PhD defence by Songda Wang on Advanced Control Strategies of Modular Multilevel Converters
21.12.2020 kl. 13.00 - 16.00
Songda Wang, Department of Energy Technology, will defend the thesis "Advanced Control Strategies of Modular Multilevel Converters"
Advanced Control Strategies of Modular Multilevel Converters
Professor Remus Teodorescu
Associate Professor Sanjay Chaudhary
Associate Professor Tamas Kerekes
Professor Frede Blaabjerg, Dept. of Energy Technology, Aalborg University (Chairman)
Kalle Ilves, ABB Sweden
Professor Marco Liserre, Christian-Albrechts-Universität of Kiel,
Voltage source converter (VSC) based high voltage direct current (HVDC) transmission systems have many merits including the flexibility to control the active/reactive power transmission of the system and to connect two power systems of different frequencies, etc. In the VSC-HVDC system, the converters are the most important part. The Modular Multilevel Converter (MMC) is the most popular converter topology for the VSC-HVDC system. With the help of the MMC, the HVDC system will have high scalability, low harmonics, and good fault tolerance. However, Because of the numerous submodules in the MMC, this can pose some challenges for MMC-HVDC systems. In this thesis, two following challenges of the MMC-HVDC system are researched: (1). The submodule capacitor voltage ripple problem under unbalanced grid conditions; (2). The high computation burden problem of the model predictive control of MMC.
To address those problems, this Ph.D. project proposes analytical models, necessary grid conditions to analyze the submodule capacitor voltage ripple of MMC under unbalanced grid conditions. The analytical equations of the submodule capacitor voltage under an unbalanced grid fault are derived to help to understand the behavior of the submodule capacitors under unbalanced grid conditions. Then, two capacitor voltage balancing methods are proposed to maintain the submodule-capacitor voltages balanced and within the desired limits, avoiding the tripping of the converter due to overvoltages. Then, a machine learning based capacitor ripple reducing method is proposed. With the machine learning network approach, the complicated non-linear capacitor voltage ripple model is replaced by neural networks based on deep learning method with a low computational burden. What is more, the circulating current reference is determined by the machine learning network to reduce and balance the ripple.
A machine learning based model predictive (MPC) control emulation for MMCs is proposed in this Ph.D. project to achieve fast dynamic response with a low computational burden. Two machine learning networks are applied to replace the traditional proportional–integral (PI)/ proportional-resonant (PR) controller to control the MMC systems, then the complicated control parameter design is avoided. The machine learning based controller can achieve the same dynamic response as the MPC, thus the proposed machine learning based controller could be suitable for the applications which need high response speed. What is more, the high computation burden problem of the MPC controller is also addressed by the proposed machine learning based method. The computation-light machine learning networks emulate the behavior of the MPC controller to control the MMC with a lower computational burden.
THE DEFENCE IN ENGLISH - all are welcome.
Department of Energy Technology