摘要
为提升超短期电网负荷预测精度,提出基于Stacking多模型融合的超短期电网负荷预测法。首先,结合5-折交叉验证法分别训练第一层的LSTM、LightGBM、XGBoost三个初级学习器,将训练结果进行Stacking融合;然后将融合结果作为新特征用于训练第二层LightGBM次级学习器,使用次级学习器得到电网负荷预测的最终结果;最后利用山东省公共数据开放平台提供的某市实际超短期电网数据验证所提方法的有效性。实验结果表明,比起单一模型预测,所提的Stacking多模型融合预测法,在预测结果的平均精度与峰谷变化的适应能力方面更具优势。
In order to improve the accuracy of ultra-short-term power grid load forecasting, an ultra-shortterm power grid load forecasting method based on Stacking multi-model fusion is proposed. Firstly, combine the 5-fold cross-validation method to train the three primary learners of LSTM, LightGBM and XGBoost in the first layer, and stack the training results;then use the fusion results as new features to train the second layer of LightGBM secondary learning;Finally, the actual ultra-short-term power grid data of a city provided by the public data open platform of Shandong Province is used to verify the effectiveness of the proposed method. The experimental results show that the proposed Stacking multi-model fusion prediction method has more advantages in the average accuracy of the prediction results and the adaptability to peak-to-valley changes than the single model prediction.
作者
张中健
高士亮
张露
安润鲁
张中城
周子力
ZHANG Zhongjian;GAO Shiliang;ZHANG Lu;AN Runlu;ZHANG Zhongcheng;ZHOU Zili(School of Cyberspace Security,Qufu Normal University,Jining Shandong 273165;School of Mathematics and Statistics,Northeast Petroleum University,Daqing Heilongjiang 163711)
出处
《软件》
2022年第8期131-134,178,共5页
Software
