Scholarlink

To enhance the accuracy and reliability of short-term photovoltaic (PV) power forecasting and to mitigate the challenges posed by the inherent volatility and nonlinearity of PV data, this paper proposes a novel hybrid prediction framework: CEEMDAN-PE-VMD-IHHO-GRU. First, the historical PV power series are decomposed into multiple intrinsic mode functions (IMFs) using Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN). The complexity of each IMF is quantitatively evaluated via Permutation Entropy (PE), and the IMFs are subsequently classified and reconstructed based on their entropy values. High-frequency components are further refined through Variational Mode Decomposition (VMD) to capture more detailed features. Finally, the hyperparameters of the Gated Recurrent Unit (GRU) model are optimized using an Improved Harris Hawks Optimization (IHHO) algorithm, which incorporates good point set initialization, adaptive energy decay, and Gaussian random walk strategies. Experimental results demonstrate that the proposed model effectively captures nonlinear temporal patterns in PV power series and significantly improves forecasting performance compared to conventional and existing hybrid models.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Short-Term PV Power Prediction Method Based on Two-Stage Decomposition and IHHO-GRU

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Qiquan Wu,
  • Guixue Cheng,
  • Hanyun Wang

摘要

To enhance the accuracy and reliability of short-term photovoltaic (PV) power forecasting and to mitigate the challenges posed by the inherent volatility and nonlinearity of PV data, this paper proposes a novel hybrid prediction framework: CEEMDAN-PE-VMD-IHHO-GRU. First, the historical PV power series are decomposed into multiple intrinsic mode functions (IMFs) using Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN). The complexity of each IMF is quantitatively evaluated via Permutation Entropy (PE), and the IMFs are subsequently classified and reconstructed based on their entropy values. High-frequency components are further refined through Variational Mode Decomposition (VMD) to capture more detailed features. Finally, the hyperparameters of the Gated Recurrent Unit (GRU) model are optimized using an Improved Harris Hawks Optimization (IHHO) algorithm, which incorporates good point set initialization, adaptive energy decay, and Gaussian random walk strategies. Experimental results demonstrate that the proposed model effectively captures nonlinear temporal patterns in PV power series and significantly improves forecasting performance compared to conventional and existing hybrid models.