负荷的耦合特性并通过多元相空间重构(multivariate phase space reconstruction,MPSR)丰富特征信息。最后,构建基于TCN的修正预测模型。以校园综合能源系统算例对比不同预测模型,结果显示所提修正预测框架的电、冷和热负荷预测均具有较低的平均绝对百分比误差,有效解决了预测中模态分解的模态混叠以及模态高频分量问题,实现预测误差修正。展开更多
Understanding the characteristics of the dynamic relationship between the onshore Ren- minbi （CNY） and the offshore Renminbi （CNH） exchange rates considering the impact of some extreme events is very important and...Understanding the characteristics of the dynamic relationship between the onshore Ren- minbi （CNY） and the offshore Renminbi （CNH） exchange rates considering the impact of some extreme events is very important and it has wide implications in several areas such as hedging. For better esti- mating the dynamic relationship between CNY and CNH, the Granger-causality test and Bry-Boschan Business Cycle Dating Algorithm were employed in this paper. Due to the intrinsic complexity of the lead-lag relationships between CNY and CNH, the empirical mode decomposition （EMD） algorithm is used to decompose those time series data into several intrinsic mode function （IMF） components and a residual sequence, from high to low frequency. Based on the frequencies, the IMFs and a residual sequence are combined into three components, identified as short-term composition caused by some market activities, medium-term composition caused by some extreme events and the long-term trend.The empirical results indicate that when it only matters the short-term market activities, CNH always leads CNY; while the medium-term impact caused by those extreme events may alternate the lead-lag relationships between CNY and CNH.展开更多
文摘为增强综合能源系统负荷精细化分解水平,充分利用误差信息以进一步提升预测性能,提出一种基于聚合混合模态分解和时序卷积神经网络(temporal convolutional network,TCN)的综合能源系统负荷修正预测框架。首先,采用改进完全集合经验模态分解对电、冷和热负荷初步分解处理,随后利用变分模态分解对具有强复杂性的子序列进一步分解。然后,依据最大信息系数(maximum information coefficient,MIC)分析多元负荷的耦合特性并通过多元相空间重构(multivariate phase space reconstruction,MPSR)丰富特征信息。最后,构建基于TCN的修正预测模型。以校园综合能源系统算例对比不同预测模型,结果显示所提修正预测框架的电、冷和热负荷预测均具有较低的平均绝对百分比误差,有效解决了预测中模态分解的模态混叠以及模态高频分量问题,实现预测误差修正。
基金partially supported by the National Natural Science Foundation of China under Grant Nos.71390330,71390331,71390335the National Nature Science Foundation of China for financial support to this study+1 种基金supported by the Postdoctorate Programme of Centre University of Economics and Financethe Postodctorate Programme of China Great Wall Asset Management Corporation
文摘Understanding the characteristics of the dynamic relationship between the onshore Ren- minbi （CNY） and the offshore Renminbi （CNH） exchange rates considering the impact of some extreme events is very important and it has wide implications in several areas such as hedging. For better esti- mating the dynamic relationship between CNY and CNH, the Granger-causality test and Bry-Boschan Business Cycle Dating Algorithm were employed in this paper. Due to the intrinsic complexity of the lead-lag relationships between CNY and CNH, the empirical mode decomposition （EMD） algorithm is used to decompose those time series data into several intrinsic mode function （IMF） components and a residual sequence, from high to low frequency. Based on the frequencies, the IMFs and a residual sequence are combined into three components, identified as short-term composition caused by some market activities, medium-term composition caused by some extreme events and the long-term trend.The empirical results indicate that when it only matters the short-term market activities, CNH always leads CNY; while the medium-term impact caused by those extreme events may alternate the lead-lag relationships between CNY and CNH.
