短期负荷局部线性嵌入流形学习预测法

考虑短期日负荷预测各时刻点之间的整体性和相关性，提出一种从整体上刻画和预测短期日负荷的新方法。将日24点负荷数据值看作一个24维数据集，从多维角度挖掘负荷复杂的变化规律，建立高维预测模型。利用流形学习理论对建立的高维模型进行有效降维，从而提取高维空间数据的固有属性和整体几何规律，揭示其蕴含的有效信息。采用局部线性嵌入法（locallyl Inearembedding，LLE）对24维负荷数据进行非绳l生降维，在低维空间内进行负荷预测，再用LLE重构得到24个时刻的预测值。仿真结果表明本文提出方法相比于传统一维分量预测法精度更高、速度更快。...

Full description

Saved in:
Bibliographic Details
Published in电力系统保护与控制 Vol. 40; no. 7; pp. 25 - 30
Main Author 黄静 肖先勇 刘旭娜
Format Journal Article
LanguageChinese
Published 四川大学电气信息学院,四川成都,610065%四川大学电气信息学院,四川成都610065 2012
智能电网四川省重点实验室,四川成都610065
Subjects
局部线性嵌入
最小二乘支持向量机
流形学习
负荷预测
非线性降维
流形学习
最小二乘支持向量机
负荷预测
局部线性嵌入
非线性降维
Online AccessGet full text
ISSN1674-3415
DOI10.3969/j.issn.1674-3415.2012.07.005

Cover

More Information
Summary:考虑短期日负荷预测各时刻点之间的整体性和相关性，提出一种从整体上刻画和预测短期日负荷的新方法。将日24点负荷数据值看作一个24维数据集，从多维角度挖掘负荷复杂的变化规律，建立高维预测模型。利用流形学习理论对建立的高维模型进行有效降维，从而提取高维空间数据的固有属性和整体几何规律，揭示其蕴含的有效信息。采用局部线性嵌入法（locallyl Inearembedding，LLE）对24维负荷数据进行非绳l生降维，在低维空间内进行负荷预测，再用LLE重构得到24个时刻的预测值。仿真结果表明本文提出方法相比于传统一维分量预测法精度更高、速度更快。
Bibliography:load forecast; manifold learning; locally linear embedding; nonlinear dimension reduction; least square support vector machine
Considering the integrity and strong relevance of each point in short-term daily load value, the paper proposes a new model exposing and forecasting the daily load as a whole. We consider the 24 daily load values as a 24-dimensional data set, study the complex load changing rule from the multi-dimensional perspective and establish a high-dimensional load forecasting model. We use manifold theory to make an effective dimension reduction of the high-dimensional model, thus extract the inherence characteristics and overall regularity of the high-dimensional data, revealing the useful information it contains. We adopt locally linear embedding （LLE） to make a nonlinear dimension reduction for the 24-dimendional load data, and forecast the load in low dimensionality space, and then reconstruct the 24 load prediction based on the LLE reconstruction algorithm. Simulation results show that the
ISSN:1674-3415
DOI:10.3969/j.issn.1674-3415.2012.07.005