Increasing the Accuracy of Hourly Multi-Output Solar Power Forecast with Physics-Informed Machine Learning