Development of a Multi-Objective Cleaning Decision Model for Photovoltaic Rooftops Based on MARS–NSGA-II

Environmental ash accumulation significantly degrades PV rooftop performance. An efficient cleaning decision-making model is therefore essential to maintain operational efficiency. In order to solve this problem, this paper proposes a multi-objective cleaning decision model that combines Multivariate Adaptive Regression Spline (MARS) and Improved Non-Dominance Ordering Genetic Algorithm (NSGA-II). The MARS module is used to fit the nonlinear influence function of environmental factors such as temperature, humidity, wind speed and PM<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">10</inf> on photovoltaic power generation loss, and build a high-precision prediction model, which is further embedded in the multi-objective optimization framework to achieve a dynamic trade-off between saving cleaning water and improving power generation recovery through the improved NSGA-II algorithm. The core innovation is three rule-based adaptive cleaning modes—post-rain roller-brush, robotic water-spray, and winter dry-vacuum cleaning—with automatic selection based on weather forecasts and climate risk. In addition, in order to improve the search efficiency and the diversity of solutions, a custom crossover and mutation strategy is introduced into NSGA-II, which is especially robust for the environment of low rain and extremely low temperature. Experimental results show that compared with the benchmark method, the model can increase the water saving rate by 19.2% and the power generation recovery efficiency by 46%.