L/LC-filtered voltage source converters (VSCs) are two of the most fundamental converters in an AC power system, which have various industrial applications such as active power filter, grid-supported converter, grid-forming converter, motor drive, battery charger, ...etc. These applications make L/LC-filtered VSCs become essential parts which are indispensable. In order to drive L/LC-filtered VSCs, many control methods have been proposed such as proportional integral (PI) control, proportional resonance (PR) control, and model predictive control (MPC)...etc. Among them, finite control set MPC (FCS-MPC) is one of the most effective and simplest control method for L/LC-filtered VSCs due to its excellent dynamic performance, straight-forward operation principle, easy handling of system non-linearities and application constraints, and the capability of simultaneously controlling multiple objectives.
Besides these above advantages, FCS-MPC for L/LC-filtered VSCs also has two main drawbacks such as performance dependency on model parameter accuracy, and high hardware cost because a large number of sensors are required. To solve these problems, this thesis proposes some robust FCS-MPC strategies to drive L/LC-filtered VSCs with a minimum number of sensors.
Firstly, a new FCS-MPC branch known as finite control set model-free predictive power control (FCS-MFPPC) is presented to avoid model parameter mismatch for L-filtered VSCs. However, since conventional FCS-MFPPC suffers from another problem of stagnant current variation update, its performance is very poor. To solve these problems, an enhanced FCS-MFPPC strategy is proposed to eliminate the stagnant current variation update. Because not only the model parameter mismatch but also the stagnant current variation update are removed, the proposed FCS-MFPPC method shows high performance and robustness, which is superior to previous methods.
Secondly, a grid-voltage sensorless finite control set model-free predictive current control (FCS-MFPCC) strategy is developed for L-filtered VSCs, where not only model parameter mismatch and stagnant current variation update are avoided, but also grid-voltage sensor is removed. To eliminate the stagnant current variation update, current variations are kept updated at every sampling period by updating the intermediate terms to reconstruct all current variations. On the other hand, grid-voltage sensorless operation is guaranteed by detecting the phase angle of the grid voltage from a current variation rather than from grid-voltage measurement used in conventional FCS-MFPCC. In addition, measurement noise, which highly influences the performance of FCS-MFPCC, is also considered in the proposed method. Compared to previous methods, the proposed method has superior features as follows: (1) Strong robustness against model parameter mismatch and measurement noise (2) High performance with the elimination of stagnant current variation update (3) Grid-voltage sensorless operation.
Thirdly, this thesis proposes a simple FCS-MPC strategy to improve the robustness against model parameter mismatch for LC-filtered VSCs. Unlike conventional FCS-MPC, the predictive schemes for inductor current and capacitor voltage are treated separately to avoid model inductance dependency. Consequently, the capacitor voltage is easily predicted from a C-filtered model instead of a typical LC-filtered model used in conventional FCS-MPC. And, the capacitance remains as only model parameter that affects the control performance. To solve model capacitance mismatch, the least-squares algorithm is utilized to find the accurate model capacitance parameter. As a result, the high performance is achieved with the proposed method without the degradation from mismatched model parameters.
Fourthly, this thesis proposes a new predictive model and an improved FCS-MPC strategy for LC-filtered VSCs. Specifically, a new predictive model reduces the number of current sensors from 6 to 3, and the model parameters are precisely identified by means of Ohm's law and least-squares principle. Even though 3 current sensors are reduced in the proposed predictive model, it shows the high prediction accuracy, which is comparable to conventional predictive model and outperforms the previous predictive model obtained from Euler method. As a result, the proposed method becomes more reliable and much cheaper in contrast to conventional ones.
Simulation and experiment were carried out to verify the effectiveness of the proposed strategies.