The Use of Machine Learning Algorithms in Optimising SGS for Synchronising

An examination of the relative efficacy of machine learning (ML) algorithms for power supply and demand prediction in smart grid systems is warranted by their growing application. Through meticulous hyperparameter tweaking, we assessed six well-known machine learning algorithms: support vector machine (SVM), random forest, k-nearest neighbor, linear regression, artificial neural network, and Gaussian regression (GR). When the ML algorithms were appropriately calibrated, we found little to no significant difference in performance among them when we used Eskom database data for day-ahead forecasting, including hourly demand and renewable energy generation. On the other hand, wind power forecast proved more difficult because of its unpredictable nature than photovoltaic (PV) electricity. With the exception of GR, statistical analysis showed no discernible changes in timing performance amongst the algorithms, despite SVM’s somewhat quicker processing time. The implication is that, with appropriate hyperparameter fine-tuning, using different machine learning algorithms for power prediction—particularly for sources with regular patterns, such as solar PV or daily consumption rates-may not always result in statistically significant variations in prediction quality.