Loading...
A Machine Learning Model Based Schedule of Photovoltaic Solar Plant Dust Cleaning
Abuzaid, Haneen Mohammad Faleh
Abuzaid, Haneen Mohammad Faleh
Date
2024-03
Authors
Advisor
Type
Dissertation
Degree
Citations
Altmetric:
Description
A Doctor of Philosophy Dissertation in Engineering Systems Management by Haneen Mohammad Faleh Abuzaid entitled, “A Machine Learning Model Based Schedule of Photovoltaic Solar Plant Dust Cleaning”, submitted in March 2024. Dissertation advisor is Dr. Mahmoud Awad and dissertation co-advisor is Dr. Abdulrahim Shamayleh. Soft copy is available (Dissertation, Completion Certificate, Approval Signatures, and AUS Archives Consent Form).
Abstract
Photovoltaic (PV) systems are widely utilized renewable energy resources, playing a vital role in sustainable energy generation worldwide. Nonetheless, their performance is significantly reduced by dust accumulation, emphasizing the importance of effective cleaning strategies. This research aims to enhance the overall performance of PV systems by improving maintenance practices, particularly focusing on the methods and schedules for PV cleaning. To achieve this, a comprehensive literature review spanning from 2010 to 2024 is conducted, concerning factors influencing PV performance, and current PV cleaning methods, with ongoing discussion and validation from PV experts, actively engaged in the PV field, to bridge theory and practice. The most influential factors include meteorological variables, PV specifications, system design, dust characteristics, sustainability considerations, and operational factors. Using a Multi-Criteria Decision-Making (MCDM) model, specifically the Analytic Network Process (ANP), the study proposes an optimal PV cleaning method, identifying partially automated cleaning as the most suitable for utility-scale PV projects in MENA. Additionally, the research offers a robust cleaning schedule by predicting the PV Performance Ratio (PR), a standardized metric that is widely used in performance-guaranteed contracts, utilizing machine learning algorithms. Two prediction models are proposed: time-series prediction models using LSTM, ARIMA, and SARIMAX algorithms for predicting the PR, and a threshold-based ensemble voting classifier using RF, Log, and GBM for predicting the cleaning process, with three large and distinct datasets validation. Results indicate the efficacy of SARIMAX in PR prediction, with high R2 values and low errors across case studies. The ensemble voting classification model achieves satisfactory performance accuracy in predicting cleaning processes. Notably, PR predictive models outperform cleaning process classification models. Moreover, varying features’ importance outcomes across case studies highlight the necessity of including location-specific conditions for optimal PV cleaning strategy. The findings provide robust guidelines for PV system stakeholders, aiding informed decision-making and enhancing the sustainability of PV cleaning processes.