An adaptive ensemble approach for solar irradiance tilt adjustment / Serina Lim Chee Lian

Submitted in fulfilment of the requirements for Degree of Science (by Research)

ABSTRACT This research enhances solar photovoltaic (PV) system efficiency and reliability by integrating advanced machine learning models and adaptive solar tracking mechanisms. The study employs Long Short-Term Memory (LSTM) networks, Extreme Learning Machines (ELM), and an ensemble approach for the precise forecasting of solar irradiance. An integration model combining the ensemble LSTM-ELM with Liu-Jordan and Badescu models was implemented to optimize solar panel tilt angles using Particle Swarm Optimization (PSO), leading to a closer alignment with actual solar data and enhancing energy capture. The proposed system utilizes a motor that adjusts the solar panel's tilt angle based on the monthly tilt angle derived from the integration model. This adjustment is performed once at the beginning of each month, after which the panels remain in the adjusted position to maximize energy capture for the rest of the month. Comprehensive testing across various environmental scenarios confirmed the system's robustness and reliability. The ensemble models demonstrated superior accuracy following thorough testing and optimization, achieving an MAE of 59.5, RMSE of 133.1, and an R2 value of 0.85, surpassing individual models. The model's predictions were validated against NASA data, revealing a significant improvement in accuracy over time. The precision tilt adjustment mechanism resulted in a 16.3% increase in energy output compared to a fixed system. These findings underscore the effectiveness of combining machine learning with optimization techniques to maximize solar power generation efficiency.

Dissertation (Degree of Master of Science by Research) - Universiti Teknologi Brunei (2024)

Includes references pages 122-131 and appendix pages 132-214