Search

This paper presents a control scheme based on the predictive control theory and Petri nets (PNs) formalism to control a flying capacitor multilevel converter. The proposed scheme takes advantage of the benefits offered by the finite control set-model predictive control (FCS-MPC) methodology and the versatility of PNs to regulate the capacitor voltages and output current, simultaneously. The inclusion of PNs allows us to simplify the control scheme. A first PN is used to select, from a subset of switching states, the switching state that accomplishes both control objectives. This approach reduces the computational burden if compared with a typical predictive control scheme. Furthermore, a PN-based prediction scheme is used to predict the capacitor voltages. This prediction scheme uses only measurements of output voltage and current, which reduces the number of required sensors. This approach allows us to implement the internal capacitor voltages control independently if a multileg configuration is needed. The scheme is experimentally validated in a five-level flying capacitor multilevel converter. [ABSTRACT FROM PUBLISHER]

In this paper, an improved model predictive control (MPC) of the modular multilevel converter (MMC) with reduced computational burden is proposed. A mathematical model of the MMC system based on the sum and difference of arm voltages are derived. Instead of determining the switching state of individual submodule (SM), the voltage levels of MMC are considered as control options based on the assumption that the SM capacitor voltages are well balanced. The further reduction of calculation effort is realized by using the tolerance band of capacitor voltages. The proposed MPC has a hierarchical structure. The cost function taking into account the ac-side current control, circulating current elimination and arm energy balancing is presented. The optimal voltage level, selected by the cost function, provides the voltage reference for the pulse width modulation modulator. The SM capacitor voltage balancing is done using a separate control loop. The proposed control strategy is investigated using an MMC high-voltage direct current system with 200 SMs in each arm in real-time simulation and hardware-in-the-loop tests. The performance of proposed method is verified by both steady-state and transient-state operations. [ABSTRACT FROM PUBLISHER]

In recent years, modular multilevel converters (MMCs) are very popular in medium/high-voltage motor drive systems and high-voltage direct-current (HVDC) applications. The conventional model predictive control (MPC) strategies for the MMC are not practical due to their substantial calculation requirements, especially under high number of voltage levels. To solve this problem, a fast MPC strategy combined with a submodule (SM)-voltage sorting balancing method is proposed based on the discrete mathematical model derived for the MMC in this paper. By optimizing the implementation of the control objectives and simplifying the rolling optimization, the proposed strategy is simpler to expand to a system with larger number of SMs, with the amount of calculation reduced and the advantages of the conventional MPC algorithm reserved. In addition, the proposed control can minimize the $dv/dt$ of the output voltages. Both steady-state and transient performances are evaluated by a down-scaled three-phase MMC prototype under various experimental conditions, which validates the proposed fast MPC strategy. [ABSTRACT FROM PUBLISHER]

A single-objective predictive control method that deals with four main control objectives applied to a multivariable single-phase three-level neutral-point-clamped converter operating as an active power filter is proposed in this paper. The four control objectives are to self-support the dc-bus voltage under load variations, to compensate the reactive power and the current harmonics, and to balance the dc capacitor voltages by using a predefined combination of the redundant switching states of the converter. The main contribution of the proposed method is that these objectives are accomplished without using weighting factors in the cost function, which eliminates problems such as multiobjective optimization or additional empirical procedures for determination of these factors. As a result, the method is easy to implement and rapidly selects the optimal voltage to improve the dynamic-state performance. Experimental results from a 2-kVA prototype are presented to prove that the method is valid for single-phase compensation. The well-known effect of model parameter errors' issue, which is inherent in predictive control methods, is also tested to confirm that the harmonic distortion in the grid current is below 5% even when the predictive model has a 25% error between actual and theoretically estimated grid impedance values. [ABSTRACT FROM PUBLISHER]