A Convolutional Transformer-based Model for Estimation of Soiling-driven PV Power Loss

Using a well-built estimation model, one effective way of diagnosing PV systems is to compare the estimated PV power output and the true PV power output. Considering that PV power output is affected by various external factors, we first need to build a reliable model for PV power output estimation. In this paper, we propose a model using a convolutional transformer for the estimation of PV power loss by soiling effects. The input to the model is an RGB image and the output is the estimated power loss class. Using the five power loss classes, through experiments, we show that our model achieves 83.37% classification accuracy.