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.

What problem does this paper attempt to address?

The key problem that this paper attempts to solve is the intermittency of solar radiation, specifically the instability of solar power due to cloud occlusion. This has hindered the wide application of solar energy in commercial and residential environments. Therefore, in order to better schedule and allocate grid resources, real - time solar radiation prediction for photovoltaic systems connected to the grid is required. ### Main problems 1. **Solar radiation fluctuations caused by cloud occlusion**: The occlusion, reflection or scattering of clouds directly affects the amount of received solar radiation, making the solar energy output unstable. 2. **Challenges of long - term prediction**: Existing optical - flow - based methods have deficiencies in predicting cloud movement, especially the cloud movement near the horizon, resulting in inaccurate long - term predictions (more than 30 seconds). 3. **Image distortion problem**: Sky images captured by ground - imaging devices are severely compressed near the horizon, affecting the accurate prediction of cloud movement. ### Solutions To solve these problems, the paper proposes the following methods: 1. **Spatial Warping**: - By introducing a deep - learning method to predict future sky - image frames and using spatial warping techniques to correct image distortion near the horizon, thereby improving resolution and prediction accuracy. - The core idea of spatial warping is to map the original image to a new space, in which the apparent movement of clouds remains consistent at the zenith and the horizon, thus ensuring accuracy over a longer prediction time. 2. **Multi - frame input prediction model (Multi - image Prediction)**: - Use multiple past image frames as input to provide a larger temporal context, thereby improving the robustness and accuracy of prediction. - Specifically, in order to predict an image at a future moment, the model will consider images at multiple past time points, such as {t−5, t−3, t−1, t}, and use this information to predict future image frames. 3. **Improved loss function**: - Introduce a composite loss function that combines intensity, gradient and motion to ensure that the predicted image is as close as possible to the real image in terms of pixel values, edge details and motion consistency. - The loss function includes: - Intensity loss \( L_{\text{int}}(bI, I) = \| bI - I \|_2^2 \) - Gradient loss \( L_{\text{gd}}(bI, I) = \sum_{i,j} \| |bI_{i,j} - bI_{i - 1,j}| - |I_{i,j} - I_{i - 1,j}| \|_1 + \| |bI_{i,j} - bI_{i,j - 1}| - |I_{i,j} - I_{i,j - 1}| \|_1 \) - Motion loss \( L_{\text{op}} = \| f(bI_{t + 1}, I_t) - f(I_{t + 1}, I_t) \|_1 \) ### Summary Through the above methods, the paper aims to improve the prediction accuracy of future sky images, especially for long - term predictions (more than 30 seconds), so as to better support the scheduling and resource allocation of solar energy systems.

Precise Forecasting of Sky Images Using Spatial Warping