Abstract
The clouds’ position in the sky, their creation, movements and dissipation have a significant impact on the energy produced in photovoltaic (PV) parks. To improve the power supply and the grid balance, the sky cameras will provide the images of the sky and edge computing devices will estimate the clouds’ coverage in the sky and consequently compute the upper amount of energy that can be produced. For the execution of these processes the PV plants depend on systems of considerable cost that include the cameras and the supporting software most often executed on edge computing devices. The requirements for extreme weather conditions of these application specific cameras and their supporting hardware and software lead to high-cost figures. These facts decrease the efficiency defined as the results accuracy over the system’s cost. Aiming at efficient solutions in the course of developing a PV park controller [1], the current study presents and compares the results of applying a cloud coverage process on systems with cameras of different cost and specifications. Moreover, it proposes an effective image segmentation technique for the cloud coverage executed on edge devices and it compares its accuracy to that of commercially available software. The comparison analyzes the differences in the accuracy results of three system configurations using each of two different cameras and either the proposed cloud segmentation algorithm or the commercial software. The real time performance of the proposed system algorithmic techniques on the edge-oriented NVIDIA Jetson Nano [2] validates the techniques.
