Performance Evaluation of Probabilistic Methods Based on Bootstrap and Quantile Regression to Quantify PV Power Point...

Area of Science:

• Renewable Energy Systems
• Artificial Intelligence in Energy
• Statistical Modeling

Background:

• Accurate solar photovoltaic (PV) power forecasting is crucial for grid integration.
• Quantifying forecast uncertainty is essential for reliable energy management.
• Existing methods often lack robust uncertainty estimation for PV power.

Purpose of the Study:

• To present and compare two probabilistic approaches for estimating uncertainty in solar PV power point forecasts.
• To evaluate the effectiveness of a hybrid intelligent model for PV power forecasting.
• To assess the reliability of uncertainty quantification methods for PV power predictions.

Main Methods:

• A hybrid intelligent model combining Wavelet Transform (WT) for data filtering and a Radial Basis Function Neural Network (RBFNN) optimized by Particle Swarm Optimization (PSO) was developed for point forecasts.
• Two probabilistic methods, bootstrap and quantile regression (QR), were employed to estimate forecast uncertainty and generate prediction intervals (PIs).
• The performance of the hybrid model and the uncertainty quantification methods were evaluated using real-world PV power data.

Main Results:

• The hybrid WT+RBFNN+PSO model demonstrated strong point forecast capabilities compared to other models.
• Both bootstrap and QR methods effectively quantified the uncertainty associated with PV power forecasts.
• Numerical tests confirmed the reliability of the proposed uncertainty quantification approaches.

Conclusions:

• The proposed hybrid intelligent model provides accurate point forecasts suitable for uncertainty quantification.
• The bootstrap and QR methods offer reliable ways to estimate the uncertainty in solar PV power predictions.
• Accurate uncertainty estimation is vital for the effective integration and management of solar PV energy.