Optimized solar power forecasting: A multi-decomposition framework using VMD and swarm techniques

Accurate short-term forecasting of photovoltaic (PV) power generation is crucial for optimizing solar plant operations and ensuring grid stability. This study proposes an advanced VMD-SD-LSTM forecasting model with reconstruction, integrating Variational Mode Decomposition (VMD) and Swarm Decomposition (SD) to enhance predictive accuracy. High-frequency components extracted by VMD undergo SD for further refinement before being processed by independent Long Short-Term Memory (LSTM) networks, while low-frequency components are directly fed into LSTM models. The proposed method was evaluated against LSTM, VMD-LSTM, and SD-LSTM models using R2, RMSE, and nRMSE metrics. Results demonstrate that VMD-SD-LSTM with reconstruction outperforms all baseline models, achieving the highest R2 of 99.842% (winter), 99.360% (spring), 99.619% (summer), and 99.711% (autumn), while significantly reducing RMSE. The proposed framework effectively captures both short-term fluctuations and long-term trends, proving its robustness for real-world PV power forecasting.