基于风速-功率组合分解重构的海上风电功率预测  

OFFSHORE WIND POWER FORECASTING BASED ON WIND SPEED-POWER COMBINATION DECOMPOSITON

作　　者：傅质馨[1] 王宝池 刘皓明[1] 王健[1] 朱俊澎 袁越[1] Fu Zhixin;Wang Baochi;Liu Haoming;Wang Jian;Zhu Junpeng;Yuan Yue(School of Electrical and Power Engineering,Hohai University,Nanjing 211100,China)

机构地区：[1]河海大学电气与动力工程学院,南京211100

出　　处：《太阳能学报》2024年第11期418-426,共9页Acta Energiae Solaris Sinica

基　　金：国家自然科学基金青年项目(52207091)。

摘　　要：为提高海上风电功率预测精度,提出一种基于数据组合分解重构和贝叶斯优化的深度学习预测模型。首先,利用改进的自适应噪声完全集合经验模态分解和麻雀搜索算法优化的变模态分解将海上风速数据和历史功率数据进行分解,降低信号波动性和单一分解不彻底对预测结果的干扰。然后,根据分量的模糊熵计算结果进行分组重构,简化模型。最后,对每个功率分量建立基于贝叶斯算法和长短期记忆神经网络的预测模型,将各分量的结果叠加得到海上风电功率预测值。经海上风电实测数据验证表明,与单一的海上风速或功率数据分解预测模型相比,所提出模型可降低原始数据的强波动性对预测结果的干扰,在单步预测和多步预测上都有更高的精度。To enhance prediction accuracy,this study proposes a novel deep learning model that leverages data combination decomposition and Bayesian optimization.Initially,utilizing improved complete ensemble empirical mode decomposition with adaptive noise and variable modal decomposition optimized by the sparrow search algorithm,offshore wind speed data and historical power data are subjected to composite decomposition.Subsequently,the modal components obtained from composite decomposition are reconstructed using fuzzy entropy analysis to simplify the model.Finally,Bayesian-optimized long short-term memory neural networks are employed to forecast each power component,and the results of these components are aggregated to obtain the prediction of offshore wind power generation.Experimental data from offshore wind farms indicate that the proposed method efficiently mitigates the interference resulting from significant fluctuations in the original data,resulting in higher accuracy for both single-step and multi-step predictions compared to a conventional single-model approach utilizing offshore wind speed or power data decomposition.

关 键 词：海上风电场 预测 模态分解 长短期记忆 模糊熵 贝叶斯优化 

分 类 号：TM614[电气工程—电力系统及自动化]