Impact of Correlation-based Feature Selection on Photovoltaic Power Prediction

被引：0

作者：

Kwon, Jung-Hyok ^{[1
]}

Lee, Sang-Woo ^{[2
]}

Lee, Sol-Bee ^{[2
]}

Kim, Eui-Jik ^{[2
]}

机构：

[1] Hallym Univ, Smart Comp Lab, Chunchon, South Korea

[2] Hallym Univ, Sch Software, Chunchon, South Korea

来源：

2019 4TH TECHNOLOGY INNOVATION MANAGEMENT AND ENGINEERING SCIENCE INTERNATIONAL CONFERENCE (TIMES-ICON) | 2019年

基金：

新加坡国家研究基金会;

关键词：

Correlation coefficient; feature selection; machine learning; photovoltaics power prediction; weather variables; ENSEMBLE;

D O I：

10.1109/times-icon47539.2019.9024493

中图分类号：

TP39 [计算机的应用];

学科分类号：

081203 ; 0835 ;

摘要：

This paper empirically presents the impact of the correlation-based feature selection on the accuracy of the photovoltaic (PV) power prediction, and then selects the weather variables that maximize prediction accuracy. To this end, the experiments are conducted using the weather dataset consisting of eighteen weather variables (i.e., features). For experiments, we first calculate a correlation coefficient of each weather variable by analyzing the correlation between PV power and each weather variable. Then, we create the subsets of weather variables considering the absolute value of correlation coefficient and generate the multiple prediction models using the created subsets. Finally, the accuracy of the generated prediction models is compared with each other to find the most accurate prediction model. The experiment results provide a reference guideline for selecting the weather variables that maximize the accuracy of PV power prediction.

引用

页数：4

共 11 条

[1] An analog ensemble for short-term probabilistic solar power forecast [J].