Page - 124 - in Adaptive and Intelligent Temperature Control of Microwave Heating Systems with Multiple Sources

4. ControlSystemDesign period. Compared with the first raising-temperature period, what is more important to the final temperature distribution is the flat- temperature period (the second red part in figure 4.11). In this case, the TD learning controller is perfectly suitable because of its ability to dealwith intelligentcontrol tasksandhighlynonlinearplants. Finally, the entire control task of the TD learning controller is signif- icantly simplified by the hybrid control structure, from following a varying temperature in a large temperature range to converging to a fixedtemperature inasmall temperaturerange. Asaresult, thestruc- ture of the TD learning controller is also simplified and correspond- ingly thewhole learningprocess is shortened. LookuptablebasedQ(λ)-learning After defining its applicable region, the next step is to determine the form of the TD learning controller. In principle, all kinds of TD learn- ingcontrollerscouldbeappliedinHEPHAISTOS.Anintuitivewayto generate the MDP is to treat the temperature value as the state vari- ableandthecontrol inputas theactionvariable. Bothstateandaction variables are all continuous and it follows logically that the function approximationbasedACmethods(boththecriticandtheactorusethe functionapproximationapproach)shouldbeimplemented. However, it is also natural to convert the continuous states and actions into dis- Time Temperature Raising-temperature period Flat-temperature period Target Temperature Curve Figure4.11. Target temperaturecurveduringtheheatingprocess. 124