Precise Forecasting of Sky Images Using Spatial Warping

TL;DR

The main contribution is to derive an optimal warping method to counter the adverse affects of clouds at the horizon, and learn a framework for future sky image prediction which better determines cloud evolution for longer time horizons.

Abstract

The intermittency of solar power, due to occlusion from cloud cover, is one of the key factors inhibiting its widespread use in both commercial and residential settings. Hence, real-time forecasting of solar irradiance for grid-connected photovoltaic systems is necessary to schedule and allocate resources across the grid. Ground-based imagers that capture wide field-of-view images of the sky are commonly used to monitor cloud movement around a particular site in an effort to forecast solar irradiance. However, these wide FOV imagers capture a distorted image of sky image, where regions near the horizon are heavily compressed. This hinders the ability to precisely predict cloud motion near the horizon which especially affects prediction over longer time horizons. In this work, we combat the aforementioned constraint by introducing a deep learning method to predict a future sky image frame with higher resolution than previous methods. Our main contribution is to derive an optimal warping method to counter the adverse affects of clouds at the horizon, and learn a framework for future sky image prediction which better determines cloud evolution for longer time horizons.

Figures (7)

• Figure 1: Predicting future sky images. Left-to-right: Results from SkyNet-UNet (the proposed technique), PhyD-Net-Dual 9150809, optical flow, and ground truth images. The methods take in as input images $[I_{t-5}, I_{t-3},I_{t-1},I_{t}]$, and predict future frames $[ \widehat{I}_{t+1}, \widehat{I}_{t+2},...,\widehat{I}_{t+5}]$; From top-to-bottom.
• Figure 2: Sample images captured by a TSI TSI.
• Figure 3: A cloud subtends a smaller angle when it is further away from the zenith. This results in the nonlinear spatial warping that is seen in Figure \ref{['fig:TSI_collage']}, and poses critical challenges for effective forecasting of cloud movement.
• Figure 4: Overview of how the 3D position of a cloud in the world space gets mapped to a point on the image plane using a hemispherical mirror.
• Figure 5: We show how the appearnce of an input image changes under the proposed warping. The plot on the left visualizes how we map from radial distances on the image to radial distances in the proposed representation. Choosing different values of the range of $\widetilde{\rho}$ produces different FOVs and associated distortions. This is visualized in the center column. The right column shows the image after inverting the warp to obtain the original image.