Renewable energy power output forecasting

• Main Author: Wang, Zhiwei
• Other Authors: Xu Yan
• Format: Thesis-Master by Coursework
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Electrical and electronic engineering
• Online Access: https://hdl.handle.net/10356/171435
• Institution: Nanyang Technological University

This study focuses on utilizing machine learning algorithms to predict solar power generation. The validation dataset utilized in this study was gathered at the St. Lucia Campus, University of Queensland, Australia. Two algorithms, Gradient Descent and Long Short-Term Memory (LSTM), were employed to predict point values and generate prediction intervals using probabilistic analysis of training errors. The outcomes of the point forecasting and interval forecasting were evaluated. By comparing the results of Gradient Descent and LSTM, parameter tuning is instrumental in the performance of forecasting model. As a result, Gradient Descent (GD) exhibited enhanced understanding of the relationships between parameters and final outcomes, long short- term memory (LSTM) does not show a clear relationship due to the problem of overfitting. However, the prediction result of LSTM is better than GD according to the evaluation metrics and prediction graphics. However, with ample computational resources, it becomes feasible to ascertain the optimal parameters for the LSTM algorithm and address the challenge of overfitting. Given the extensive computational demands of the LSTM algorithm, the ideal configurations for hidden neurons and epochs remain undetermined. It will take more future works in this field.