Comparing Photovoltaic Power Prediction: Ground-Based Measurements vs. Satellite Data Using an ANN Model

Abstract

Accurate prediction of photovoltaic (PV) power output is crucial for assessing the feasibility of early-stage projects in relation to specific site weather conditions. While various mathematical models have been used in the past for PV power prediction, most of them only consider irradiance and ambient temperature, neglecting other important meteorological parameters. In this article, a 1-year dataset from a high-precision meteorological station at the Green Energy Research facility is utilized, along with electrical parameters from a polycrystalline silicon c-Si PV module exposed during the study period, to forecast PV power. In addition, the accuracy of using satellite data for PV power forecasting is investigated, considering the growing trend of its utilization in recent research. Regression techniques, such as linear regression with interaction, tree regression, Gaussian process regression, ensemble learning for regression, response surface methodology, SVM cubic, and artificial neural network (ANN), are employed for PV power prediction, using both ground measurement data and satellite data. Comparatively lower accuracies are observed when using satellite data across all regression methods, in contrast to the higher accuracies achieved with ground-based measurements. Notably, the Gaussian process regression method demonstrates high accuracy ( R ² = 0.25 for satellite data and R ² = 0.94 for ground-based data). Furthermore, the ANN approach further enhances the accuracy of PV power forecasting, yielding R ² = 0.42 for satellite data and R ² = 0.96 for ground-based data. These findings emphasize the need for caution when relying on satellite data for PV power forecasting, even when employing advanced ANN approaches. It underscores the importance of considering ground-based measurements for more reliable and accurate predictions.