Hour-Ahead Photovoltaic Power Prediction Combining BiLSTM and Bayesian Optimization Algorithm, with Bootstrap...

Area of Science:

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

Background:

• Accurate photovoltaic (PV) power prediction is essential for integrating renewable energy into the grid.
• Existing forecasting models face challenges with accuracy and robustness under varying conditions.
• Deep learning offers potential for enhanced PV power forecasting.

Purpose of the Study:

• To develop and evaluate a novel deep learning model for precise one-hour ahead PV power forecasting.
• To optimize the model's performance through hyperparameter tuning and robust data preprocessing.
• To compare the proposed model's efficacy against established forecasting techniques.

Main Methods:

• Implementation of a Bidirectional Long Short-Term Memory (BiLSTM) deep learning model.
• Application of a robust data preprocessing algorithm for historical PV data.
• Synergistic combination of BiLSTM with Bayesian Optimization Algorithm (BOA) for hyperparameter tuning.
• Utilization of bootstrap resampling for interval prediction.

Main Results:

• The proposed BiLSTM-BOA model demonstrated superior deterministic forecasting performance.
• Achieved significant reductions in normalized mean absolute error (nMAE): 75.03% vs. MLP and 77.01% vs. RF.
• Interval prediction provided robust and flexible forecasts adjusted to desired confidence levels.

Conclusions:

• The BiLSTM-BOA model offers a highly effective solution for short-term PV power forecasting.
• The model's performance is validated across diverse meteorological and seasonal conditions.
• This approach enhances the reliability of PV power integration into the energy grid.