Abstract
With the proliferation of converter-interfaced generation in modern power systems, grid-forming converters are viewed as a solution to improve system stability and resilience in weak power grids. However, the dynamic behaviour of the grid-converter systems is strongly influenced by inevitable disturbances and transients in weak power grids (e.g. short circuit faults). Moreover, grid-converter systems are prone to harmonic instability due to the interactions between the converters and passive elements. These issues pose security risks and limit the further integration of renewable generation into the modern power system. Therefore, this thesis aims to improve the transient stability and harmonic stability of grid-converter systems by employing deep-learning and analytic methods for developing power synchronization control (PSC) in weak power grids. First, the internal structure of the synchronization loop in PSC is modified to reduce vulnerability to grid transients by utilizing a back-calculation scheme. Also, the damping characteristics of PSC are enhanced to mitigate the decaying DC offset current of the converter. Second, the internal reference calculation is developed by embedding a long short-term memory (LSTM) neural network into PSC. The LSTM neural network is trained to extract and predict the grid voltage trajectory based on the converter dynamics and grid strength. Thus, the control system updates the internal references dynamically to meet the low-voltage ride-through (LVRT) requirements and prevent synchronization loss. Third, by employing deep learning methods and neural networks, an encoder-stacked classifier is introduced for early detection of synchronization instability. This allows time for corrective control actions to be taken and prevents synchronization loss in the grid-converter system. The applied neural networks are trained to be robust against data corruption and added noise. Finally, the admittance characteristics of converters are studied and necessary conditions are outlined for achieving harmonic stability with PSC in weak grids. Moreover, a 12.5-kVA three-phase back-to-back converter system is implemented under weak grid conditions for the experimental evaluation of the results and future works.
Translated title of the contribution | Machine Learning Approaches to Improving the Transient Stability of Voltage-Source Converters in Weak Grids |
---|---|
Original language | English |
Qualification | Doctor's degree |
Awarding Institution |
|
Supervisors/Advisors |
|
Publisher | |
Print ISBNs | 978-952-64-1208-5 |
Electronic ISBNs | 978-952-64-1209-2 |
Publication status | Published - 2023 |
MoE publication type | G5 Doctoral dissertation (article) |
Keywords
- deep learning
- grid faults
- grid-forming converters
- harmonic stability
- machine learning
- neural networks
- power electronics
- power synchronization control
- transient stability
- weak power grids