Short-term microgrid load probability density forecasting method based on k-means-deep learning quantile regression

• DOI: 10.1016/j.egyr.2022.03.117
• Publication Date: 2022-03-25T09:50:57Z
• Authors: Zilong Zhao; Jinrui Tang; Jianchao Liu; Ganheng Ge; Binyu Xiong; Yang Li
• Keywords: Microgrid; Quantile; Extreme Learning Machine; Kernel density estimation; Quantile regression

Traditional short-term load forecasting (STLF) methods for large utility grid systems usually provide the forecasted with deterministic points. However, cannot reveal pattern and uncertainty of controllable in a microgrid, where prediction errors may exceed expected range due to high volatility strong randomness. In order deal this matter, probability density method is proposed predict microgrid robust power scheduling paper. The effectively combines several data-driven statistical algorithms, including k-means algorithm, quantile regression long memory neural network (QRLSTM), kernel estimation (KDE). Firstly, similar days related day are selected through historical data these divided into two subsets training dataset testing dataset. Secondly, QRLSTM-based model established used different quantiles. Finally, function predicted points obtained by KDE on target day. accuracy evaluated roundly results demonstrate that can reproduce distribution noticeably better performance than some benchmark methods.