This repository is an improved method of Deep-Solar-Eye
As the photovoltaic (PV) power has a very low carbon footprint, the use of solar panels is becoming widespread. However, the soiling of solar panels caused by severe weather will reduce up to 50% power generations. This challenge is considered by an existing method for quantifying the solar power loss. Whereas this method utilized a classification method, which is not sufficient for quantification resolution. To solve this, this project makes contribution on modifying the classification problem to a quantile regression problem based on the convolution neural network (CNN), which will increase the resolution of the quantification result.
This project is compiled on Visual Studio 2019
- If Visual Studio 2019 is available, please load the .sln file, then run SolarEye_main.py.
- If you don't have Visual Studio 2019, try any way you want to run SolarEye_main.py.
- Cuda is used, please check if cuda can be used on running GPU: Cuda support. If your GPU is unavailable, we reconmend you run the .ipynb file on colab.
A first-of-its-kind dataset, Solar Panel Soiling Image Dataset, comprising of 45,754 images of solar panels with power loss labels. From Deep-Solar-Eye Data has already been processed to binary data, please download from Binary dataset. Extract file, and put all of them to the same directory of .py files.
Pre-trained model, SolarQRNN.pth, is provided.