Robust analysis of photovoltaic plants: A framework based on prediction uncertainties by machine learning

Abstract

Uncertainty in photovoltaic plants significantly affects performance and economic viability, making its management crucial for optimal design and operation. In this context, the present study offers a novel framework for comprehensively assessing the photovoltaic project development. Iran was selected as a case study, and using the k-means algorithm, the area was divided into eight clusters. Subsequently, a ten Megawatts photovoltaic plant was assumed in each region, and a comprehensive feasibility assessment was conducted, encompassing energy, economic, environmental, exergy, exergoeconomic, enviroeconomic, and energoeconomic aspects. Finally, the uncertainties were simulated using the Monte Carlo method. The gradient boosting regressor machine learning algorithm was predicted the PV plant for 10,000 different scenarios (R² ∼ 98 %). The results showed the capacity factor and exergy efficiency of photovoltaic panels across Iran ranged from 23-27 % and 19.3–20.6 %, respectively. Additionally, the Levelized cost of electricity and payback period were in the range of 48–56 $/MWh and 13–16 years, respectively. Also, Monte Carlo analysis showed the probability of project failure for developing photovoltaic power plants in Iran was between 14–26 %. This assessment demonstrates the PV project development in Iranian cities is feasible and can serve as a reference for fossil-fuel-rich countries.