Males Tomlinson

Model predictive control with a fixed switching frequency for an AC-to-AC converter

This paper discusses the control of an AC-to-AC converter that forms part of an 11 kV electronic voltage regulator. A novel finite-set model predictive control technique is presented to allow fixed frequency switching. A fixed switching period is defined and divided into smaller evaluation steps. The switching behaviour is obtained by using the analogy of PWM with a triangular carrier wave. During the first half of the switching period, the system may only switch from the on-state to the off-state and during the second half, only from the off-state to the on-state. To account for the repetitive nature of the switching waveform, an average receding horizon prediction method is introduced. An anticipated prediction is made by calculating the average value for one switching period ahead, according to the position and state in the current switching period. The method presented good practical results and proved to be effective in maintaining a fixed switching frequency.

A generic approach to implementing finite-set model predictive control with a fixed switching frequency

This paper presents a generic approach to practically implementing a finite-set model predictive control strategy with a fixed switching frequency for high sampling rates. With conventional FS-MPC methods, the equations for describing the predictions of the controlled variables are derived specifically for each given topology. In this paper, a more generic approach is presented by using established state space control theory to develop a method for modelling and implementing an arbitrary topology. Numerical methods and the use of a symbolic computational toolbox is used for the calculation of off-line prediction equations and using a lookup table to reduce on-line computation time. An implementation scheme for an FPGA is discussed and experimental results are provided to confirm a successful design.

Increasing the performance of Finite-Set Model Predictive Control by oversampling

This paper presents an FPGA-based implementation of Finite-Set Model Predictive Control (FS-MPC) which makes use of oversampling in order to achieve a better control result: In general the maximum IGBT switching frequency is limited to half the sampling frequency for all FS-MPC methods. In simulations and practical implementations, however, it can be seen that the real switching frequency is far lower than its theoretical maximum. Especially for lower power applications where a good quality of the output variables is important, FS-MPC shows very high ripples compared to modulation-based approaches. The main reason for this is that conventional control systems have sampling cycle times which are normally in the range of 20 μs to 200 μs. By using FPGAs, the sampling frequency of the control system can be significantly increased, as the control is completely implemented in hardware. Experimental results for current control of a single-phase resistive-inductive load and for output voltage control of a single-phase Uninterruptible Power Supply (UPS) with a Neutral Point Clamped (NPC) inverter demonstrate the power of the proposed method. A voltage balancing algorithm for the DC link capacitors of the two inverter legs is included in the cost function as well as certain weighting factors to limit the IGBT switching frequency.

A Fixed Switching Frequency Scheme for Finite-Control-Set Model Predictive Control—Concept and Algorithm

A fixed switching frequency scheme for finite-control-set model predictive control is presented to achieve an output waveform quality that compares well to that of a pulse-width-modulator-based linear controller, while retaining the benefits of model predictive control. The controller is divided into two parts: the offline calculation and storage of the coefficients of the prediction equations and the online evaluation of the controller. The controller is experimentally evaluated on a five-level flying-capacitor converter.

Predictive torque control of an induction machine fed by a flying capacitor converter

This paper presents a model predictive control strategy for an induction machine fed by a three-level flying capacitor converter. In order to reduce the calculation effort (64 different switching possibilities for one prediction step) in a first step the best one of the 19 voltage vectors is determined. After that the best switching state which delivers this voltage vector is determined regarding the flying capacitor voltage balancing and the switching effort, making use of a hysteresis-based algorithm. The proposed control algorithm is verified by several simulations which clearly verify that an effective control of speed, torque, flux and of the flying capacitor voltages is possible.

Model predictive control of a flying capacitor converter with output LC filter for UPS applications

This paper presents a model predictive control strategy for a three-level flying capacitor converter with output LC filter for UPS applications. In order to reduce the calculation effort (64 different switching possibilities for one prediction step) in a first step the best one of the 19 voltage vectors is determined. After that the best switching state which delivers this voltage vector is determined regarding the flying capacitor voltage balancing and the switching effort, making use of a hysteresis-based algorithm. A disturbance observer is used to estimate the output currents of the LC filter which are normally not measured. The proposed principle can be easily extended to flying capacitor converters with more than three voltage levels.

Series-stacked medium voltage electronic voltage regulator

This paper discusses the development of a series-stacked medium-voltage electronic voltage regulator for the purpose of regulating the voltage on a 22 kV three-phase transmission line. The three-phase regulator consists of three independent 12.75 kV regulator cells, connected in a star configuration. Each regulator cell consists of an autotransformer with one common and two series windings. An AC-to-AC voltage converter is connected to each series winding in order to accurately regulate the amount with which the line voltage is boosted. This allows for a 10% regulation of the line voltage using two AC-to-AC converter modules. With the modular approach, higher voltages can be attained by simply adding more AC-to-AC converter modules to an appropriate transformer. An alternative AC-to-AC converter topology is implemented and appropriate protection mechanisms are discussed. The system is verified using simulation and practical results and the concept proves to be feasible and practical.

Model predictive control of an AC-to-AC converter with input and output LC filter

The use of finite state model predictive control (FS-MPC) is investigated for an AC-to-AC converter with LC output filter for voltage regulation applications. The equivalent input inductance of the source is included in the model of the converter which results in a second equivalent LC filter at the input. This introduces oscillations in the system. FS-MPC is used to add control to the input without the need for complex control loops. Furthermore, evaluating and switching the system at equidistant instants in time, causes the average switching frequency to decrease for certain voltage references. A new prediction method is proposed to maintain the average switching frequency, which improves the THD of the output voltage. Simulation results are provided to verify the performance of the proposed prediction method.

Heuristic variable switching point predictive current control for the three-level neutral point clamped inverter

This paper presents a variable switching point predictive current control (VSP2CC) method for induction machines (IMs) driven by a three-level neutral point clamped (NPC) inverter with a heuristic preselection of the optimal voltage vector. Enumeration-based model predictive control (MPC) methods are very simple, easy to understand and, in general, offer the possibility to control any nonlinear system with arbitrary user-defined terms in the cost function. However, the two most important drawbacks are the increased computational effort which is required and the high ripples on the controlled variables which limit the applicability of these methods. These high ripples result from the fact that in enumeration-based MPC algorithms the actuating variable can only be changed at the beginning of a sampling interval. However, by changing the applied voltage vector within the sampling interval, a voltage vector can be applied for a shorter time than one sample, which results in a reduced ripple. Since this strategy leads to an additional overhead which is crucial especially for multilevel inverters, it is combined with a heuristic preselection of the optimal voltage vector to reduce the calculation effort. Experimental results are provided to verify the proposed strategy. Furthermore, it will be shown experimentally that a conventional enumeration-based MPC method will lead to very low switching frequencies and high current ripples at low machine speeds; this significant drawback can be overcome with the proposed VSP2CC strategy.

Finite-control-set model predictive control with a fixed switching frequency vs. linear control for current control of a single-leg inverter

In this paper a finite-control-set model predictive control (FCS-MPC) scheme for achieving a fixed switching frequency is compared with a linear regulator for current control of a single-leg inverter driving a resistive-inductive load with a back-EMF. This comparison is motivated by the fact that finite-control-set MPC is often regarded as having an inferior output waveform quality compared to PWM due to a variable switching frequency. In this paper, an oversampled FCS-MPC control scheme that achieves a fixed switching frequency through hard constraints is compared to an oversampled and well-tuned linear resonant regulator. It will be shown that by oversampling the predictive controller and forcing a fixed switching frequency scheme, the output waveform quality of FCS-MPC can produce results similar to that of a PWM-based controller. It will further be shown that the typical one-sample delay during reference step changes and the effects of model uncertainties such a DC bus ripple are greatly reduced if the predictive controller is sampled at a much higher rate than the switching frequency.