This thesis proposes an intelligent approach to solve the problem of maximum power point tracking (MPPT) control for thermoelectric generator (TEG). The approach is based on a off-policy model-free reinforcement learning algorithm named Deep Deterministic Policy Gradient (DDPG). The fundamental of TEG, DC/DC converters and MPPT are introduced. The concepts and approaches of reinforcement learning are also analyzed. Then the crucial modification and improvement are implemented with respect to the original DDPG method so as to create the proposed control algorithm for this specific MPPT issue in TEG circuit. The modified RL algorithm possesses several advantages compared with other reinforcement learning methods, such as continuous-action control, low computational cost, high stability, etc. The control system with the proposed RL algorithm is superior to traditional MPPT methods, e.g. P\&O, in many aspects such as rapid response to optimal point and low steady-state fluctuation. In addition, the simulation and experiment systems are designed and established to test the proposed RL MPPT control system for TEG. The analysis of the results shows the novel method promote the control performance obviously.
This thesis proposes an intelligent approach to solve the problem of maximum power point tracking (MPPT) control for thermoelectric generator (TEG). The approach is based on a off-policy model-free reinforcement learning algorithm named Deep Deterministic Policy Gradient (DDPG). The fundamental of TEG, DC/DC converters and MPPT are introduced. The concepts and approaches of reinforcement learning are also analyzed. Then the crucial modification and improvement are implemented with respect to the original DDPG method so as to create the proposed control algorithm for this specific MPPT issue in TEG circuit. The modified RL algorithm possesses several advantages compared with other reinforcement learning methods, such as continuous-action control, low computational cost, high stability, etc. The control system with the proposed RL algorithm is superior to traditional MPPT methods, e.g. P\&O, in many aspects such as rapid response to optimal point and low steady-state fluctuation. In addition, the simulation and experiment systems are designed and established to test the proposed RL MPPT control system for TEG. The analysis of the results shows the novel method promote the control performance obviously.
Intelligent maximum power point tracking for thermoelectric generator based on a reinforcement learning method
WEI, XING
2017/2018
Abstract
This thesis proposes an intelligent approach to solve the problem of maximum power point tracking (MPPT) control for thermoelectric generator (TEG). The approach is based on a off-policy model-free reinforcement learning algorithm named Deep Deterministic Policy Gradient (DDPG). The fundamental of TEG, DC/DC converters and MPPT are introduced. The concepts and approaches of reinforcement learning are also analyzed. Then the crucial modification and improvement are implemented with respect to the original DDPG method so as to create the proposed control algorithm for this specific MPPT issue in TEG circuit. The modified RL algorithm possesses several advantages compared with other reinforcement learning methods, such as continuous-action control, low computational cost, high stability, etc. The control system with the proposed RL algorithm is superior to traditional MPPT methods, e.g. P\&O, in many aspects such as rapid response to optimal point and low steady-state fluctuation. In addition, the simulation and experiment systems are designed and established to test the proposed RL MPPT control system for TEG. The analysis of the results shows the novel method promote the control performance obviously.File | Dimensione | Formato | |
---|---|---|---|
2018_12_Xing_Wei.pdf
non accessibile
Descrizione: Thesis text
Dimensione
12.31 MB
Formato
Adobe PDF
|
12.31 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/144039