基于季节性STM模型的中国电力消费特征及预测研究

发布时间：2018-03-17 23:36

  本文选题：季节性结构时间序列模型　切入点：电力消费特征　出处：《华侨大学》2017年硕士论文　论文类型：学位论文

【摘要】：在我国,电力消费是能源消费的主要方式。一方面,电力消费通常具有周期性和趋势性两种特征,如何合理并准确刻画电力消费特征,一直以来是经济核算部门和一些科研机构的重要研究课题之一。另一方面,在“十三五”电力规划期间,国内电力消费情况面临许多严峻挑战。当务之急是对未来的电力需求进行预测,如何准确地预测我国电力消费需求,实现经济平稳增长,同样也是一个重要研究课题之一。在这种大背景下,本文从一个新视角对我国电力消费特征就测度和预测两个方面进行讨论。本文从成分分解和状态空间模型的角度,详细介绍四种静态STM的基本构造方法理论,并应用于我国电力消费需求特征场景,对2005年至2016年我国电力季度消费需求进行特征测度,评选出最优的静态STM。之后在其基础上进行推广,构造固定参数形式的STM+βX模型和时变参数形式的tSTM+β_1X,应用于我国电力消费需求预测场景,对2016年电力消费做出预测。最后对比STM+X和静态STM,基于最优模型选取指标,从拟合效果和预测精度两个维度比较这几种STM的模型效果。本文的主要结论主要有:第一,通过构造Ⅰ(1)和Ⅰ(2)过程两种趋势成分形式,季节虚拟变量和季节三角函数两种季节成分形式,通过两两组合构造出四种状态空间形式的静态STM模型,利用极大似然法对超参数进行估计,刻画出四种我国电力消费需求特征。这四种静态STM较好地刻画了我国电力消费特征,表明我国电力消费具有趋势性和季节性两种特征。根据四种误差评估指标结果,发现Ⅰ(2)过程趋势成分加时变季节三角函数的季节成分组合的静态STM模型,能够非常好地刻画我国电力消费的季节性特征。第二,在特征测度方面得到最优的静态STM模型的基础上,对模型进行推广,得到固定参数形式的STM+βX和时变参数形式的tSTM+β_1X两种状态模型框架下的季节性STM+X模型,引入经济增长和气候因素两种外生变量,先利用极大似然法对超参数进行估计,再利用这两种STM+X模型刻画电力消费特征。这两种STM+X也同时能很好的刻画出我国电力消费需求特征,根据四种误差评估指标结果,发现时变参数形式的tSTM+β_1X在拟合和预测两方面都具有显著的优势,即电力消费在预测方面的最优模型为Ⅰ(2)过程趋势成分加时变季节三角函数组合的时变参数tSTM+β_1X模型。第三,实证研究结果表明,我国电力消费需求具有稳定的内在增长趋势,同时容易受到外部冲击的影响。此外,我国电力消费需求还具有明显的季节效应,受全球气候变暖的影响,这种季节效应还存在显著的季节扩张性。第四,实证研究结果表明,我国电力消费还受我国宏观经济内生性发展和我国气候条件程度决定。同时,经济增长和气候条件对我国电力消费需求存在一定的冲击效应,且这种冲击效应是时变的。
[Abstract]:In China, electricity consumption is the main way of energy consumption. On the one hand, electricity consumption usually has two characteristics: periodicity and trend. It has always been one of the important research topics for economic accounting departments and some scientific research institutions. On the other hand, during the 13th Five-Year Plan, The domestic electric power consumption is facing many severe challenges. The urgent task is to forecast the future power demand, how to accurately forecast our country's electricity consumption demand, and how to realize the steady economic growth. It is also one of the important research topics. Under this background, this paper discusses the measurement and prediction of the characteristics of power consumption in China from a new perspective. This paper introduces in detail the basic construction method theory of four static STM, and applies them to the feature scenario of electric power consumption demand in our country, and measures the characteristics of our country's electric power quarterly consumption demand from 2005 to 2016. The optimal static STM is selected. Then the STM 尾 X model with fixed parameters and the tSTM 尾 1X with time-varying parameters are constructed on the basis of this model, which is applied to the power consumption demand prediction in China. Finally, compared with STM X and static STM, the model effects of these STM are compared from two dimensions of fitting effect and prediction precision based on the optimal model selection index. The main conclusions of this paper are as follows: first, By constructing two kinds of trend component forms, seasonal virtual variable and seasonal trigonometric function, the static STM model of four state space forms is constructed by constructing two kinds of trend component forms in the process of 鈪，

本文编号：1627026