Gaussian process for learning solar panel maximum power point characteristics as functions of environmental conditions

This paper proposes a method to learn the variation of solar panel Maximum Power Point (MPP) parameters as functions of environmental conditions using Gaussian Process (GP) based machine learning. As a result of using GP, functions are learned along with the additional information of their uncertainty margins. The paper discusses about learning three functions specifically, where each of them take the two environmental variables ´solar irradiance´ and ´cell temperature´ as arguments and map these environmental variables to the corresponding MPP parameters, namely, the maximum power, the MPP voltage and the MPP current. Learned functions presented in the paper have been trained for a commercially available solar panel using MPP data generated using a previously published solar panel simulator. The learned function for maximum power has been validated by comparing the function outputs (GP results) against the manufacturer specified power values. A discussion about how learning such functions can help in advancing MPP Tracking (MPPT) is also provoked while highlighting the impact machine learning can make to the field of photovoltaics.