Page - 101 - in Adaptive and Intelligent Temperature Control of Microwave Heating Systems with Multiple Sources

4.1. AdaptiveControl For example, according to empirical evidences from numerous exper- iments of HEPHAISTOS, a good upper power limit Umax can be ob- tained by using a simple proportional-integral (PI) controller. In this case, the number of totally activated feeding sources in the GA con- troller isalwaysboundedbythepowercalculatedbythePIcontroller, which not only guarantees the control stability but also reduces the searching range of the input space. The practical GA based nonlinear MPCsystemisshownasfigure 4.3. Plant (HEPHAISTOS) GA Controller Nonlinear System Identification Target Temperature Yt Measured Temperature Yr Control Input U PI Controller Power Limit Umax Estimated System Dynamics [A], [Φ] Figure4.3. Practically implementedGAbasednonlinearMPCsystem. 4.1.2. NeuralNetworkbasedControl The idea of using neural networks as the controller has been stud- ied and implemented massively in numerous applications [PSRJG00] [LP02] [PW08]. In general there are two main structures that apply neural network based control (NNC) [LV09], such as in figure 4.4. The first structure ( 4.4a) is called indirect NNC, which has a simi- lar topology with traditional feedback control system and involves both the NN estimator and the NN controller. The estimator learns the dynamics of the unknown system and the controller uses the es- timated model to control the real plant. The second structure ( 4.4b ) contains only a NN controller which directly controls the real plant. 101