基于LASSO的光伏发电功率短期时间序列预测  

Short-term solar power forecasting based on LASSO

作　　者：刘金裕 赵磊 朱道立[1,2] LIU Jinyu;ZHAO Lei;ZHU Daoli(Sino-US Global Logistics Institute,Shanghai Jiao Tong University, Shanghai 200030, China;Antai College of Economics and Management,Shanghai Jiao Tong University, Shanghai 200030, China)

机构地区：[1]上海交通大学中美物流研究院,上海200030 [2]上海交通大学安泰经济与管理学院,上海200030

出　　处：《电力科学与工程》2019年第7期13-22,共10页Electric Power Science and Engineering

基　　金：国家自然科学基金资助项目(71871140);国家自然科学基金资助项目(71471112)

摘　　要：传统的短期直接预测方法在处理高维、复杂、多变的大数据方面具有一定的局限性,较难适用大数据环境下的工程应用。针对这一问题,本文提出了一种基于降维估计方法LASSO的光伏发电功率短期时间序列预测模型。该模型利用经典最小二乘模型进行参数估计,并结合选择算子,压缩参数空间以处理高维复杂数据。根据统计学习中常见的2种正则化表达和3类稀疏结构,本文设计了6种VAR-LASSO预测模型,并给出求解这6种模型的一阶多项式时间求解方法。最后仿真实验表明本文提出的VAR-LASSO模型在预测效果上优于经典VAR-最小二乘模型。The short-term solar power generation is the technical prerequisite for photovoltaic grid connection, and this technology is of great significance for improving the safety and stability of grid operation. The traditional short-term direct prediction method (VAR-least squares estimation model) has certain limitations in dealing with high-dimensional, complex and variable big data, and it is difficult to apply engineering applications in big data environment. Aiming at solving this problem, this paper proposes a short-term time series prediction model of solar power based on LASSO. The model uses the classical least squares model for parameter estimation and combines the selection operator (LASSO) to compress the parameter space to process high-dimensional complex data. According to the two regularized expressions (Tikhonov, Ivanov) and three types of sparse structures commonly used in statistical learning, six VAR-LASSO models are designed and the first-order polynomial time for solving these six models is given. Finally, this paper uses the data of 28 solar power generation stations in Delaware, USA. The experimental results show that the proposed VAR-LASSO model is superior to the classical VAR-least squares model in predicting the effect.Comparing with Tikhonov, Ivanov regularization expression method has more convenient advantages in parameter adjustment.

关 键 词：LASSO VAR-LASSO模型 Tikhonov Ivanov 短期预测 

分 类 号：TM73[电气工程—电力系统及自动化] N945.24[自然科学总论—系统科学]