摘要
针对光伏功率预测中极端天气样本不足和模型表达能力有限等问题,提出基于VMD-IKmeans-QLSTM的相似日预测方法。首先,设计多维特征体系的K-means聚类策略,确保极端天气样本数量充分。然后,基于Pearson相关系数构建多维加权特征矩阵实现精准相似日选取。采用变分模态分解(VMD)对功率信号多尺度分解,并设计量子比特变分量子电路的量子长短期记忆网络(QLSTM),利用量子叠加态增强非线性建模能力进行预测。该组合模型在新疆某光伏电站的应用结果表明,该模型与传统的模型相比,在多云、晴天和雨雪情况下,R2分别提升41.85%、8.06%和48.46%。
Aiming at the problems of insufficient extreme weather samples and limited model expression ability in PV power prediction,a similar-day prediction method based on VMD-IKmeans-QLSTM is proposed.Firstly,a K-means clustering strategy with a multi-dimensional feature system is designed to ensure a sufficient number of extreme weather samples.Then,a multi-dimensional weighted feature matrix is constructed based on the Pearson correlation coefficient to realize accurate similarday selection.The multi-scale decomposition of power signals is performed using Variational Mode Decomposition(VMD),and the Quantum Long and Short-Term Memory network(QLSTM)with quantum bit-variational quantum circuits is designed to utilize the quantum superposition state to enhance the nonlinear modeling capability for prediction.The application results of this combined model in a PV power station in Xinjiang show that the model improves R2 by 41.85%,8.06%and 48.46%in cloudy,sunny and rainy/snowy conditions,respectively,compared with the conventional model.
作者
秦汉森
郭欢
QIN Hansen;GUO Huan(School of Artificial Intelligence,Jianghan University,Wuhan 430056,China)
出处
《现代信息科技》
2025年第24期120-129,137,共11页
Modern Information Technology
基金
江汉大学2024年研究生科研创新基金项目(KYCXJJ202443)。
