Scholarlink

<p>Efficient solar energy utilization can alleviate energy shortages and global warming. However, the intrinsic intermittency and non-stationarity of photovoltaic (PV) power generation complicate accurate forecasting, which is essential for reliable grid operation and intelligent energy dispatch. This study presents an advanced forecasting framework that integrates refined datal decomposition, deep learning, and intelligent parameter optimization to improve prediction accuracy, robustness, and generalization. A cascaded decomposition strategy is proposed to mitigate non-stationarity and mode aliasing in PV power series. It combines wavelet-based subband decomposition with selective MVMD refinement on low-frequency components, enhancing temporal representation and improving forecasting accuracy. To further improve the stability and efficiency of model training, a meta-heuristic algorithm called Chaotic Elite Porcupine Optimization (CECPO) is developed. This algorithm incorporates chaotic mapping, Levy flight, and elite reverse learning to enhance population diversity and reduce the risk of premature convergence. The proposed method is thoroughly evaluated from three perspectives: model architecture, decomposition strategy, and optimization algorithm. Experiments on three real-world PV datasets demonstrate significant improvements over four state-of-the-art benchmarks, with WMAPE reduced by 4.08% to 25.81% and RMSE reduced by 5.90% to 18.41%. These results demonstrate the effectiveness of the proposed approach in capturing complex temporal structures in PV data and highlight its practical potential for improving forecasting reliability and enabling smarter energy management in modern power systems.</p>

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

A Cascaded DWT-MVMD-Based Decomposition Framework for Short-Term Photovoltaic Power Forecasting with Deep Learning and Chaotic Optimization

  • 期刊论文

  • 发表日期：

    2026年7月10日

查看全文
  • Hang Wan,
  • Siyuan Chen,
  • Jiasong Wang,
  • Rui Quan,
  • Yufang Chang,
  • William Derigent

摘要

Efficient solar energy utilization can alleviate energy shortages and global warming. However, the intrinsic intermittency and non-stationarity of photovoltaic (PV) power generation complicate accurate forecasting, which is essential for reliable grid operation and intelligent energy dispatch. This study presents an advanced forecasting framework that integrates refined datal decomposition, deep learning, and intelligent parameter optimization to improve prediction accuracy, robustness, and generalization. A cascaded decomposition strategy is proposed to mitigate non-stationarity and mode aliasing in PV power series. It combines wavelet-based subband decomposition with selective MVMD refinement on low-frequency components, enhancing temporal representation and improving forecasting accuracy. To further improve the stability and efficiency of model training, a meta-heuristic algorithm called Chaotic Elite Porcupine Optimization (CECPO) is developed. This algorithm incorporates chaotic mapping, Levy flight, and elite reverse learning to enhance population diversity and reduce the risk of premature convergence. The proposed method is thoroughly evaluated from three perspectives: model architecture, decomposition strategy, and optimization algorithm. Experiments on three real-world PV datasets demonstrate significant improvements over four state-of-the-art benchmarks, with WMAPE reduced by 4.08% to 25.81% and RMSE reduced by 5.90% to 18.41%. These results demonstrate the effectiveness of the proposed approach in capturing complex temporal structures in PV data and highlight its practical potential for improving forecasting reliability and enabling smarter energy management in modern power systems.