Sim-to-Real Transfer with Domain Randomization for Maximum Power Point Estimation of Photovoltaic Systems

Simulations are widely used in the field of photovoltaic systems as they provide an abundant source of data for the building and training of numerical methods or artificial intelligence techniques. However, the strategies that succeed in simulation may not be victoriously transferred to the real world due to the modeling errors. In this paper, we propose a Gaussian process regression with domain randomization, which is able to bridge the ‘Sim-to-Real’ gap in the application of maximum power point estimation. By randomizing the parameters of the models for the training process, the Gaussian process regression models can minimize the ‘Sim-to-Real’ transfer cost and adapt the dynamics of the real-world environment.