A photovoltaic power forecasting method based on the LSTM-XGBoost-EEDA-SO model

Area of Science:

• Renewable Energy Systems
• Artificial Intelligence in Power Engineering
• Signal Processing for Time Series Analysis

Background:

• Photovoltaic (PV) power generation is highly sensitive to meteorological conditions, necessitating accurate forecasting for stable power system operation and economic dispatch.
• Existing PV power forecasting methods often struggle with the complex, non-linear dynamics influenced by weather variability.
• Enhancing the precision of PV power forecasts is crucial for grid integration and reliable energy supply.

Purpose of the Study:

• To develop a hybrid framework for high-precision photovoltaic power forecasting.
• To improve the accuracy of PV power prediction by integrating advanced signal decomposition, parallel forecasting techniques, and adaptive weight optimization.
• To address the challenges posed by meteorological fluctuations in PV power generation.

Main Methods:

• A hybrid framework combining Thompson-Tau-Newton interpolation for data preprocessing and Pearson correlation for feature selection.
• Ensemble Empirical Mode Decomposition (EEMD) for decomposing PV power sequences into multi-scale low- and high-frequency components based on sample entropy.
• A parallel XGBoost-LSTM forecasting structure where XGBoost models low-frequency components and LSTM models high-frequency components.
• Snake Optimization (SO) algorithm for dynamic optimization of the combination weights for adaptive fusion of forecasting results.

Main Results:

• The proposed hybrid model significantly outperforms standalone benchmark methods in PV power forecasting accuracy.
• The SO algorithm achieved lower forecasting errors compared to Particle Swarm Optimization (PSO), Sparrow Search Algorithm (SSA), and equal-weight assignment.
• The decomposition and parallel forecasting approach effectively captured both trend and temporal dependencies in PV power data.
• The adaptive weight optimization ensured superior fusion of forecasting results from different components.

Conclusions:

• The proposed hybrid framework offers a novel and effective approach for high-precision PV power forecasting.
• Integrating multi-modal feature fusion and optimized weight allocation is key to improving forecasting performance.
• The method demonstrates superior capability in handling meteorological influences on PV power generation.
• This research contributes to more reliable and efficient integration of solar energy into power grids.