Load Forecasting Using Independent Component Analysis as Preprocessing

Abstract

In this letter, we propose a preprocessing method for Multi Layer Perceptron (MLP) forecasting method using Independent Component Analysis (ICA). We use the ability of ICA capturing underlying structures in time-series data, and apply ICA to MLP forecasting problem. Specifically, we try to improve accuracy of forecasting by dimension reduction of input variables using Complexity Pursuit which is one of the ICA algorithms based on spatial and temporal structures in the data.