Towards Solar Irradiance Forecasting Using Deep Learning on the Edge

The appearance of clouds in the sky and their movement can cause significant differences in the sun's direct irradiance on the surface of the earth and significantly affect the energy production of photovoltaic (PV) parks. Irradiance forecasting can prevent perturbations on the power supply and the grid balance. This process utilizes a module that has as its input a sky image and estimates the irradiance with image processing algorithms based on neural network techniques. Aiming at a performance effective irradiance estimation module that is also efficient with respect to the hardware requirements and therefore suitable for the edge computing paradigm, the current paper presents the study towards the development of a Convolutional Neural Network (CNN) based model that is able to process a sky image and produce a single continuous value that represents the estimated irradiance for the particular sky image. The real-time performance is validated by the implementation on the edge-oriented Xilinx Zynq UltraScale+ MPSoC FPGA.