Marcus Bellander	

present his master thesis

Non-linear optimization using Constrained Neural Networks with Applications on
the Nordic Energy Market

Abstract

In this thesis we study time series modeling and forecasting by using
feed-forward neural networks. Neural networks are data-driven
universal approximators, i.e. black box-models. The scope of the
thesis has been to incorporate a priori knowledge in the training of a
neural network, thus creating a gray box-model. A method of analyzing
such a model is presented, as well as a demonstration of these methods
on both synthetic and real problems from the Nordic Energy Market.