Seite - 102 - in Adaptive and Intelligent Temperature Control of Microwave Heating Systems with Multiple Sources

4. ControlSystemDesign Plant (HEPHAISTOS) NN Controller NN Estimator Target Temperature Yt Measured Temperature Yr Control Input U Estimated Weights Vector or Matrix (a) Indirect control structure Plant (HEPHAISTOS) NN Controller Target Temperature Yt Measured Temperature Yr Control Input U (b)Direct control structure Figure4.4. Twocontrol structuresofNNcontroller. Theweightsof thisNNcontroller isusually tunedusingtheunsuper- vised learning algorithms. Unlike in supervised learning, where the targetoutputof theNNisgiveninadvance, inunsupervisedlearning there is no explicit target output provided. Therefore the main chal- lenge in using a direct NN controller is to determine a suitable and realistic learning algorithm that can effectively tune the weights and guarantee thecontrolperformance. Thedynamicsoftherealplanttobecontrolledcanbedescribedbythe expression, suchas Y(k+1) =F (Y(k),U(k)), (4.20) 102