Daily Peak Load Forecasting by a Correntropy based Artificial Neural Network using an Adaptive Kernel Size Method Considering Outliers

Abstract

This paper proposes daily peak load forecasting by a correntropy based Artificial Neural Network (ANN) using an adaptive kernel size method for reduction of engineering loads considering outliers. When outliers exist in the training data, estimation accuracy of daily peak load forecasting using a conventional least mean square (LMS) based ANN can be affected by the outliers. Therefore, engineers have to remove the outliers in order to improve estimation accuracy and it is a heavy burden to engineers. Although Correntropy has a possibility to solve this problem, adjustment of a kernel size has been a big challenge for correntropy. Effectiveness of the proposed method is verified by comparison with a conventional LMS based ANN using Stochastic Gradient Descent (SGD), a Correntropy based ANN using SGD with a fixed kernel size and a Correntropy based ANN using SGD with the conventional adaptive kernel size method.