基于CEEMDAN和BiLSTM-AM的超短期风速预测方法

尹元亚; 潘文虎; 赵文广; 苏志朋; 韩屹; 吴红斌

尹元亚,潘文虎,赵文广,苏志朋,韩屹,吴红斌.基于CEEMDAN和BiLSTM-AM的超短期风速预测方法[J].电测与仪表,2024,61(9):77-84.
YIN Yuanya,PAN Wenhu,ZHAO Wenguang,SU Zhipeng,HAN Yi,WU Hongbin.Ultra-short-term wind speed prediction method based on CEEMDAN and BiLSTM-AM[J].Electrical Measurement & Instrumentation,2024,61(9):77-84.

基于CEEMDAN和BiLSTM-AM的超短期风速预测方法

Ultra-short-term wind speed prediction method based on CEEMDAN and BiLSTM-AM

DOI：10.19753/j.issn1001-1390.2024.09.010

中文关键词: 超短期风速预测自适应噪声完备集合经验模态分解双向长短时记忆注意力机制贝叶斯优化

英文关键词:ultra-short-term wind speed prediction, complete ensemble empirical mode decomposition with adaptive noise, bidirectional long short-term memory, attention mechanism, Bayesian optimization

基金项目:国网安徽省电力有限公司科技项目（B312C0230007）

作者中文名	作者英文名	单位
尹元亚	YIN Yuanya	国网安徽省电力有限公司芜湖供电公司
潘文虎	PAN Wenhu	国网安徽省电力有限公司芜湖供电公司
赵文广	ZHAO Wenguang	国网安徽省电力有限公司电力科学研究院
苏志朋	SU Zhipeng	国网安徽省电力有限公司芜湖供电公司
韩屹	HAN Yi	合肥工业大学电气与自动化工程学院
吴红斌	WU Hongbin	合肥工业大学电气与自动化工程学院

摘要点击次数: 439

中文摘要:

精准的风速预测结果可以推进风电的高效消纳以及增强新型电力系统的安全稳定性。为进一步挖掘风速序列的非线性特征，提升风速预测精度，提出了一种基于CEEMDAN与BiLSTM-AM的超短期风速预测方法。针对风速的随机波动性，采用自适应噪声完备集合经验模态分解(complete ensemble empirical mode decomposition with adaptive noise, CEEMDAN)对风速序列进行分解，转化为一系列较为平稳的子模态，从而降低预测的复杂度；采用具有双向信息流结构的双向长短时记忆网络(bidirectional long short-term memory,BiLSTM)挖掘各分量的变化规律，同时注意力机制(attention mechanism, AM)为神经网络的隐藏层状态分配相应权重，突出长时间序列中的关键信息，并利用贝叶斯优化对模型超参数进行寻优；将各分量的预测结果进行叠加作为最终结果。通过实际算例对比分析可知，该模型在单步与多步预测任务中均展现出良好的预测性能。

英文摘要:

Accurate wind speed prediction results can promote the efficient consumption of wind power and enhance the safety and stability of novel power system. To further explore the nonlinear characteristics of wind speed sequences and improve the accuracy of wind speed prediction, an ultra-short-term wind speed prediction method based on CEEMDAN and BiLSTM-AM is proposed. In response to the random fluctuation of wind speed, the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) is used to decompose the wind speed sequence into a series of relatively stable sub modes, thereby reducing the complexity of prediction. The bidirectional long short-term memory network (BiLSTM) with a bidirectional information flow structure is used to mine the change law of each component, and attention mechanism assigns corresponding weights to the hidden layer state of the neural network, highlights the key information in the long time series, and uses Bayesian optimization to optimize the model hyperparameter. The predicted results of each component are superimposed as the final result. Through comparative analysis of actual examples, the model exhibits good predictive performance in both single step and multi-step prediction tasks.

查看全文查看/发表评论下载PDF阅读器

关闭