Deep-learning-based probabilistic estimation of solar PV soiling loss

Abstract

Although the integration of solar photovoltaic (PV) systems is gaining widespread acceptance, the intermittency and instability of PV power generation lead to several operational challenges. PV power generation can be impacted by multiple environmental factors, such as the soiling of solar PV panels. There are some conventional methods proposed to deterministically estimate the solar power loss caused by soiling. However, the error of deterministic estimation cannot be eliminated due to the inherent volatility of solar power. Therefore, this paper proposes a probabilistic quantification method, namely SolarQRNN, to estimate the power loss by leveraging images captured by surveillance cameras. Specifically, the proposed model employs a novel quantile loss function and deep learning structures (backbone networks based on residual convolution units), which combines quantile regression and computer vision models for the first time. The proposed method has been extensively tested on a solar panel soiling image dataset. Test results indicate that SolarQRNN outperforms benchmark classification models with at least 51% improvements in evaluating metrics.