Predictive Control of Three-Level Active NPC Converter with Evenly Energy Losses Distribution Daniel Andler ∗ , Marcelo Perez ∗ , Jos´ e Rodr´ ıguez ∗ , Steffen Bernet † ∗ Technical University Federico Santa Mar´ ıa, Valpara´ ıso, Chile. Email: marcelo.perez@usm.cl † Technical University Dresden, Germany. Email: daniel.andler@mailbox.tu-dresden.de Abstract—Active NPC voltage source converters feature a higher degree of controllability than standard NPC convert- ers due to the additional power switches which allows several different connections of the neutral point. In this paper, these switching redundancies are used to minimize and evenly distribute the energy losses. The predictive control scheme proposed is separated in two stages: The first one controls the output reference currents and keeps the dc voltages balanced. The second stage uses the switching redundancies given by the ANPC VSC to distribute the energy losses evenly. The energy balance is performed with a model-based estimation of the switching and conduction energy losses. In order to test the proposed algorithm, a comparison of the energy distribution between the proposed strategy and a NPC converter is presented. I. I NTRODUCTION Neutral Point Clamped voltage source converters (NPC VSC) have started the research and development of multilevel converters in the early 1980s [1] and have gained more and more importance on the market of the medium voltage converters during the last decades [2]. Nowadays this technology is mature and can be found in several industrial applications [3]. This converter has several attractive features like high reliability and availability, however it has also important disadvantages like the dc voltage unbalance and the uneven distribution of semiconductor losses among the switching devices. As in every converter, the reduction of power losses is an essential aspect to increase efficiency, reducing costs and failures. When the energy losses are not well distributed, as in the NPC converter, the junction temperature can be very different among the semiconductors. The highly stressed devices are prone to failures reducing also the overall performance of the cooling system. The difference of energy losses increases with the switching frequency making the problem even worse. The critical operating points are located at the boundaries of the converter’s op- erating area, at maximum modulation depth (m =1.15), and unity power factors (λ = ±1). Nevertheless, if active switches are placed instead of the clamping diodes the loss distribution can be substantially improved. The active NPC voltage source converter (ANPC VSC) features more degrees of freedom and, in contrast to the conventional NPC converter, it has more than one way to clamp the midpoint. Compared to the conventional NPC topology, the total losses in the ANPC VSC can be maintained but better balanced, hence a better thermal balance can be obtained [4] [5]. This paper presents a Model Predictive Control (MPC) applied to a 3 level ANPC VSC. This control method performs a high dynamic current control [6] and also the DC-voltage balance can be controlled. Furthermore, a novel predictive loss balance algorithm is implemented and tested with simulations. II. ENERGY LOSS ESTIMATION The semiconductor losses are approximated by analyt- ical expressions in terms of voltages and currents. The losses are separated in switching and conduction losses. The conduction power losses are described by (1), where V Γ0 and r Γ0 are the forward conduction voltage and the equivalent differential resistance of the device respectively [4]. P cond (I ph )= I ph · (V Γ0 + r Γ0 · I ph ) (1) The conduction energy losses can be calculated in each sampling interval as E sw = ∫ Ts p cond (t)dt = I ph · (V Γ0 + r Γ0 · I ph ) · T s (2) where T s is the sampling time. The switching losses depend on the current and the voltage involved in the switching process as well as the switching frequency. To estimate the switching losses it is possible to detect in each sampling interval if a switching transition occurs and approximate the energy losses by a polynomial function as E loss = K 1 ∆i c ∆V ce + K 2 ∆i c ∆V 2 ce + K 3 ∆i 2 c ∆V ce +K 4 ∆V 2 ce + K 5 ∆i 2 c ∆V 2 ce , (3) where K i ,i =1, 2...5 are obtained from a least-square approximation of the measured data. This approach con- siders all physically reasonable terms of the switched voltage and current. It is possible to neglect terms in order 754 The 2010 International Power Electronics Conference 978-1-4244-5393-1/10/$26.00 ©2010 IEEE