Seite - 123 - in Adaptive and Intelligent Temperature Control of Microwave Heating Systems with Multiple Sources

4.2. IntelligentControl Plant (HEPHAISTOS) Adaptive Controller (MPC or NNC) Q(λ) Learning Controller Target Temperature Yt Measured Temperature Yr Control Input U No Power Input Controller Selection Figure4.10. HybridTDlearningcontrol system. later. This special hybrid control structure is developed because of several reasons. First, in most control tasks the target temperature is varying along time such as shown in figure 4.11. Apparently, a randomly initialed TD learning controller has the worse ability to follow the target tem- perature change than conventional adaptive controllers. Because it does not use the same system estimation approach as in conventional adaptive control systems and it has to take a large number of explo- ration actions to get the accurate state values. Besides, it is not rec- ommended to use a varying target in TD learning controllers, and the highly varying and stochastic environment/plant will also influence its learning results. Based on these reasons, it is better to use a con- ventionaladaptivecontrollertohandletheraising-temperatureperiod (thebluepart infigure 4.11). Second, according to numerous experimental data (can be found in thenextchapter),nomatterwhichcontrolmethodisapplied, thecon- trolled temperature distributions of different control algorithms are very similar to each other during the first raising-temperature period. Allcontrolalgorithmshavetocontrolbasedoninaccuratesystemesti- mations and no effective control actions can be made. In other words, as long as the target temperature curve is defined, it is not likely to improve the temperature distribution on the first raising-temperature 123