An enhanced approach to cost-effective cleaning optimisation for enhanced solar PV power production in arid areas

Dust accumulation in arid environments results in energy losses that affect solar module performance and increase operational expenses. Real-time operational and environmental data were used to evaluate soiling losses ranging from 5% to 20% at the 300 MW Sakaka Solar Power Plant in Saudi Arabia between 2020 and 2024. In this paper, the soiling ratio (SR) is first calculated from operational and environmental data to quantify the impact of dust on the performance of solar photovoltaic systems. Subsequently, the grey wolf optimiser was employed to develop a method for determining the most effective cleaning schedule that balances soiling-related energy income losses and cleaning costs. In addition, a prediction-based approach, i.e., long short-term memory, was developed to forecast soiling behaviour by estimating the SR, resulting in cleaning strategies that were both more proactive and more successful. This incorporates the stochastic and temporally variable process of soiling to maintain flexible cleaning schedules while minimising operational costs and energy losses. Further, the model outperformed traditional intelligent models, with a root mean squared error of 0.0144, a mean absolute error of 0.0110, and a coefficient of determination (R²) of 0.79. The optimisation results show that mechanised cleaning consistently outperforms manual methods across all cases (dry, wet, and yearly). In calculation-driven optimisation, the optimal intervals for mechanised cleaning were 15, 30, and 18 d, respectively, and for manual cleaning, 37, 55, and 45 d; prediction-based optimisation refined these to 16, 46, and 17 d, and 38, 60, and 37 d, respectively. These results show that the prediction framework not only maximises cost efficiency by adjusting the cleaning pattern based on environmental conditions but also minimises unnecessary cleaning of solar panels, thereby reducing excess soiling losses.