Please use this identifier to cite or link to this item: http://hdl.handle.net/123456789/1088
Title: Solar Power Generation Forecasting Using Ensemble Approach Based on Deep Learning and Statistical Methods.
Authors: Mariam AlKandari 
Supervisor: Prof. Imtiaz Ahmad
Degree Awarded: M.Sc. Degree in: Computer Engineering
Keywords: Solar Power Forecasting,Machine Learning,Statistical Methods,Renewable Energy,Photo voltaic
Issue Date: 2019
Publisher:  Kuwait university - college of graduate studies
Abstract: Solar power forecasting will have a significant impact on the future of large-scale renewable energy plants. Accurate solar forecasting will assist grid operators to obtain better management of the power grid and minimize the operational costs of generating electrical energy from burning fossil fuel, a natural energy source. Predicting solar photovoltaic power usage depends heavily on climate conditions, which fluctuate over time. To assist with this, machine learningis a viable solution to learn from historical weather data to predict the future power generation. Moreover, statistical methods proved to provide better accuracy of prediction in various aspects. In this research, we propose a hybrid model (MLSHM) that combines machine-learning methods with statistical methods for more accurate and precise prediction of future solar power generation from renewable energy plants. In addition, we developed a new machine-learning model, Auto-GRU, that predicts the solar PV power by learning from encoded historical weather data. To enhance and boost the accuracy of the proposed MLSHM, we employ two diversity techniques, i.e. structural diversity which combines two differently structured models (machine-learning models and a statistical model), and data diversity by dividing the training set among machine learning models. To combine the prediction of the ensemble members in the proposed MLSHM, we exploit four different combining methods: simple averaging approach, weighted averaging using linear approach, weighted averaging using non-linear approach, and combination through variance using inverse approach. The proposed hybrid model was validated on two real-time series datasets, Shagaya in Kuwait and Cocoa in the USA. The experiments show that the proposed hybrid model, using all the combination methods, achieved considerably higher accuracy compared to the prediction of the traditional individual models such as LSTM, GRU, Auto-LSTM and theta model. The results demonstrated that a hybrid model combining machine-learning methods with statistical methods outperformed a hybrid model that only combinesmachine-learningmodelswithoutstatisticalmethods.
URI: http://hdl.handle.net/123456789/1088
Appears in Programs:0612 Computer Engineering

Files in This Item:
File Description SizeFormat 
Thesis.pdf2,35 MBAdobe PDFView/Open    Request a copy
Show full item record

Page view(s)

33
Last Week
1
Last month
2
checked on Sep 26, 2020

Download(s)

1
checked on Sep 26, 2020

Google ScholarTM

Check


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.