A Machine Learning Frontier for Predicting LCOE of Photovoltaic System Economics

Abstract

In this research article, the objective is to determine the return on investment (ROI) of photovoltaic (PV) power plants by employing machine learning (ML) techniques. Special focus is done on the levelized cost of electricity (LCOE) as a pivotal economic parameter crucial for facilitating economic decision‐making and enabling quantitative comparisons among different energy generation technologies. Traditional methods of calculating LCOE often rely on fixed singular input values, which may fall short in addressing uncertainties associated with assessing the financial feasibility of PV projects. In response, a dynamic model that integrates essential demographic, energy, and policy data, is introduced encompassing factors such as interest rates, inflation rates, and energy yield, which are anticipated to undergo changes over the lifetime of a PV system. This dynamic model provides a more accurate estimation of LCOE. The comparative analysis of ML algorithms indicates that the auto‐regression integration moving average (ARIMA) model exhibits a high accuracy of 93.8% in predicting consumer electricity prices. The validation of the model is highlighted through two case studies in the United States and the Philippines underscores the potential impact on LCOE values. For instance, in California, LCOE values could vary by nearly 30% (5.03 cents kWh−1 for singular values vs 7.09 cents kWh−1 using our ML model), influencing the perceived risk or economic feasibility of a PV power plant. Additionally, the ML model estimates the ROI for a grid‐connected PV plant in the Philippines at 5.37 years, in contrast to 4.23 years using traditional methods.

Keywords

• auto‐regression integration moving average (ARIMA) model
• levelized cost of electricity (LCOE)
• machine learning
• photovoltaics