Short-Term Hours Ahead Forecast of Expected Available Solar Power Using Linear Regression Machine Learning Scheme

Integration of renewable energy sources which are intermittent in nature can exacerbate the inherent challenges in hours ahead operational planning. In this work a machine learning scheme based on linear regression is used to forecast hours ahead generation using a year data that is partitioned into daily and weekly categories. The performance measures of L
2
norm of error, Root Mean Square Error (RMSE), and Mean Absolute Error (MAE) are used to assess the scheme. The results show that forecast error based on RMS, of 1 hour ahead is lower (RMS = 0.19717) than that of 3 hours ahead (RMS = 0.86876) for all categories of data partitions. In addition, the performance errors are lower when weekly partition (RMS = 0.05575) is used compared to a daily partition (RMS = 0.19717) is used in making a forecast for 1 hours ahead forecast.