A Correntropy Based Artificial Neural Network using Early Stopping for Daily Peak Load Forecasting

This paper proposes a correntropy based Artificial Neural Network (CANN) using early stopping (ES) for daily peak load forecasting. The proposed method utilizes stochastic gradient descent with momentum (SGDM) as a learning algorithm. In order to provide an electric power generation schedule for pumped-storage hydro plants and gas turbine power plants operated in peak load periods, it is important to forecast the daily peak load accurately in advance. Correntropy and ES are applied in order to prevent over-fitting of ANNs. Effectiveness of the proposed method (CANN using SGDM and ES) is verified by comparison with the conventional least mean square based ANN (LANN) using SGD, a LANN using SGDM, a LANN using Adam, a LANN using SGD and ES, a LANN using SGDM and ES, a LANN using Adam and ES, and a CANN using Adam and ES.