Seite - 110 - in Adaptive and Intelligent Temperature Control of Microwave Heating Systems with Multiple Sources

4. ControlSystemDesign generatedas JIC= 1 N N∑ i=1 (Yi−Yt)2+αmax(Ymax−Yt)2+αmin(Ymin−Yt)2 , (4.30) whereαmax andαmin are coefficients to adjust weights of Ymax and Ymin, respectively. In equation 4.30, the maximum and the minimum temperatures of the whole load are also included in the cost function, whichismoreaccurate toreflect thetruetemperaturehomogeneityof the heated load. However, the problem regarding this cost function is that the maximum and the minimum temperatures are not fixed in certain locations, and their locations are varying during the heating process. It is not feasible to build a model to describe the relationship between the maximum and the minimum temperatures and individ- ual microwave feeding power. As a result, normal control techniques aredifficult tobeapplied in this situation. 4.2.1. ReinforcementLearning The development and implementation of reinforcement learning (RL) in the control field provides a perfect alternative to deal with afore- mentioned intelligent control tasks. The idea of reinforcement learn- ing was originally inspired by the biological learning process [LV09], like the learning behaviors of human beings and other animals. In RL ifanaction is followedbyasatisfactorystateofaffairsoran improve- ment in the state of affairs, it will receive a positive (or less negative) rewardandthetendencytoproducethatactionisstrengthened,which is reinforced. Otherwise if an action is followed a non-satisfactory state of affairs, it will receive a negative (worse than the first case) reward and the tendency to produce that action is weakened [Bar94]. TheultimateobjectiveofRLis tofindtheoptimalaction(control)pol- icythatmaximizestheoverallrewardsobtainedduringtheentirecon- trolprocess. Unlike the supervised learning approach introduced in the previous chapter 3, where the learning is based on datasets provided by a knowledgablesupervisor, there isnopredefinedlearningtargetgiven 110