An accurate long-term energy demand forecasting is essential for energy planning and policy making. However, due to the immature energy data collecting and statistical methods, the available data are usually limited i...An accurate long-term energy demand forecasting is essential for energy planning and policy making. However, due to the immature energy data collecting and statistical methods, the available data are usually limited in many regions. In this paper, on the basis of comprehensive literature review, we proposed a hybrid model based on the long-range alternative energy planning (LEAP) model to improve the accuracy of energy demand forecasting in these regions. By taking Hunan province, China as a typical case, the proposed hybrid model was applied to estimating the possible future energy demand and energy-saving potentials in different sectors. The structure of LEAP model was estimated by Sankey energy flow, and Leslie matrix and autoregressive integrated moving average (ARIMA) models were used to predict the population, industrial structure and transportation turnover, respectively. Monte-Carlo method was employed to evaluate the uncertainty of forecasted results. The results showed that the hybrid model combined with scenario analysis provided a relatively accurate forecast for the long-term energy demand in regions with limited statistical data, and the average standard error of probabilistic distribution in 2030 energy demand was as low as 0.15. The prediction results could provide supportive references to identify energy-saving potentials and energy development pathways.展开更多
时间序列数据广泛来源于社会各个领域,从气象学到金融学再到医学,准确的长期预测是时间序列数据分析、处理与研究中的一个关键问题。针对时间序列数据中存在的不同尺度相关性的挖掘与利用,提出一种基于神经网络的多尺度信息融合时间序...时间序列数据广泛来源于社会各个领域,从气象学到金融学再到医学,准确的长期预测是时间序列数据分析、处理与研究中的一个关键问题。针对时间序列数据中存在的不同尺度相关性的挖掘与利用,提出一种基于神经网络的多尺度信息融合时间序列长期预测模型ScaleNN,旨在更好地处理时间序列数据中的多尺度问题,从而实现更准确的长期预测。首先,结合全连接神经网络和卷积神经网络,有效提取全局信息与局部信息,并将2种信息聚合后进行预测;其次,通过在全局信息表征模块中引入压缩机制,以更轻量化的结构接受更长的序列输入,增大模型的感知范围并提高模型效能。大量实验结果表明,ScaleNN在多个真实世界数据集上的性能优于当前该领域的优秀模型PatchTST(Patch Time Series Transformer),在运行时间降低35%的同时仅需19%的参数量。可见,ScaleNN可广泛应用于不同领域的时间序列预测问题,为交通流量预测、天气预报等领域提供预测的基础。展开更多
基金Project(51606225) supported by the National Natural Science Foundation of ChinaProject(2016JJ2144) supported by Hunan Provincial Natural Science Foundation of ChinaProject(502221703) supported by Graduate Independent Explorative Innovation Foundation of Central South University,China
文摘An accurate long-term energy demand forecasting is essential for energy planning and policy making. However, due to the immature energy data collecting and statistical methods, the available data are usually limited in many regions. In this paper, on the basis of comprehensive literature review, we proposed a hybrid model based on the long-range alternative energy planning (LEAP) model to improve the accuracy of energy demand forecasting in these regions. By taking Hunan province, China as a typical case, the proposed hybrid model was applied to estimating the possible future energy demand and energy-saving potentials in different sectors. The structure of LEAP model was estimated by Sankey energy flow, and Leslie matrix and autoregressive integrated moving average (ARIMA) models were used to predict the population, industrial structure and transportation turnover, respectively. Monte-Carlo method was employed to evaluate the uncertainty of forecasted results. The results showed that the hybrid model combined with scenario analysis provided a relatively accurate forecast for the long-term energy demand in regions with limited statistical data, and the average standard error of probabilistic distribution in 2030 energy demand was as low as 0.15. The prediction results could provide supportive references to identify energy-saving potentials and energy development pathways.
文摘时间序列数据广泛来源于社会各个领域,从气象学到金融学再到医学,准确的长期预测是时间序列数据分析、处理与研究中的一个关键问题。针对时间序列数据中存在的不同尺度相关性的挖掘与利用,提出一种基于神经网络的多尺度信息融合时间序列长期预测模型ScaleNN,旨在更好地处理时间序列数据中的多尺度问题,从而实现更准确的长期预测。首先,结合全连接神经网络和卷积神经网络,有效提取全局信息与局部信息,并将2种信息聚合后进行预测;其次,通过在全局信息表征模块中引入压缩机制,以更轻量化的结构接受更长的序列输入,增大模型的感知范围并提高模型效能。大量实验结果表明,ScaleNN在多个真实世界数据集上的性能优于当前该领域的优秀模型PatchTST(Patch Time Series Transformer),在运行时间降低35%的同时仅需19%的参数量。可见,ScaleNN可广泛应用于不同领域的时间序列预测问题,为交通流量预测、天气预报等领域提供预测的基础。
