Short-term photovoltaic forecasting model with parallel multi-channel optimization based on improved dung beetle...

Area of Science:

• Renewable Energy Systems
• Artificial Intelligence in Power Engineering

Background:

• Accurate photovoltaic (PV) power generation prediction is crucial for power grid stability and efficient dispatching.
• Existing prediction models face challenges in achieving high accuracy under diverse weather conditions.

Purpose of the Study:

• To enhance the accuracy of PV generation prediction.
• To improve the reliability of power grid operations through better forecasting.

Main Methods:

• An improved Rhino beetle optimization algorithm (LSDBO) incorporating Logistic chaos mapping and sine function strategies was developed.
• A parallel Convolutional Neural Network-Multi-Head Attention (CNN-MHA) and Long Short-Term Memory-Multi-Head Attention (LSTM-MHA) model, termed PCL-MHA, was optimized.
• Experiments were conducted using historical data from the Alice Springs PV system, featuring monocrystalline and polycrystalline silicon panels.

Main Results:

• The proposed PCL-MHA model achieved superior performance across 16 metrics under various weather conditions.
• Average performance metrics included R2 of 98.43%, Mean Squared Error (MSE) of 2.69%, Mean Absolute Error (MAE) of 7.8%, and Root Mean Squared Error (RMSE) of 15.09%.
• The model demonstrated an average improvement of 2.41% in R2, 6.6% in MSE, 7.77% in MAE, and 11.21% in RMSE compared to other models.

Conclusions:

• The enhanced LSDBO algorithm and PCL-MHA model significantly improve PV generation prediction accuracy.
• The findings contribute to more stable and efficient power grid management through reliable renewable energy forecasting.