A hybrid framework for estimating photovoltaic dust content based on UAV hyperspectral images

Abstract

Dust is one of the key factors influencing photovoltaic (PV) power generation. The ability to accurately capture PV dust information is essential for PV operation and sustainable utilization. This study employs uncrewed aerial vehicle (UAV) hyperspectral imaging to monitor PV dust deposition. To address the problems of information redundancy in hyperspectral data and the backpropagation neural network (BPNN) easily falling into local optimum, a high-precision UAV hyperspectral PV dust estimation method is proposed. The fractional order derivative (FOD) is applied to the spectral reflectance of PV dust accumulation, and a PV dust estimation model with sine map tuna swarm optimized backpropagation neural network (STSO-BPNN) is established, which is validated using UAV hyperspectral images and ground measured dust data. The results show that FOD improves the spectral signal-to-noise ratio, and the 0.2 order STSO-BPNN model achieves higher accuracy (R² = 0.95, RMSE = 0.79 g/m², RPIQ = 7.98). These findings provide a scientific basis for the rapid and accurate estimation and mapping of PV dust accumulation while proposing a novel strategy for efficient PV implementation and management.