Combining Forecasts with Blind Signal Separation Methods in Electric Load Prediction Framework

R. Szupiluk, P. Wojewnik, and T. Zabkowski (Poland)


Forecasts combining, neural networks, blind signal separation, model improvement


In this paper we present a novel method for prediction improvement when many models are used. Our aim is to find in the modeling results the common basis components and process them to filter the noises and destructive signals. The basis components are found by blind separation methods like PCA or ICA. The constructive signals are integrated using an inverse system to decomposition or neural network. We check the validity of our methodology on load prediction task.

Important Links:

Go Back