Parallel boosting neural network with mutual information for day-ahead solar irradiance forecasting

Area of Science:

• Renewable Energy Systems
• Artificial Intelligence in Energy
• Computational Science

Background:

• The global shift towards sustainable energy necessitates reliable renewable sources like solar power.
• Intermittency of solar energy poses challenges for photovoltaic (PV) system integration, requiring accurate solar irradiance (SI) forecasting.
• Existing deep learning (DL) and decision tree (DT) models have limitations in computational cost, data requirements, and generalizability.

Purpose of the Study:

• To develop a novel Parallel Boosting Neural Network (PBNN) framework to overcome limitations of traditional SI forecasting models.
• To enhance the accuracy and reliability of solar irradiance forecasting for integrated PV systems.
• To improve the efficiency and generalizability of forecasting algorithms.

Main Methods:

• A hybrid PBNN framework integrating boosting algorithms (Extreme Gradient Boosting, Categorical Boosting, Random Forest) with a feedforward neural network (FFNN).
• Parallel processing of base learner outputs (DT algorithms) with subsequent weighting by the FFNN for final prediction.
• Mutual Information (MI) algorithm employed for feature selection to identify crucial SI forecasting variables.

Main Results:

• The PBNN framework demonstrated significant improvements in forecasting accuracy, reducing Mean Absolute Percentage Error (MAPE) by [Formula: see text] and [Formula: see text] on Islamabad and San Diego datasets, respectively.
• Feature selection using MI algorithm enhanced the performance of the PBNN model.
• Robustness analysis through literature comparison confirmed the effectiveness of the proposed PBNN.

Conclusions:

• The proposed PBNN framework offers a superior approach to solar irradiance forecasting compared to existing methods.
• The PBNN effectively addresses the computational and generalizability issues associated with traditional DL and DT models.
• Accurate SI forecasting using PBNN is crucial for the reliable and efficient operation of solar energy systems.