| dc.description.abstract | Climate change has forced the global energy sector to transition from energy production based on fossil fuels such as oil, natural gas, and coal towards renewable energy sources i.e., solar energy, wind energy, biomass, etc. Countries across the world have set targets to reduce greenhouse gas (GHG) emissions significantly by 2050. For example, the EU is committed to reducing GHG emissions by 85-90% by 2050. Renewable energy would serve as a key driver in achieving these ambitious targets.
To meet these ambitious goals, many projects have been launched in recent years, for example, Supergrid with the aim to interconnect various European countries. However, this is not a simple task. Cross-border electricity flow would require high voltage direct current (HVDC) technology, which can carry more electricity efficiently over long distances. Many HVDC projects interconnecting different countries have been recently launched. For instance, a 190 km HVDC interconnection (Piedmont-Savoy HVDC Line) links the French and Italian electricity networks. Similarly, other projects such as NorthSea DC grid, Medtech, Desertec, etc have been undertaken.
Modular multilevel converters (MMCs) have emerged as one of the most promising technologies for HVDC applications owing to their remarkable features such as their modular nature, reduced or no filter requirements, and redundancy. However, the control of MMC is a difficult problem as in addition to the control of output current, the internal dynamics of the MMC need to be well controlled. Moreover, the bilinear model of the MMC introduces non-linearity through the product of the states and inputs.
At the start of this thesis work (early 2020), many control methods had already been investigated for MMC. These included conventional cascade control methods which due to their linear nature result in sluggish dynamic performance as compared to advanced control methods such as model predictive control. Moreover, due to the presence of multiple control loops, their design becomes complex. Then many non-linear methods such as backstepping, sliding mode control, feedback linearization, and model predictive control had also been tried.
Among the existing methods, model predictive control is an effective method to deal with the multi-input multi-output (MIMO) nature, nonlinearity of MMC and can easily handle the constraints. For power converters generally, finite control set model predictive control (FCS-MPC) is used. The existing methods based on FCS-MPC for MMC were not able to achieve high performance with low computational complexity simultaneously. The methods with low computational complexity resulted in sluggish dynamic response whereas the methods which offered high dynamic performance had high computational complexity making the real-time application of FCS-MPC difficult. As a result, an extended prediction horizon which generally improves the performance of the system as it can prepare early for the effects in the future could not be used with existing methods.
Moreover, at the start of this work, there was no published work based on MPC that considered all the practical delays i.e., forward (control) and backward (feedback) for MMCs. Only a forward delay of one sampling instant is considered mostly in the existing literature.
The steady-state performance of FCS-MPC for MMC is also an issue that has been investigated. In the existing literature, this is handled by modulated MPC. These modulated MPC techniques are either based on continuous control set MPC (CCS-MPC) or indirect FCS-MPC methods where duty cycles are calculated for each SM by considering the two or three voltage levels that give the minimum cost. However, CCS-MPC methods have very high computational complexity due to non-convexity (introduced by the nonlinear model) and having very small sampling times. On the other hand, the methods that are based on indirect FCS-MPC strategies do not have very good control of the circulating current.
In addition, when this work was started, all the methods based on MPC for MMC required an outer loop to regulate the summation voltages. The methods that proposed an equivalent of the outer loop within MPC suffered from continuous ripples in the circulating current. The methods without the outer loop or its equivalent within MPC resulted in a slow deviation of summation voltages away from reference.
Another issue that has been investigated for FCS-MPC for MMC is the variable and high switching frequency. The existing methods modify the sorting algorithm to lower the switching frequency, however, the variable switching frequency is not addressed by modification of the sorting algorithm. Methods that result in fixed switching frequency are based on modulated MPC with carrier phase-shifted PWM. However, as the number of SMs increases, the number of carriers also increases which makes it implementation complicated for applications with hundreds of SMs.
Finally, the existing methods for fault detection, localization, and tolerance for open circuit switch failure in MMCs are mainly based on conventional cascade control. Very few works based on MPC have addressed this issue. These works based on MPC either have very high computational complexity or their operating range is limited.
The main contributions of this work can be summarized as:
• Various methods with high performance and low computational complexity based on FCS-MPC are proposed
• Method to make the computational complexity of MPC for MMCs independent of the number of SMs is proposed.
• Methods to improve the steady-state performance of FCS-MPC are developed.
• A novel cost function is proposed to remove an outer loop or any kind of additional control over circulating current reference for summation voltage regulation.
• A novel circulating current reference is proposed to remove the need for an outer loop to regulate the summation voltages.
• Experimental verification of proposed methods by considering all the practical delays i.e., forward and backward is done
• A method for fault detection, localization, and tolerance using Indirect FCS-MPC for open circuit switch fault in MMCs is developed
• A modified sorting algorithm is also proposed that results in low and almost fixed switching frequency on an average of a few cycles. This ensures that the losses between the SMs are balanced.
• The transition from simulations to experimental verification led to some problems. As a solution to these problems, a very simple method for grid voltage estimation was developed. In addition, a virtual acside voltage approach is also developed in this thesis. | en_US |
