Examining effects of air pollution on photovoltaic systems via interpretable random forest model

Renewable energy plays a vital role in power generation and solar photovoltaic systems due to resource availability throughout the year. This work aims to investigate the impact of air pollutants and meteorological parameters on the performance of the photovoltaic systems locally, taking into consideration the advantages of the photovoltaic power potential of the SW part of Romania, where Craiova is located (average solar radiation intensity >1350 kWh/m2/year). This study is based on a one-year dataset provided by a sensor that monitors particulate matter concentrations, volatile organic compounds, dioxide of carbon, ozone, noise, formaldehyde and three climate parameters (temperature, pressure, and relative humidity). The research methodology applies an innovative interpretable random forest model emphasising the implications of air pollution for photovoltaic systems. The proposed machine learning model was trained to predict the particulate matter level in air based on the basic environmental variable measurements. The study presents six random forest models of varying complexity, which reach the accuracy of classification for the selected problem up to 99 %, and applies the Shapley Additive Explanations technique to interpret the decision-making model. The observation regarding the highest concentration of particulate matter occurring during cold months, which typically do not align with peak solar irradiance, underscores the importance of considering various environmental factors in solar energy planning. With its practical implications, this insight offers decision-makers valuable information about the feasibility of optimising solar energy generation despite seasonal variations in air pollution levels, directly addressing their needs and concerns.