Seite - 166 - in Adaptive and Intelligent Temperature Control of Microwave Heating Systems with Multiple Sources

5. ExperimentalResults adaptivesystemidentificationalgorithms. However,whenseveralac- cidents occur in a row during a short time, it could cause much larger unexpected system estimation errors and corresponding wrong con- trol inputs (see theresults later) From the results in figure 5.25, it is clear that both the linear and the nonlinearMPCmethodsaremuchbetterthanthePIDcontrolmethod, withrespecttothefinaltemperaturewindow. Bothofthemhavemuch smallerfinal temperaturewindowscomparedwiththePIDcontroller. More important, the temperature curves in figures 5.25 have obvious converging behaviors, which indicate temperatures of different loca- tions are effectively controlled towards the target temperature value. For example, in the first figure 5.25a, the temperature window in the beginning of the flat-temperature period is more than 10◦C. Dur- ingtheheatingprocess,alldifferent temperaturecurvesaregradually converging to the target temperature value (75◦C) and the tempera- ture window is also decreasing. In the end of the flat-temperature period, the final temperature window∆T1 is between3.6◦Cand4◦C. This converging behavior occurs in both the linear and the nonlinear MPC methods, and it reflects the active temperature distribution im- provingabilityofbothMPCmethods. ForthelinearMPC,therearetwosystemidentificationmethodsavail- able such as introduced in chapter 3. Both RLS and RKF have been tested and compared in our experiments, and corresponding results can be found in figures 5.27. There is no obvious difference between the performance of RLS and RKF, which also coincides with the sys- tem identification results shown before. It should also be noted that thecontrolperformanceof the linearMPCmethodwithmoreheating sources (in figure 5.27) is comparable to the results in the new CA3 withfewerheatingsources(infigure 5.25a). It indicates that thenum- berofheatingsourcesdoesnot influencethecontrolperformanceand final temperature homogeneity of the linear MPC method, which is similarwith thesituation in thePIDcontrol. The control performance of nonlinear MPC is similar with that of the linear MPC. The converging behavior is obvious and the final tem- perature window is also much smaller than in the PID controller. A problem occurred in the nonlinear MPC is that the large temperature overshoot in the beginning of the flat-temperature period (see figure 166