The variability of solar generation presents challenges to the management of PV projects, and there is no benchmark for the forecasting of solar irradiance in Jordan, despite the country’s renewable energy programme. To overcome the challenges of solar generation variability, machine learning models are proposed in this paper to estimate solar irradiance in Ma’an, using satellite data from NASA POWER in MATLAB. The compact ANN model produces the best results for irradiance estimation (R² = 0.986, MAE = 23.5 Wh/m²). Furthermore, by using the irradiance estimations to schedule maintenance for the Shams Ma’an plant, costs due to plant outages were reduced by 56%, saving the plant 625 MWh per year. Results of this study establish the first benchmark for solar irradiance forecasting in Ma’an.