电网技术Issue(2)：307-311,5.DOI:10.13335/j.1000-3673.pst.2015.02.002
基于机器学习的自适应光伏超短期出力预测模型
Machine Learning-Based Adaptive Very-Short-Term Forecast Model for Photovoltaic Power
摘要
Abstract
For the reason that forecast performance of solar radiation and cloud cover information is still very poor in China, the accuracy improvement of direct prediction for photovoltaic (PV) power by introducing meteorological data is limited. To further improve the prediction accuracy, a self-adapting very-short-term forecast model for PV power is proposed based on the feature mining of historical PV output data. Firstly, the parameters of support vector machine (SVM) classifier are calculated using wavelet analysis and characteristics of the historical PV output data. Then, based on the established SVM classifier, the power curve type of the sequential 15 minutes is determined through historical PV output data of the previous 30 minutes. An appropriate forecast method, which is selected between auto-regressive and moving average model (ARMA) and artificial neural network model (ANN) according to the type of power curve, is finally obtained for power prediction. Real data based experiments are performed to compare the performance of ARMA, ANN and the proposed self-adapting forecast model. The experimental results show that adaptive forecast has the best performance in terms of root mean square error (RMSE), the mean absolute percentage error (MAPE) and Theil inequality coefficient(TIC).关键词
自适应预测/自回归和滑动平均模型/神经网络/小波分析/超短期光伏出力预测Key words
self-adapting forecast/auto-regressive and moving average model/artificial neural network model/wavelet analysis/very-short-term forecast for photovoltaic power分类
信息技术与安全科学引用本文复制引用
高阳,张碧玲,毛京丽,刘勇..基于机器学习的自适应光伏超短期出力预测模型[J].电网技术,2015,(2):307-311,5.基金项目
北京邮电大学青年科研创新计划专项(2013RC100)。 （2013RC100）
