Performance Evaluation of AI-Driven Photovoltaic Output Forecasting

Solar Photovoltaic is considered as a clean source of energy and the most cost-effective renewable energy source in many developed countries. Nevertheless, photovoltaic power generation is unpredictable and imprecise due to its high variation that can be caused by the non-stationary multivariate time series data making their forecasting challenging. Therefore, in this work, we evaluate off-the-shelf AI-driven models to PV energy production forecasting based on historical production data and meteorological data by applying three pre-processing components, namely, outliers and missing data processing, data normalization and features reduction. Experiments and statistical analysis are conducted on three DKASC datasets collected from a desert climate region and our WATT 3.0 solar dataset collected from a Mediterranean climate region. The models' performances are compared to state-of-the-art methods and reveal lower error rates.