Short‐term solar power forecasting based on convolutional neural network and analytical knowledge.

Summary: With a high proportion of variable renewable energy integration, accurate forecasting approach is of vital importance in ensuring the reliable and economic operation of power system. Therefore, in this article, a novel method for predicting photovoltaic (PV) power generation based on convolutional neural network (CNN) is proposed. Analytical models of PV systems are formulated, thereby providing physical knowledge about the relationship between PV output and critical meteorological features. To explore the nonlinear and time‐varying properties of PV output, CNN is adopted in this article, which matches the patterns of similar days. Case studies based on realistic datasets in Australia demonstrate that the forecasting performance for solar power can be effectively improved by taking advantage of the proposed CNN‐based learning method. [ABSTRACT FROM AUTHOR]