Reinforcement learning-based control of improved hybrid current modulated dual active bridge AC/DC converter
Küçük Resim Yok
Tarih
2022
Yazarlar
Dergi Başlığı
Dergi ISSN
Cilt Başlığı
Yayıncı
Springer London Ltd
Erişim Hakkı
info:eu-repo/semantics/closedAccess
Özet
The DC bus voltage of AC/DC converters is conventionally regulated by proportional plus integral (PI)-based controllers. However, such controllers can't provide good performance over the entire operating region and their performance is also affected by the change of parameters of passive components depending on temperature during the converters' operation. On the other hand, reinforcement learning (RL), which is one of the machine learning methods, can become immune to parameter changes as it maintains training during the converter operation. In this paper, an RL-based control algorithm is proposed for an AC/DC dual active bridge (DAB) converter which operates with improved hybrid current modulation (iHCM). In the proposed method, the model-free Q-learning algorithm of RL is used to train an agent to regulate the DC bus voltage. The proposed algorithm is verified for various load and disturbance conditions by MATLAB/Simulink simulations, and it is compared with a PI controller which is tuned with Internal Model Control (IMC) method. According to the simulation results, besides the online learning advantage, the proposed method creates a small settling time and overshoot at the start-up for light load conditions, unlike the PI controller. On the other hand, during the change of dynamic load and AC grid voltage, it creates smaller voltage oscillations in the output DC voltage and regulates it faster. Furthermore, since the proposed method keeps the duty cycle value constant in each grid period, it produces lower total harmonic distortion (THD) than the PI controller.
Açıklama
Anahtar Kelimeler
Reinforcement learning, Q-learning, AC, DC converter, Dual active bridge (DAB) converter, Dc Converter, Scheme
Kaynak
Neural Computing & Applications
WoS Q Değeri
Q2
Scopus Q Değeri
Q1
Cilt
34
Sayı
7
