Scholarlink

<p>The objective of this study is to improve short-term (half-hour-ahead) forecasting accuracy for photovoltaic (PV) power by developing a hybrid deep learning framework that integrates variational mode decomposition (VMD), wavelet packet decomposition (WPD), and long short-term memory (LSTM) networks, along with a reconstruction process. Accurate PV forecasting is crucial for efficient energy management and grid stability; however, the stochastic and nonlinear nature of solar generation presents significant challenges. The proposed framework decomposes the original PV power signal into intrinsic sub-components using VMD, refines high-frequency modes through WPD, and models each sub-signal with LSTM networks. A reconstruction stage then aggregates the predicted components into a unified forecast. Using real-world half-hourly data collected from a 20 MW PV plant in Algeria (2018–2019), the baseline LSTM model achieved a coefficient of determination (<i>R</i><sup>2</sup>) of 89.58%, a root-mean-square error (RMSE) of 355.83 kW, and a normalized RMSE (nRMSE) of 9.44%. In contrast, the proposed hybrid model achieved an <i>R</i><sup>2</sup> of 99.75%, RMSE of 54.14 kW, and nRMSE of 1.43%. These results confirm the effectiveness of the reconstruction-based approach in capturing complex temporal dynamics and reducing forecast errors, demonstrating its strong potential for application in intelligent energy management 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

Short-Term Photovoltaic Power Forecasting Using a Hybrid VMD-WPD-LSTM Model with Reconstruction

  • 期刊论文

  • 发表日期：

    2026年1月12日

查看全文
  • Khaled Ferkous,
  • Belgacem Bekkar,
  • Sarra Menakh,
  • Mawloud Guermoui,
  • Fatima Zohra Oulad Laid,
  • Soundous Sekayar

摘要

The objective of this study is to improve short-term (half-hour-ahead) forecasting accuracy for photovoltaic (PV) power by developing a hybrid deep learning framework that integrates variational mode decomposition (VMD), wavelet packet decomposition (WPD), and long short-term memory (LSTM) networks, along with a reconstruction process. Accurate PV forecasting is crucial for efficient energy management and grid stability; however, the stochastic and nonlinear nature of solar generation presents significant challenges. The proposed framework decomposes the original PV power signal into intrinsic sub-components using VMD, refines high-frequency modes through WPD, and models each sub-signal with LSTM networks. A reconstruction stage then aggregates the predicted components into a unified forecast. Using real-world half-hourly data collected from a 20 MW PV plant in Algeria (2018–2019), the baseline LSTM model achieved a coefficient of determination (R2) of 89.58%, a root-mean-square error (RMSE) of 355.83 kW, and a normalized RMSE (nRMSE) of 9.44%. In contrast, the proposed hybrid model achieved an R2 of 99.75%, RMSE of 54.14 kW, and nRMSE of 1.43%. These results confirm the effectiveness of the reconstruction-based approach in capturing complex temporal dynamics and reducing forecast errors, demonstrating its strong potential for application in intelligent energy management systems.