Solar Radiation Forecasting Using Artificial Neural Networks Considering Feature Selection

Abstract

Due to various factors, including worries about greenhouse gas emissions, supporting government policies, and decreased equipment costs, the expansion of solar-based energy generation, notably in the form of photovoltaics, has accelerated significantly in recent years. Solar panels continue to face several challenges regarding their practical integration and reliability. These concerns originate from the variable nature of the solar resource. Solar generation has inherent variability, which poses problems associated with the costs of supplemental generation and grid reliability. Therefore, high accuracy solar forecasting is required. Several machine learning strategies are broadly employed for solar power forecasting. However, analyzing solar radiation characteristics in order to select the features that have a meaningful correlation between inputs and outputs of machine learning algorithms has received less attention. This study uses a multilayer perceptron (MLP) artificial neural network (ANN) with Bayesian optimization to forecasting solar radiation. The Pearson Correlation Coefficients (PCCs) are used to select effective features. The simulation findings reveal that the accuracy assessment metrics are higher when employing feature selection for prediction.