Dung beetle optimization algorithm-based hybrid deep learning model for ultra-short-term PV power prediction.

Area of Science:

• Renewable Energy Systems
• Artificial Intelligence in Energy
• Machine Learning for Power Grids

Background:

• Accurate ultra-short-term photovoltaic (PV) power prediction is crucial for grid stability and efficient solar energy integration.
• Existing prediction models often struggle with the complex temporal dynamics and correlations inherent in PV power generation.
• The need for advanced forecasting techniques is driven by the increasing penetration of variable renewable energy sources.

Purpose of the Study:

• To develop and evaluate a novel hybrid model for improving ultra-short-term PV power prediction accuracy.
• To leverage the strengths of self-attention temporal convolutional networks (SATCN) and bidirectional long short-term memory (BiLSTM) networks for feature extraction.
• To optimize model performance using a metaheuristic algorithm for enhanced forecasting.

Main Methods:

• A hybrid model integrating SATCN for temporal and correlation feature extraction with BiLSTM networks was proposed.
• The self-attention mechanism was employed within SATCN to capture intricate temporal dependencies.
• Hyperparameter optimization was performed using the dung beetle optimization algorithm on a year-long PV power dataset.

Main Results:

• The proposed hybrid model demonstrated superior performance compared to standalone CNN, BiLSTM, and TCN models.
• A significant reduction in root-mean-square error (RMSE) by 33.1% was achieved relative to other models.
• The model attained a mean absolute error (MAE) of 0.175, a weighted mean absolute percentage error (wMAPE) of 4.821, and an R-squared (R2) value of 0.997.

Conclusions:

• The hybrid SATCN-BiLSTM model offers a highly accurate solution for ultra-short-term PV power forecasting.
• The integration of self-attention mechanisms and BiLSTM effectively captures complex PV power patterns.
• The optimized model shows significant potential for practical applications in solar energy development and grid management.