Because of Korea's rapid expansion in photovoltaic (PV) generation, forecasting long-term PV generation is of prime importance for utilities to establish transmission and distribution planning. However, most previous studies focused on long-term PV forecasting have been based on parametric methodologies, and most machine learning-based approaches have focused on short-term forecasting. In addition, many factors can affect local PV production, but proper feature selection is needed to prevent overfitting and multicollinearity. In this study, we perform feature-selective long-term PV power generation predictions based on an ensemble model that combines machine learning methods and traditional time-series predictions. We provide a framework for performing feature selection through correlation analysis and backward elimination, along with an ensemble prediction methodology based on feature selection. Utilities gather predictions from various sources and need to consider them to make accurate forecasts. Our ensemble method can produce accurate predictions using various prediction sources. The model with applied feature selection shows higher predictive power than other models that use arbitrary features, and the proposed feature-selective ensemble model based on a convolutional neural network shows the best predictive power. INDEX TERMS Ensemble learning, forecasting, long-term forecast, machine learning, power system planning.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.