Short-term Solar Irradiance Forecasting Based on Multi-Branch Residual Network

For having a stable and reliable smart grid system, an accurate short-term solar irradiance forecasting is necessary. The challenges arise while solar energy has variable and fluctuating nature due to complex weather conditions, temperature, and meteorological factors. Machine learning and deep learning techniques besides traditional time-series forecasting methods are considered to be effective tools to have precise short-term solar forecasting system, while they are analyzing the time-series solar data at single resolution and unable to capture their sudden and short variations. In this paper, we propose a novel deep architecture consisting multi-branch residual network (ResNet) to model the solar irradiance data at different resolutions and extract hierarchical features to improve the forecasting accuracy. We evaluate the performance of the proposed model relative to other deep learning models, ResNet, and long short-term memory (LSTM), using seventeen years of data from twelve different sites in Philadelphia. Numerical results show the state-of-the-art performance on half-day-ahead forecasting.