Methodology and practical aspects of signal filtration based on Blind Source Separation

This paper presents the methodological and practical aspects of signal filtration using Blind Signal Separation. The paper describes the whole process and discusses various methods and algorithms used in signal filtration: ICA FASTICA, JADE, AMUSE and Nonnegative Matrix Factorization (NMF). A practical experiment has been conducted to demonstrate the overall filtration framework and show the effectiveness of three chosen algorithms: JADE, AMUSE and NMF. The practical goal of this experiment was to improve the prediction of short-term energy consumption. While using NMF does not provide an improvement, the other two algorithms perform well and lead to a significant reduction in the prediction error.