Welcome to the Solar Irradiance Classification project page!

The solar irradiance data classification is greatly helpful in identifying the perturbation sizes for perturb-and-observe (P&O) maximum power point tracking (MPPT) method. In this project, we observe the solar irrandiance variation in two location in the United States: Humboldt State University (HSU) and University of Nevada, Las Vegas (UNLV). The solar irradiance data is freely available at MIDC Homepage.

The solar irradiance patterns for the two locations are completely different:

The data can be easily categorized as coastal area patterns and desert area patterns. About 60% of the data collected in this project can be easily labeled manually; however, the rest can be hardly to tell due to the specific weather conditions.

We propose an automatic solar irradiance data classification system to automatically label the ambiguous data samples according to machine learning confidences. We compared various machine learning techniques and chose support vector machine (SVM) as the base classifier. The classification accuracy can achieve as high as 99.7% under 10-fold cross-validation. Readers who interest in the classification project can download and try out our demo code.

The source code and a demo of the solar irradiance data classification are available here.

 


Back to Yan Ke's Homepage.