Seite - 129 - in Adaptive and Intelligent Temperature Control of Microwave Heating Systems with Multiple Sources
Bild der Seite - 129 -
Text der Seite - 129 -
4.2. IntelligentControl
of possible states forYmin is 4. The number of total position states is
5×4 =20.
Based on the discretization schemes of temperatures and positions,
the overall number of possible states for the modeled MDP should be
4×20 = 80. This number could be further reduced using the state
abstractionstrategy[Die99]. Thecontrolobjective is toconstraintboth
Ymax andYmin in the predefined temperature range. As long as they
are in the range, the states of positions ofYmax andYmin become less
important. Therefore all states withSmax = 1 andSmin = 1 can be
combined into one state. Although this state abstraction makes state-
actionvaluerelatedwiththisstatemorestochastic, thenumberoftotal
state-action pairs can be largely reduced due to this abstraction. As a
result, the number of total states after the state abstraction is |S| =
20×3+1 = 61.
Besides the states, the number of total actions is also reduced based
on the empirical knowledge. In the new HEPHAISTOS cavity 3, there
are 18 microwave heating sources. Apparently there is no way to in-
clude all of them in theQ(λ) learning controller (218 actions). Consid-
ering the applicable scenario of this controller is the flat-temperature
periodwherenohighpowerheatingisneeded,severalrulesaremade
toscreenfor thesuitablecontrolactionsas following.
1. Each HEPHAISTOS module has 6 heating sources, but only 4 of
themareused. Heatingsourcesno. 5andno. 8 (seefigure 2.5)are
R1 R2
R4 R5
R3
Figure4.13. Areadiscretization.
129
Adaptive and Intelligent Temperature Control of Microwave Heating Systems with Multiple Sources
- Titel
- Adaptive and Intelligent Temperature Control of Microwave Heating Systems with Multiple Sources
- Autor
- Yiming Sun
- Verlag
- KIT Scientific Publishing
- Ort
- Karlsruhe
- Datum
- 2016
- Sprache
- englisch
- Lizenz
- CC BY-SA 3.0
- ISBN
- 978-3-7315-0467-2
- Abmessungen
- 14.8 x 21.0 cm
- Seiten
- 260
- Schlagwörter
- Mikrowellenerwärmung, Mehrgrößenregelung, Modellprädiktive Regelung, Künstliches neuronales Netz, Bestärkendes Lernenmicrowave heating, multiple-input multiple-output (MIMO), model predictive control (MPC), neural network, reinforcement learning
- Kategorie
- Technik