Implementation of Load Demand Prediction Model for a Domestic Load Center Using Different Machine Learning Algorithms—A Comparison

Abstract

AbstractTo comply with the advanced smart grid operations such as Artificial Intelligence (AI) based Distributed Generation (DG) Integration and Load schedules, learning of future load and supply availability is inevitable. Specifically, the use of Big Data analytics and prediction is very crucial as they have changed the paradigm of Conventional Grid operations. Since last two decades, research on Load demand (PT) forecasting is on high pedestal. And there has been more than a dozen Machine Learning (ML) algorithms reported in the literature. But, features/predictors selection was always a critical call in any ML based prediction. Not only that effective comparison and choice between numerous ML algorithms has always been a research challenge. To adress the said challenges, this article presents the load forecasting of a domestic load center using Feed Forward Artificial Neural Networks (FF-ANN) and nineteen different ML algorithms trained by the combination of weather and time stamp features/predictors. ML algorithm driven MATLAB-SIMULINK prediction model designed and developed can predict the Load demand for any given date if weather parameters are fed to it. In adition, an extensive comparison between different ML algorithms in terms of training time, prediction speed, Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Square Error (MSE), R2, Training Time in seconds and Prediction Time in Obs/Sec presented paves a way for researchers in selecting right ML algorithm for load forecasting problem concerning domestic load centers. Among all ML algorithms trained and tested, Rotational Quadratic Gaussian Process Regression (RQ-GPR) ML algorithm is witnessed to be with higher accuracy and lower RMSE. MATLAB 2018b licensed user added with Statistics and ML Tool box is used for the whole implementation.KeywordsLoad forecastingMachine learning algorithmsGaussian process regressionSupport vector regressionTree regressionEnsembled algorithmsArtificial neural networks