PV Generation Forecasting With Missing Input Data: A Super-Resolution Perception Approach

For practical PV generation forecasting, it is sometimes the case that the input data is missing or incomplete due to measurement and recording errors, which makes the application of a machine learning-based PV forecasting model difficult or impossible. This letter presents a new forecasting framework to address this practical and important issue. The super-resolution perception convolutional neural network (SRPCNN) is used to recover the missing data and the stochastic configuration network (SCN) is used to forecast PV generation with the recovered data. The proposed method is tested on an open dataset, and two existing methods are compared as benchmarks. The testing results verify its excellent missing-data recovery ability for accurately forecasting PV generation even under severe missing-data conditions.