You are on page 1of 13

1826

IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 56, NO. 6, JUNE 2009

Model Predictive Control—A Simple and Powerful Method to Control Power Converters
Samir Kouro, Member, IEEE, Patricio Cortés, Member, IEEE, René Vargas, Student Member, IEEE, Ulrich Ammann, Member, IEEE, and José Rodríguez, Senior Member, IEEE

Abstract—This paper presents a detailed description of Finite Control Set Model Predictive Control (FCS-MPC) applied to power converters. Several key aspects related to this methodology are, in depth, presented and compared with traditional power converter control techniques, such as linear controllers with pulsewidth-modulation-based methods. The basic concepts, operating principles, control diagrams, and results are used to provide a comparison between the different control strategies. The analysis is performed on a traditional three-phase voltage source inverter, used as a simple and comprehensive reference frame. However, additional topologies and power systems are addressed to highlight differences, potentialities, and challenges of FCS-MPC. Among the conclusions are the feasibility and great potential of FCS-MPC due to present-day signal-processing capabilities, particularly for power systems with a reduced number of switching states and more complex operating principles, such as matrix converters. In addition, the possibility to address different or additional control objectives easily in a single cost function enables a simple, flexible, and improved performance controller for power-conversion systems. Index Terms—Adjustable-speed drives, current control, model predictive control (MPC), power converters, pulsewidth modulation (PWM).

I. I NTRODUCTION OWER converters and drive control technology have been in continuous development since the second half of the 20th century and have proven to be enabling technologies in practically every application area (energy, communications, medicine, mining, transportation, etc.). In particular, converter control techniques have been a very active research topic in the field of power electronics, covering countless topologies for low-, medium-, and high-power applications [1]–[4]. Classic linear controllers, together with modulation schemes and nonlinear controllers based on hysteresis comparators, have been the most widely analyzed and developed control strategies for power converters [2]. This can be explained by the
Manuscript received May 9, 2008; revised September 30, 2008. First published November 18, 2008; current version published June 3, 2009. This work was supported in part by the Universidad Técnica Federico Santa María, in part by Ryerson University, and in part by the Chilean National Fund of Scientific and Technological Development (FONDECYT) under Grant 1080582. S. Kouro was with the Electronics Engineering Department, Universidad Técnica Federico Santa María, 2390123 Valparaíso, Chile. He is now with the Department of Electrical and Computer Engineering, Ryerson University, Toronto, ON M5B 2K3, Canada (e-mail: samir.kouro@ieee.org). P. Cortés, R. Vargas, and J. Rodríguez are with the Electronics Engineering Department, Universidad Técnica Federico Santa María, 2390123 Valparaíso, Chile (e-mail: patricio.cortes@usm; rene.vargas@usm.cl; jrp@usm.cl). U. Ammann is with the Institute of Power Electronics and Control Engineering, Universität Stuttgart, 70569 Stuttgart, Germany (e-mail: ammann@ilea. uni-stuttgart.de). Digital Object Identifier 10.1109/TIE.2008.2008349

P

limitation imposed initially by the analog implementation of control loops. These control methods were later on adapted to discrete time digital implementations, which are currently widely accepted as the industry standard. On the other hand, digital signal processing has experienced an even more explosive evolution over the last decades, enabling the implementation of emerging and, usually, more complex control techniques. Some of these methods have been applied to power converters, among them fuzzy [5], [6], adaptive [7], [8], sliding-mode [9], [10], and predictive controls. The latter is a wide control concept that can be further subdivided or classified into several categories [11], depending on the operating principle and other characteristics. In general terms, predictive control can be considered as any algorithm that uses a model of the system to predict its future behavior and selects the most appropriate control action based on an optimality criterion. One of the earlier predictive controllers used in power converters is the so-called dead-beat control, which eliminates the classic linear controller by using a predictive model of the system. This model is used to calculate the required reference voltage in order to reach the desired reference value for a certain variable (usually the current). The predicted reference voltage is later generated by the converter via a modulation stage. This scheme has been applied for current control of inverters [12]–[19], rectifiers [20], [21], active filters [22], [23], and uninterruptible power supplies (UPSs) [24], [25]. A different approach is Model Predictive Control (MPC) [26], in which a model of the system is considered in order to predict the future behavior of the variables over a time frame (integer multiple of the sample time). These predictions are evaluated based on a cost function, and then, the sequence that minimizes the cost function is chosen, obtaining, in this way, the future control actions. Only the first value of the sequence is applied, and the algorithm is calculated again every sampling period. MPC has several advantages, such as the easy inclusion of nonlinearities and constraints. This scheme has few applications in power converter control and drives due to the high amount of calculations needed in order to solve the optimization problem online, which is incompatible with the small sampling times used in converter control. One solution in order to reduce the calculation time is to solve the optimization problem offline, as presented in [27], where MPC is implemented as a search tree and the calculation time is reduced, making it possible to use the MPC in drive control. Another solution is the use of Generalized Predictive Control (GPC) [28], where the optimization is solved analytically, obtaining a linear controller. Nevertheless, with GPC, it is very difficult to include system constraints and nonlinearities.

0278-0046/$25.00 © 2009 IEEE

Authorized licensed use limited to: Ryerson University Library. Downloaded on June 5, 2009 at 10:13 from IEEE Xplore. Restrictions apply.

and controlled instantly in t = tk is not realizable in real-time applications. In this paper. the control action leading to the minimum cost (min{gi }. the predicted value xp3 (tk+1 ) is the closest to the reference x∗ (tk+1 ). . n. a decision or cost function fg can be defined. . (a) Ideal theoretical case. in part. they can be evaluated together with the measured value x(tk ). Authorized licensed use limited to: Ryerson University Library.: MODEL PREDICTIVE CONTROL—SIMPLE AND POWERFUL METHOD TO CONTROL POWER CONVERTERS 1827 Another approach for implementing MPC for power converters and drives is to take advantage of the inherent discrete nature of power converters. If needed for highly dynamic systems. S3 is selected and applied in t = tk . the reference can be considered constant over Ts . as shown in Fig. Naturally. this problem can be overcome if a two-step-ahead prediction is considered. n) is selected to control the system. n. . based on a prediction function fp . In addition. xpi (tk+1 )}. and industrial presence. Then. . S2 is selected and applied in t = tk+1 . providing a fair reference frame for the analysis. Now. which means that it has no special features or favorable characteristics that will make one method outperform another. n. FCS-MPC O PERATING P RINCIPLE The power converter or drive control problem can be defined as the determination of an appropriate control action S (t) (usually the gate signals of the converter) that will drive a generic system variable x(t) as close as possible to a desired reference value x∗ (t). a first prediction can be made to obtain the future value x(tk+1 ) (this is the first step prediction). since Ts is sufficiently small compared with the dynamic behavior of the system. . which can be assumed to be equal to the actual value x∗ (tk ). This approach is known as a Finite Control Set MPC (FCS-MPC). Since the control actions or control set is finite in number Si . Following the same criterion. FCS-MPC operating principle. Consider the qualitative behavior of x(t) and its regularly sampled value x(tk ) over a sample period Ts for a system with a finite number of control actions n. . . the ideal theoretical case in which the variables can be measured. and applications are addressed to show the flexibility and potentiality of FCS-MPC. 1(b). . This prediction function is directly derived from the discrete model and parameters of the system. for i = 1. . hence. addressing the latest advances and promising future of this technology. 1(a). To determine which of the control actions is to be selected. . the future reference value x∗ (tk+1 ) can be estimated via appropriate extrapolation methods. and it has been successfully applied to a wide range of power converter and drive applications [29]–[56]. a comparison between FCS-MPC and classic linear controllers is presented. and control actions are performed instantly (ideal case). while x(tk+1 ) is measured to perform the algorithm again. . predicted. from the predicted value xp (tk+1 ). A typical example for Fig. and then. . . there is only one prediction for the first step. 1(a). where measurements.KOURO et al. As shown in the example in Fig. The evaluation of the cost function with the n predictions will lead to n different costs. Special attention is given to conceptual differences in the operating principle. Note that the future reference value is needed x∗ (tk+1 ). a complete sample period Ts is available to perform the algorithm. Naturally. In addition. . 1. II. in which the control action to be applied in the following sample time S (tk+1 ) is determined. given by the applied control action S (tk ) = S3 determined in the previous execution of the algorithm. with i = 1. a measurement x(tk ) is made and the previously computed control action S (tk ) is applied. With this information and the system model. and actuation times added together. 2009 at 10:13 from IEEE Xplore. controller design. fg would be the absolute error between the predictions and the reference gi = |x∗ (tk+1 ) − xpi (tk+1 )|. the sample period Ts has to be greater than the measurement. as shown in Fig. simple operating principle. and implementation. to predict all the possible system transitions xpi (tk+1 ) = fp {x(tk ). for i = 1. Based on the example shown in Fig. 1(b). Assume that on a sample time tk . at t = tk+1 . . a general approach on FCS-MPC of power converters and drives is presented. Nevertheless. Downloaded on June 5. computation. the state with minimum cost is selected and generated. the optimal selected control action S (tk+1 ) is applied. usually dependent on the desired reference value and the predictions gi = fg {x∗ (tk+1 ). this topology is somehow neutral to both control techniques. However. the MPC optimization problem can be simplified and reduced to the prediction of the system behavior only for those possible switching states. leading to one optimal selection S (tk+1 ) (this is the second step prediction). (b) Implementation case. and thus. The reasons are its more generic structure. other converter topologies. The traditional twolevel voltage-source inverter (2L-VSI) has been selected as the main topology for comparison. This paper has. each prediction is used to evaluate a cost function (also known as quality or decision function). computations. In addition. for i = 1. Both predictions are performed during the first sample period. and consequently. a survey nature. the FCS-MPC algorithm is performed for n possible control actions. This method is also known as finite alphabet MPC or simply as predictive control. Since power converters have a finite number of switching states. This way. Si }. since the possible control actions (switching states) are finite. . . Restrictions apply. drives.

VSI power circuit. system constraints. considering a generic system variable x(t). usually pulsewidth modulation (PWM) or SVM [2]. a comparison of the current control of a 2L-VSI is presented in this section. saturations. For simplicity. Since only a discrete model of the system is necessary. . which can connect the corresponding load terminals to the positive (P ) or negative (N ) bars of the converter by controlling the two power switches of a leg with Sx = 1 and Sx = 0 (x = a. For the PI controller design. a linear proportional–integral (PI) controller followed by a modulation stage. evaluating the different switching states. FCS-MPC generic control diagram. A simplified control block diagram and the corresponding algorithm for the real-time implementation of FCS-MPC are shown in Figs. the converter is Authorized licensed use limited to: Ryerson University Library. 2 and 3. 6. 4. For this comparison. then. c). multiple variables. 2009 at 10:13 from IEEE Xplore. b. NO.1828 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS. basically. The converter consists of three phase legs. while S (tk+1 ) = S2 is selected from the n predictions for the second step. respectively. . VOL. method feasible to implement with present-day available microprocessing resources. Moreover. as shown in Fig. . perturbations. 2. If a voltage space vector is defined as vs = 2 Vdc (Sa + aSb + a2 Sc ) 3 (1) Fig. a linear resistive inductive RL load is considered. there are 23 = 8 different switching states. 4. B. Power Converter Overview The power circuit of a 2L-VSI is shown in Fig. eight voltage space vectors can be generated by the converter (v0 . on the contrary. C OMPARISON OF C LASSIC L INEAR PI AND FCS-MPC To highlight the conceptual differences of the FCS-MPC approach with traditional control methods. Since there are three phases and each phase features two control actions. 3. The 2L-VSI has been selected since it is one of the most widespread converter topologies in the industry and features a generic structure and operating principle that can be easily extended to other converter topologies. Fig. It is worth mentioning that this control method is not limited to a single variable. v7 ). . . a passive-diode front-end rectifier is assumed to provide the dc-link voltage Vdc . the fact that power converters have a reduced and limited number of switching states (or control set) makes this where a = ej 2π/3 . rather than approximated linear models together with control system design theory and modulation algorithms. This is the basis of the great flexibility and control potential that can be achieved with FCS-MPC. FCS-MPC generic algorithm. III. every characteristic that can be mathematically modeled and measured can be included in the predictive model and cost function. 5. Restrictions apply. 56. Linear PI Current Control With Space Vector Modulation (SVM) of a 2L-VSI The traditional current control for a VSI is composed of two stages. and. A. JUNE 2009 Fig. a simpler and more direct design and implementation of the controller can be achieved. Downloaded on June 5.

Authorized licensed use limited to: Ryerson University Library. and kept equal by v0 . while for iβ . . . The corresponding simplified block diagram of the current FCS-MPC is shown in Table I. As expected. VSI voltage space vectors. Although it cannot directly be observed in this illustrative example. one for the iα and one for the iβ component of is or id and iq components if a rotating frame for is is considered (depending on the application). the average switching frequency is variable. which can be clearly appreciated as the optimal choice in Fig. two variables are controlled with one cost function (6). Finally. and a transfer function is obtained from the RL load model vs = R i s + L di s dt =⇒ L (6) G(s) = is 1 = vs Ls + R (2) where L is the Laplace transform. the load is linear. which leads to Lis (tk ) + Ts vi (tk+1 ) RTs + L 2 vi (tk+1 ) = Vdc Si · [1 a a2 ]T 3 isp {Si } = (4) (5) Fig. In addition. note that ips = iαp + jiβp . Other converters. motor drives. the zero vectors v0 and v7 do not change iα at all. Note that for the chosen example. the control action S (t) represents the switching states (Si . respectively. the step in the α component produces a coupling effect in the β component for the PI with the SVM control algorithm. with i = 0. Fig. while v3 and v5 produce a reduction. and cost function dependence. In addition. Simulation results of the dynamic behavior of the load current shown in the table are obtained in both cases under the same operating conditions. iαp {v1 } is above all the other current predictions. the cost function that needs to be minimized is ∗ gi (tk ) = |i∗ α − iαp {Si }| + iβ − iβp {Si } . Restrictions apply. and. leading to robustness issues. . . The output of the controller is the voltage vector that is necessary to correct the current trajectory. 7. switching states. reduced by v5 and v6 . the chosen vectors are v2 and v3 . in particular. A simplified control diagram of this method is shown in Table I. These results were obtained using Matlab/Simulink. are highly nonlinear systems which complicate the obtention of an approximated linearized transfer function for the controller design. The PI controller transfer function is C (s) = Kp 1 + 1 Ti s (3) where Kp and Ti are the proportional and integral parameters. Two PI controllers are necessary. However. . for iα . . This ∗ is then modulated by the converter reference voltage vector vs with PWM or SVM [2]. Downloaded on June 5. hence.035 s is performed only on the α component of the current from 13 to 2. respectively. hence introducing higher losses and demanding more digital processing power. 2009 at 10:13 from IEEE Xplore.or SVM-based modulation stage imposes a fixed average switching frequency. neglected. . v1 and v4 produce a higher increase and higher reduction. for example. such as the one shown previously in Fig. as will be discussed later. FCS-MPC does not force a commutation in each sample period. produce an increase of iα . A summary of the most distinctive characteristics of both analyzed methods is given in Table I. The predictive model corresponds to the discrete time model of the load. 6(a). Note that. Moreover. this effect does not appear. the converter has saturation (maximum voltage and current) which is a nonlinear condition generally overcome with antiwindup algorithms. leading to a spread current spectrum. the selected vectors are v2 and v6 . which can be designed via pole placement or root locus procedures [57]. From the space vector waveforms. the implementation of the PI controller in the present-day digital platforms demands discretization of the controller. now. and v7 . A favorable characteristic is the fact that the PWM. with a sample period of Ts = 100 μs. its model and the corresponding control has no further complications. However. with i = 0. while in FCS-MPC.6 A. 6(a) and (b) shows the system behavior in time and space vector representations. In addition. 1(b). The shadowed region corresponds to a twostep current prediction. v1 . with a Euler approximation of the current derivative. Both methods seem to behave comparably under this situation. loads. it can be clearly appreciated that some pairs of vectors produce the same effect on the α component. where is (tk ) is the measured current and vi (tk+1 ) is the voltage vector generated by the switching states Si . which is true for high switching frequencies. Hence.: MODEL PREDICTIVE CONTROL—SIMPLE AND POWERFUL METHOD TO CONTROL POWER CONVERTERS 1829 C. this drawback can be easily compensated.KOURO et al. neglecting the converter assumption held during the design process is valid only if a high-bandwidth modulation stage is implemented. For example. v2 and v6 . respectively. iβ is increased by v2 and v3 . . the overall selection to fulfill both conditions is v2 . To further show how the algorithm works. 5. Nevertheless. Particularly discontinuous SVM has become the standard method for the modulation stage [58]. The variable x(t) would be the current is . Current FCS-MPC of a 2L-VSI The FCS-MPC current control problem can be easily derived from the generic operating principle analyzed in Section II. . 6(b). 7) of the converter shown in Fig. 5. this explains why in the time representation. v4 . In a similar way. A step change at t = 0. Based on the predictions shown in Fig. hence.

the mean value of the variable is closer to the reference. FCS-MPC. In essence. This can be used to control variables closer to the reference and reduce ripple amplitude. resulting in more accurate reference tracking. on the other hand. Moreover. Considering fast sample Authorized licensed use limited to: Ryerson University Library. Restrictions apply. VOL. which is not limited to the magnitude of the error of the controlled variable. 6. (b) Space vector representation of is = iα + jiβ control. it will also introduce higher switching frequencies. 6. In fact. 2009 at 10:13 from IEEE Xplore. 56. the bandwidth and robustness of the controller will depend on how dominant the nonlinearities of the system are. as analyzed in the example of Section III. Fig. uses a full prediction model of the system. Downloaded on June 5. producing a higher penalization of bigger errors compared with smaller ones. . this error measure will produce a faster controller for the specific variable in the cost function. however. JUNE 2009 TABLE I COMPARISON BETWEEN CLASSICAL LINEAR CONTROLLERS (INCLUDING MODULATION) AND FCS-MPC An in-depth comparison and assessment on this topic is presented in [59].1830 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS. and the feedback is included in the cost function. C OST F UNCTION F LEXIBILITY AND P OTENTIAL Traditional linear controllers with modulation are limited due to their classic feedback structure. FCS-MPC current control operating principle. hence. Cost function (9) takes into account the whole prediction during Ts . (a) Time representation of iα and iβ control. The difference between (7) and (8) is that the latter produces an overproportional cost (in powers of two). the errors can be measured with functions such as g = |x∗ − xp | g = [x − xp ] 1 g= Ts TS ∗ 2 (7) (8) (9) [x∗ (t) − xp (t)] dt. not only the final value at tk+1 . NO. IV.

respectively. 8. These alternatives are reflected. A strategy to reduce and control common-mode voltages from switched-mode power converters using FCS-MPC has been introduced in [55]. particularly in highpower applications. Hence.: MODEL PREDICTIVE CONTROL—SIMPLE AND POWERFUL METHOD TO CONTROL POWER CONVERTERS 1831 Fig. the method will associate a cost to switching states that produce more common-mode voltages. The prediction of the common-mode voltage vcmp is obtained by using the output voltages per phase. Hence. cost function (11). the reduction of commutations by predicting the number of switchings involved in using each switching state. 7 shows an experimental comparison between the classic controller (including PWM-based modulation) and the FCS-MPC for a current control of a neutral-point-clamped (NPC) multilevel converter. which each switching state under evaluation would produce vcmp = vaN p + vbN p + vcN p . 3 (14) g =g + λ · j =1 ¯∗ − 1 g =g + λ · f sw Δt Δt T s −1 C (tk−i ) i=0 (12) where λ is a weighting factor. In (10). In FCS-MPC. hence. is the efficiency of the power converter. to highlight the behavior before and after the compensation. the switching losses proportional to Δic · Δvce of each semiconductor are considered directly in the cost function. a reduction in the average switching frequency. This result has been obtained with a Dspace DS1103 control board with Matlab/Simulink installed on a host standard personal computer. Downloaded on June 5. resulting in g = g + λ · |vcmp |. since they are produced by the commutation of the power devices. It can be clearly appreciated that both algorithms have very similar performances. B. Switching Frequency and Efficiency An important issue in power conversion. a reduction is introduced in the carrier frequency of the PWM method. due to the amount of power loss in long-term operations and cooling system costs.035 s. Restrictions apply.03 s. In the second case. namely. periods. Fig. A. they are a direct result of the control of a modulation method. The third alternative (12) imposes an average switching frequency given ¯∗ = C ∗ /Δt. Consequently. This quality function will select those switching states that maintain the average switching frequency as close as possible to the desired value. respectively.035 s. (13) (10) Δicp (j ) · Δvcep (j ) (11) This way. in the following three additional terms in the quality functions: g = g + λ · Cp n personal computer. 13) with switching frequency reduction at t = 0. as is usual in power converters and drives control. The working principle is to add an additional term with the prediction of the common-mode voltage vcmp to the quality function g . This new component in the cost function will avoid those switching states that would produce more commutations. for simplicity and computational effort reduction. Cp is the number of the semiconductors’ commutations involved when changing from the present to the future switching state [50]. 7. switching-state transitions that imply fewer switching losses will be then preferred [56]. for both methods. where C ∗ is the number of desired as a reference f sw commutations during Δt. switching losses can be controlled by adding another term to the cost function g . affecting its lifetime and producing deterioration. The sample period used to implement this strategy is Ts = 8 μs. The sample period used to implement this strategy is Ts = 100 μs. the prediction of the switching losses caused by selecting each switching state. This result has been obtained with a Dspace DS1104 control board with Matlab/Simulink installed on a host standard Experimental results showing the FCS-MPC mitigation of the common-mode voltages in a matrix converter is shown in Fig. The corresponding weighting factor λ is set at a different value from zero at t = 0.KOURO et al. Common-Mode Voltage Common-mode voltages produced by power converters feeding electric machines cause overvoltage stress to the winding insulation. and the weight factor λ is set at a value different from zero. the three cost measures do not have noticeable differences in their performances. The reference f sw the sum of the previous number of commutations performed during Δt (including the present prediction of the switchingstate transitions under evaluation in the cost function). 2009 at 10:13 from IEEE Xplore. Experimental current control of an NPC converter (power circuit in Fig. . Switching losses are particularly relevant. an arbitrary time frame multiple of ¯∗ is subtracted with the sample period Ts . Authorized licensed use limited to: Ryerson University Library. controlling the converter’s switching frequency. achieving. considering a converter with n semiconductors. and the assurance of a number of commutations in a certain period of time to obtain an imposed switching frequency. the terms Δicp (j ) and Δvcep (j ) are the predicted changes in the collector’s current and the collector–emitter voltage of the semiconductor j . Three approaches have been analyzed. (Top) PI with PWM and (bottom) FCS-MPC. (7) will fit most of the applications. At t = 0.

this can lead to coupling effects or Authorized licensed use limited to: Ryerson University Library. respectively. V. the switching-state transitions producing harmonic components in those frequencies will be selected. Q∗ = 0. voltage. while the voltage error can be 10 V).). one switching state can be the optimal selection for two or more sample periods. For further details. This result has been obtained with a TMS320F2812 DSP. D. vibrations. reactive power. calculated based on predictions of the input current and voltage by isp } = vβp · iαp − vαp · iβp Qp = Im{vsp · ¯ (17) where vsp and isp are the input voltage and current space vector predictions. For example. 9(b). switching losses. It can be said that the filter somehow “hides” information in the cost function for those frequencies for which it is designed. As a consequence. FCS-MPC With Spectrum Shaping Since regular FCS-MPC chooses the optimal switching state that minimizes a cost function.g. considering the prediction to be equal to the last measured value of the variable. (Bottom) Common-mode voltage. C. the objective of controlling the reactive input power Q can be achieved by simply penalizing switching states that produce predictions of Q distant from the reference value Q∗ . (Top) Load current. which can cause resonances. i. where Fz is a discrete time filter. Thus.. Downloaded on June 5. an operation with unity power factor is desired. This drawback can be overcome in FCS-MPC by simply introducing frequency information into the cost function. 6. and [54]. This can be simply achieved by including in the original quality function g an additional term g = g + λ · |Q∗ − Qp | (16) where Qp is the predicted reactive input power.03 s. vsp ≈ vs . the method can be simplified. (a) Notch filter on 4 kHz. Experimental FCS-MPC of a 2L-VSI with spectrum shaping. flux. 8. the filter can be designed in such a way that the desired harmonic content can be imposed in the system variables. it could be particularly interesting to produce fixed switching frequencies for which a narrowband stop filter (Notch) can be used [38]. mainly because they are expressed in different units (e. FCS-MPC C HALLENGES As mentioned before. as a consequence. Reactive Input Power Control With the FCS-MPC approach.. enabling the spectrum shaping. Fig. NO. with the frequency weighted cost function g = Fz {x − xp } ∗ (15) Fig. the maximum current error can be 1 A. . depending on the application. Experimental current control of a matrix converter (power circuit in Fig. 2009 at 10:13 from IEEE Xplore. Generally. Restrictions apply. 17) using FCS-MPC. Since line voltages are low-frequency signals.1832 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS. with the common-mode voltage compensation activated at t = 0. torque. and acoustic noise. JUNE 2009 Fig. They can be even of different physical natures (current. 9(a) shows a filter designed to eliminate all the information around 4 kHz in the cost function. For power converters. The sample period used to implement this strategy is Ts = 30 μs. Since some variables have completely different values than others. no commutations are forced every sample period. Note that ¯ isp is the complex conjugate of vector isp . etc. In fact. please refer to [29]. [34]–[36]. This leads to a variable switching frequency. 9. it can also be of interest to reduce the harmonic content for electromagnetic interference and electromagnetic compatibility purposes. since there is no or little cost associated to them. In this case. the additional term is reduced to |Qp |. VOL.e. different variables can be included in a single cost function. Instead of fixing the switching frequency. hence. consider a general variable x(t). other filter structures can be used in the cost function [40]. Hence. This method produces. (b) Experimental results for load current and its spectrum. Using the approach of (15) in the current control cost function (6) for a 2L-VSI leads to the imposed current spectrum experimental results shown in Fig. the control of the input power factor. 56.

A PPLICATIONS OF FCS-MPC A. The active power reference for the AFE (P ∗ ) is computed by ∗ ) and the power adding the power needed by the load (Pload ∗ . Downloaded on June 5. . without current control loops. [61]. it has been shown that the performance improvement from using one to two step predictions is important but is negligible when comparing two. In [35] and [36]. usually. (18) into three parts. There is no analytical or numerical solution proposed yet to obtain an optimal solution to the problem for FCS-MPC. and is a possible field of research and development. Another unexplored issue of FCS-MPC of power converters is the effective bandwidth achieved by the controller. to changes in the relative importance of one variable over the other in the cost function. the coefficients of the cost function are determined by empirical procedure. Note that in both cases. 11. the first part shows ˜ is half of the the current obtained when the model resistance R real resistance R.: MODEL PREDICTIVE CONTROL—SIMPLE AND POWERFUL METHOD TO CONTROL POWER CONVERTERS 1833 Fig. showing appropriate system behavior in the presence of parameter variations and measurement noise [32]. more knowledge of the system is considered in the cost function. with a sample period Ts = 100 μs. in Fig. Note how directly the active power reference reacts Authorized licensed use limited to: Ryerson University Library. 12. in the same sense as the hysteresis controllers. the same application is controlled with FCS-MPC. Like in all control theories. This challenge is an open research topic for future contributions to FCS-MPC. This result has been obtained with a Dspace DS1103 control board with Matlab/Simulink installed on a host standard personal computer. The cost function used for this application is g = |Qp | + |P ∗ − Pp | (19) At the present state of the art. Nevertheless. However. controlling directly the active and reactive power of the system. Related theoretical analyses or frequency-response measurements have not been reported. This is also true for FCS-MPC. considering the fact that FCS-MPC is inherently a nonlinear control algorithm. However. Fig. which will impose a maximum number of realizable predictions. despite the huge modeling errors of 50% and 100%. as shown in Fig. The sample period used to implement this strategy is Ts = 50 μs. respectively. a frequency-response analysis based on experiments and measurements could be performed as future developments. Note that each waveform is divided where the reactive power reference is set to zero. Although several publications have addressed this issue. for example. It is expected that with more predictions in advance. (a) Error in the load resistance. In this paper. practical implementations are limited by the computational requirements of the algorithm. FCS-MPC of a 2L-VSI under parameter variation or modeling error (simulation). The experimental behavior of the system for steps in the amplitude of the load current is shown in Fig. 2009 at 10:13 from IEEE Xplore. This is generally implemented with voltage-oriented control or direct power control methods. VI. 10. These simulation results were obtained using Matlab/Simulink. some guidelines are presented to help the weighting factor design process. This is not trivial. there is still not enough comparable information to conclude whether FCS-MPC performs better or worse in this aspect than classical linear-modulation controllers. to date.KOURO et al. A simple way to address this issue is to include a coefficient or weight factor λ for each variable in the cost function g = λx |x∗ − xp | + λy |y ∗ − yp | + · · · + λz |z ∗ − zp |. The same was tested for the load inductance in Fig. In [63]. AFE Rectifiers In regenerative applications. only two-step-ahead predictions have been presented. active front-end (AFE) rectifiers.with three(or more) step-ahead predictions [26]. respectively. are used. to date. The critical part of the control scheme is to synchronize the system with the supply voltage and to control the dc-link voltage. (b) Error in the load inductance. There seems to be an amplitude error in the case of the resistance model error while higher ripple is obtained for the case of the inductance model error. in back-to-back configuration with the inverter. 10(b). the traditional concept and theory of bandwidth available for linear controllers cannot be directly applied to FCS-MPC. Moreover. hence. and the last shows that the resistance of the prediction model is the double of the real one. The prediction necessary to control the dc-link capacitor Pdc of the active power (Pp ) is obtained by the measured supply voltage vs (tk ) and current is (tk ) evaluated together with the switching states in the predictive model of the AFE and the supply interface RL branches. there is no theoretical limit to the number of predictions that can be performed. Note that the inverter part of the back-to-back configuration is controlled separately with the FCS-MPC current control seen in Section III. 10(a) and (b) shows the behavior of a current controller for a 2L-VSI with FCS-MPC under parameter variation or modeling error of the load resistance and inductance. Restrictions apply. the middle shows the model that is correct. measurement noise and parameter uncertainty (or variability under operation) can degrade the control system performance and even affect system stability. 10(a). A theoretical approach or analysis in relation to the robustness of FCS-MPC of power converters and drives has not been reported. the current is properly controlled with negligible differences.

(b) Supply reactive power. The input Authorized licensed use limited to: Ryerson University Library. which is properly followed by the FCS-MPC algorithm. despite the fact that there are no current references or dedicated control loops. NPC converter power circuit. currents do not complete a full period due to the small time frame between steps in the active power. 12. Back-to-back FCS-MPC power control block diagram. Restrictions apply. Fig. The decoupling of the reactive power is successfully achieved even during the step transition. . (a) Supply active power. Note how the input current increases to supply the demanded active power. In this case.1834 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS. 6. Fig. However. similar to the method presented for the 2L-VSI in Section III. 13. 2009 at 10:13 from IEEE Xplore. 56. Experimental FCS-MPC power control of a back-to-back converter with step change in load current. VOL. B. Downloaded on June 5. there are two additional requirements: maintaining the voltage also as a step. JUNE 2009 Fig. 11. The input currents are sinusoidal. 13. NPC Multilevel Converter The NPC converter is widely used in medium-voltage highpower applications. (c) Input currents. The power circuit is shown in Fig. the objective is to control the load current. NO.

KOURO et al. the selection of the voltage vector is made by evaluating the effect of each one on the torque and flux of the machine model [52]–[54]. D. 16. Effect of the voltage balancing mechanism. such as the one shown in Fig. are direct converter topologies which do not feature large energy storage components. The load currents appear highly sinusoidal. Using the rotor speed and considering the seven possible voltage vectors. Fig. FCS-MPC torque and flux control of an induction machine (simulation). This result has been obtained using Matlab/Simulink with a sample period TS = 50 μs. 2009 at 10:13 from IEEE Xplore. since the input and output need to be controlled simultaneously. 14. stator flux. since the complete model can be used for prediction and several variables and requirements can be included in the cost function [52]–[54]. particularly in applications where size and volume are an issue [62]. the reduction in the switching losses was tested. and load currents.: MODEL PREDICTIVE CONTROL—SIMPLE AND POWERFUL METHOD TO CONTROL POWER CONVERTERS 1835 Fig. FCS-MPC becomes very attractive. The control strategy is based on the minimization of a cost function g . 15. these experimental waveforms have been obtained with a Dspace DS1104 control board with Matlab/Simulink installed on a host standard personal computer. 16. A large value of certain λ’s implies greater priority to achieve that objective. The stator and rotor fluxes can be estimated with the measurement of the stator currents. Note that λdc = 0. 15. Restrictions apply. respectively. Downloaded on June 5. Experimental FCS-MPC current control of NPC converter. 7. torque dynamic behavior. as can be clearly appreciated even during dynamic transitions. As with results shown in Fig.4 s are shown in Fig. The sample period used to implement this strategy is Ts = 100 μs. seven predictions of the torque and stator flux are calculated. Torque and Flux Control Torque and flux control is mainly used for high-performance adjustable speed drives. Fig.3 s. with and without the dclink control term in the cost function. the cost function used is a combination of (21) where λψ is a weighting factor that handles the magnitude relation between both terms. Here. Fig. The fact that matrix converters connect directly the mains to the load adds to the complexity of traditional controllers and modulation algorithms. ∗ The torque reference Te is obtained from the outer loop speed controller.68 s. 14 shows the behavior of the control method. losing the dc-link balancing control capability. A simplified control diagram is shown in Fig. and the speed reversal at t = 0. Simulation results for a 2L-VSI-fed induction motor showing the drive startup. then a load torque step at t = 0. Here. 18 shows the application of FCS-MPC to a matrix-converter-fed ac drive. although no current controllers are included in the control algorithm. These represent an alternative to the backto-back converters. while the reference stator flux magnitude |ψs |∗ is constant and defined by the rated machine values. Those objectives are reflected in the following cost function: ∗ g = |i∗ α − iαp | + iβ − iβp + λdc |vc1p − vc2p | + λsw Cp Fig. balance of the dc-link capacitors and reducing power losses by avoiding unnecessary switchings in the power circuit. 17. C. The torque presents a very good dynamic response Authorized licensed use limited to: Ryerson University Library. which is completely decoupled from the stator flux that is kept constant at all times.1 has been changed to zero at t = 0. . (20) where vc1p and vc2p are the predicted voltages of both dc-link capacitors and λdc and λsw are the weighting factors to adjust the dc-link balance and switching losses. Speed step response. In addition. FCS-MPC torque and flux control diagram. In this application. given by ∗ − Te | + λψ ||ψs |∗ − |ψs || g = |Te Fig. As mentioned before. 7 shows the experimental output voltage and current of an NPC obtained with a traditional controller and with FCS-MPC. Matrix Converter Matrix converters.

-C. Experimental FCS-MPC of a matrix-converter-fed ac drive during speed reversal. “Discrete-time current control of voltage-fed three-phase PWM inverters. Aug. S.1836 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS. [12] O. Conf. 206–211. pp. B. Electron. no. 2nd ed. 2. J. Hong. 2000. Lau and M. S. 2. M. 627–633. pp. “Multilevel voltage-source-converter topologies for industrial medium-voltage drives. H. 2008. no. Ind. [4] B. 5–10. emerges as a promising tool for electricpower conversion. However. [6] W.—Elect. Salvatore. The sample period used to implement this strategy is Ts = 8 μs. 8. This control method is simple and allows for the control of different converters and various kinds of variables without the need of additional modulation techniques or internal cascade control loops. no. In Fig.” in Proc.). Control. Electron. (IECON). 3. Yu. 8th Int. and C. “Sliding mode control of an induction motor. 3. Rodríguez. VOL. Instrum. 82. and probably the greatest difference with traditional controllers. Fig. 1995. Kennel and A. (d) Input voltage and current. 2. 18 have been obtained with a Dspace DS1103 control board with Matlab/Simulink installed on a host standard personal computer. 4554–4557. vol. [10] F. 23. Tsang and W. etc. 2004. Conf.” in Proc. Wu. [7] P. 5. current source topologies [51]. H. 2000. 7. Cupertino. K. M.. 1996. pp. 1996. and L. Ind. 2008. 56. 2007. Pontt. Pong. 6–10. Lai. 11. Y. Power Appl. 1194–1214. switching losses. synchronized sinusoidal input currents). “Pulsewidth modulation for power converters. the discrete nature of power converters and microprocessors. Bernet. Linder. and S. advantageously. “DSP based fuzzy controlled power converter operating in both continuous and discontinuous conduction modes. 15–19.” in Proc. Rodriguez. [3] J. with constraints and additional system requirements (such as reactive power. Elect. and (13) to mitigate common-mode voltages at the converter output. Dec. pp. the load current is completely sinusoidal. K. Matrix converter power circuit. . vol. [9] S. flying capacitor controls [49]. P. 21st IEEE Int. IEEE PESC. since it considers. Restrictions apply. Authorized licensed use limited to: Ryerson University Library. 2930–2945. 55. Holtz. (c) Input reactive power. with its great potential and flexibility. [5] C. Nov. 54.” in Proc. 152. Interesting applications can be found in those topologies where the modulation techniques are more complex. 17. J. S. 2. shown earlier. Mar.” Proc. L. Jun. C ONCLUSION A general approach of FCS-MPC for power converters and drive applications has been presented. IEEE. no.” IEEE Trans. (16) to have zero reactive power (hence. pp. vol. 761–766. Wu.. and J. Conf. Bernet. A. 22nd IEEE Int. “Adaptive control of power factor correction converter using nonlinear system identification. such as for matrix converters. (a) Motor speed. One of the major advantages of FCS-MPC. M. Other interesting applications include active filters [60]. is the flexibility to control different variables.. Chan. pp. [11] R. 5th WCICA. Kouro. 260–269. 18. 1434–1439.” in Proc. 1994. Power Electron. 6. the method can be applied without significant changes to any type of power converter or drive system. pp. Again. resulting in sinusoidal input currents which are in phase with respect to the supply voltage. 8. JUNE 2009 VII. Rashid. vol.” IEEE Trans. Electron. and safety or to meet standards and operational limits demanded by the evolving industry applications. Mar. (21) for the torque and flux control. 2007. The drive signals for the power switches are generated directly by the controller. “Indirect sliding mode control of power converters via double integral sliding surface. pp.. Instrum. without current loops or modulators.. Alsina and N. Power Electron. 2009 at 10:13 from IEEE Xplore. voltage unbalance. This is why FCS-MPC. Aug. FCS-MPC is conceptually very simple yet powerful. Ireland. Jul. common-mode voltages. Variable Speed Drives. J. Fig. Control. vol.. Downloaded on June 5. vol. “Neuro-adaptive control of induction motor stator current. This can lead to improved performance. Eng. efficiency. NO. J. Lattanzi. (b) Load current.” IEEE Trans. The reactive power is properly controlled and remains zero at all times. vol. Experimental results shown in Fig. no. 600–611. Galway. no. Ed. “A simple direct-torque fuzzy control of permanent magnet synchronous motor driver. vol. “Predictive control of inverter supplied electrical drives. R EFERENCES [1] Power Electronics Handbook. New York: Academic. the effect of the common-mode voltage component can be appreciated. Ind. Jun. [8] K. and UPSs [48].” Proc. Tse. (IECON). Kukrer.” IEEE Trans. pp. 2005. Ind. 1530–1535.. Electron. Gehlot. pp. Inst. May 6. Ind. Kouro. pp. Tan. [2] J. “Current-source converter and cycloconverter topologies for industrial medium-voltage drives. Pontt. The required high calculation power that was a limitation some decades ago is overcome by today’s standard DSPs and microcontrollers. vol. and S. 2786–2797. Electron.

” in Proc. Papathanassiou. 32nd Annu. Oct.” in Proc. Blaabjerg. Ind. P. Rees. P. 2008. 41. Conf. 49. EPE-PEMC. 37. Linke. pp. 1793–1799. S. 2005. E.: MODEL PREDICTIVE CONTROL—SIMPLE AND POWERFUL METHOD TO CONTROL POWER CONVERTERS 1837 [13] H. pp. [43] D.” IEEE Trans. Jun. Rees. IEEE IAS Annu. [50] R. vol. Rodriguez. pp. Appl. “A predictive control scheme for current source rectifiers. E. Ammann. C. “Predictive direct torque control of an induction machine. J.. N. Power Electron. pp. vol. [21] W. Rees. no. PELINCEC. Poland. pp. Vargas. and J. Lezana. Rodriguez. Vargas. 2096–2102. no. Viarouge. J. Latvia. 54. 2005. pp. Vargas. Ind. J. Poznan. and P. Slimani. [40] D. [14] H. [41] M. Ambro˘ ziè. U. Nov. Rivera.” in Proc. “Analysis and implementation of a real-time predictive current controller for permanent-magnet synchronous servo drives. [38] P. and F. Univ. Kennel.and M. 3. pp. “Behavior of the predictive DTC based matrix converter under unbalanced AC supply. C. and F. Lezana. 495–503. [44] G. J. Leon. Pontt. Buso. 23.. 29. Muller. 17. Woo. Sep. 32nd Annu. 2535–2542. Vargas. Rec. Control Conf. Sep. 2005. Pontt.” in Proc. no. pp. Lehn. 1999. Papers. S. Ind. vol. “Predictive control of a three-phase neutral point clamped inverter. Kennel. Ind. Appl. Pontt. 18. “A control analysis and implementation of high voltage. Ind. Youn. Liu. [48] P. [62] P. pp. S. 2003. no. pp. 1275–1280. no. 1241–1246. IEEE PESC. Nedeljkoviè. 110– 117. J. Cortes.” IEEE Trans. 2697–2705.-M. Jun.KOURO et al. 2001. and U.” in Proc. U. Sep. Authorized licensed use limited to: Ryerson University Library. EPE-PEMC. 2006. P. 40. Correa.” in Proc. 16–19.. “Guidelines for weighting factors design in model predictive control of power converters and drives. Clare. NJ: Prentice-Hall. H. Sep. [60] D. Wheeler. and S. “New time-discrete modulation scheme for matrix converters. García. [31] J. [42] A. R. Electron. Cook. Jun.” IEEE Trans. 1692–1706.. Power Electron. pp. B. pp.. Vargas. Rodríguez. Mexico. pp. Feb. Appl. 206–212../Jul.” IEEE Trans. Nemec. Rodriguez.” IEEE Trans. F. Clare.” in Proc. Aalborg. Nov. D. 1660–1664. J. Slovenia. S. Apr. Electron.” in Conf. vol. Krzeminski. “Matrix converters: A technology review. Weinstein. Feb. [37] J.-F. pp. P. no. vol. [46] H. Malesani. [18] S. Acapulco. 1994. 2006. Cortes. pp. pp.” in Proc. Recife. Cortes. Perantzakis. [56] R. Correa.. 585–593. Roth-Stielow. pp. vol. and P. Quevedo. Cortes. Lezana. Wheeler. K. J. 1–10. “Robust dead-beat current control for PWM rectifier and active filters. 9–14. 1720–1733. pp. and S. no. Model Predictive Control. 276–288. 1997. Cortes and J. Goodwin. Jun. J. pp. and L. 45. 2007. U. 2008.” in Proc. P. Cortes. Oct. N. Ammann. Mattavelli. 2005. Conf. 202–207. Clare. Conf. [47] M. S. Silva.” in Proc. pp. Vargas. [22] S... and P. 526–530. and D. 1062–1068. Dang. 2192–2196. 4. 1364–1369. Q. Silva. Silva. 1. 36th IEEE PESC. “Direct model predictive control—A new direct predictive control strategy for electrical drives. “Predictive control of a three-phase neutral-point-clamped inverter. Denmark. Serra. and P. Oct. 2003. IEEE IAS Annu. 2905–2910. S. pp. U. vol. U. Sep. Ind. IEEE PESC. Ind. 32nd Annu. Electron. P. Espinoza. Jun. “Generalized predictive control (GPC)-ready for use in drive applications?” in Proc. IEEE PESC. Australas. pp. Jun. 2008.-T. E. Fasolo. [26] E. 2007.-G. 1 pt. Jul. [23] J. Springob and J. J. P. pp. 2009. C. 6. Meeting. 317–325. 52. Ind. Amer. New York: Springer-Verlag. Rodríguez. vol. and G. no. Rees. 2005.. May/Jun. Ammann. and P. vol. Salgado. pp. “New modulation strategy for a matrix converter with a very small mains filter. 2005. [52] J. 33rd IEEE PESC. New York. and J. 2005. vol. J. no. D. Xepapas. Power Electron. 18.” in Proc. Electron. Electron.” in Proc. “An improved deadbeat control for UPS using disturbance observers. P. [16] S. no. 503–515. Control Conf. Yang and C. Nedeljkoviè.. 2003. “Predictive current control strategy with imposed load current spectrum. [54] R. Meeting. Ammann. Sep. IEEE 36th PESC. vol. Portoro. J. pp. 2004. 2268–2273. Restrictions apply. 35th IEEE PESC. 2. 699–702. Wheeler. Rees. and M. J. Poland. 52. Cortés. 3. vol. Buso. “Predictive strategy to reduce common-mode voltages on power converters. and D.” IEEE Trans. pp. 5. Rodriguez. and R. [55] R.” in Conf. Mexico.. Eur. Rodriguez. Sep. 495–503. W.” IEEE Tran. IEEE PESC. and J. Clare. 1. J. “Predictive current control of a voltage source inverter.” IEEE Trans. and J. “Control of EMI from switch-mode power supplies via multi-step optimization. Mattavelli. [49] E. no. “Application of discontinuous PWM modulation in active power filters. 2–4.. no. Jul. Rodriguez. Mattavelli. 464–472. Silva. Rodriguez. and M. “Predictive control of a three-phase inverter. “Three-phase parallel active power filter with direct current control.. Correa. [15] L.” IEEE Trans. vol. 3763–3768. [35] J. Rodriguez. Bode. 2002. Conf. “Novel control strategy of an AC/DC/AC converter using power relations. J.” in Proc. Feng. vol. pp. S. Power Electron. and H.” in Proc. 41. 2001. Vargas. Conf. Lezana. P. 934–939. R. “Predictive current control of voltage source inverters.” in Proc.. “Predictive control strategies for ZCS direct converter HV power supply. Bordons. Manias.. 6. Oct. 23–27. Quevedo and G. 18. H. J. Correa. Ammann. 1–7. pp. “High-bandwidth current control for torqueripple compensation in PM synchronous machines..” in Proc. 2004. Abu-Rub. 2009 at 10:13 from IEEE Xplore. IEEE IECON. Cortes. 2005. B.. Amer. Pontt. Xepapas. 1998. E. Linder. Conf. Int. Kennel. P. [24] S. J. 2005. Rodríguez. 9. no. Electron. [61] V. Englewood Cliffs. Toliyat. 2.. [32] J. 15–19. Electron./Dec. Manias. Vargas. Cortés. 2007. vol.-S. Cortes. Rodriguez. Ammann. “An analytical approach to steady-state current control properties of power converters featuring discrete-time switching. R. La Rocca. Graebe. Power Electron. 10–13. 51.” IEEE Trans. 2007. “Multistep optimal analog-to-digital conversion. Brazil.-H. IEEE PESC. [51] P. 2008. vol. IEEE PESC. high frequency direct power converter. pp. [36] J. Acapulco. A. and S. Ind. C. Ammann. and S. vol. P. G. R. pp. Pontt.” Electron. P. 2002. 5. “A predictive current control technique for three-level NPC voltage source inverters. pp. Jan. no. 390–395. A. J. P. vol. Kim. “Direct current control—A new current regulation principle. Camacho and C. 2005. Ind. 52. J.” IEEE Trans. J. and J. Linder and R. Vazquez. G. no. and P. Rec. “Predictive control strategy for ZCS single stage resonant converter. Linder and R. . pp. Feb. Huerta. McGrath. Franquelo.” in Proc. Kouro. Catucci.” IEEE Trans.-H. and B. 2003. Jeong and M./Dec. 2008. Quevedo and G. and J. Wheeler. vol. Rivera. Rodríguez. Dec. Holmes.” in Proc. 4. no. Electron. 2. pp. P. pp. Espinoza. and Y. Mar. “Predictive power control of an AC/DC/AC converter.. Rodríguez. Eur. 32nd Annu. 2. Loh. Power Electron. Riga. 1839–1844. 44. Jun. Mar. Downloaded on June 5. Power Eng. and S. Eur. U. M. “Analysis and implementation of a new PFC digital control method. Wheeler. Pontt.” in Proc. E. “Predictive direct torque control of an induction motor fed by a matrix converter. Silva. Power Electron. [33] J. Appl.. and U. [30] J. “Uninterruptible power supply multiloop employing digital predictive voltage and current regulators. 1607–1615. I. U. Rodríguez. pp. Ammann.” in Proc. 1846–1854.. vol. [59] U.” IEEE Trans. 335–340. “Model predictive control for electrical drives. Jan. Pontt. Catucci. 329– 336. M. Vargas. [39] P. [20] L. 1. 713–721. Salgado. [57] G. Guzinski. “Predictive control of a flying capacitor converter. Henriksen.” IEEE Trans. “An improved robust predictive current regulation algorithm. pp. Pontt. Quevedo. Ammann.. Jun. and U. Apr. L. Muller. 1. “A deadbeat current controller for field oriented induction motor drives. C. Electron. pp. Circuits Syst. [29] S. Nov. 2003. 2005.” in Proc.” IEEE Trans. [63] P. Newman. “Efficient predictive current control technique for multilevel voltage source inverters. R. 3401–3406. 1–3.. Z. Australia. M. Le-Huy. Control System Design. [19] H.” in Proc. 1. J. Sep. IEEE ICIT.” IEEE Trans. 2005. 23.. Pontt. Cortés. “Predictive torque control with input PF correction applied to an induction machine fed by a matrix converter. Warsaw.” in Proc. Appl. and P. R. Goodwin. Goodwin. “A controller architecture for high bandwidth active power filters. 1. J. F. “DSP-based active power filter with predictive current control. Reg. pp. Asiminoaei. Eur. 612–618. 13th EPE-PEMC. P. Jun. [45] G. Rodriguez. Ind. J. EPE Conf. and V. 3. 561–562. no. Sep. Pontt. pp. 1999. “A discrete-time predictive current control for PMSM. G. Ambro˘ ziè. P. Appl. U. vol. J. Appl. Nov. Cortes. Goodwin. [27] A. Silva. 2004. I. Vargas. pp. 5. LA. [17] G. vol. P. [28] R. 54. IEEE IECON. 2006. and C. S. [25] P. Rodríguez. 2008. Ammann.” IEEE Trans. 3. IEEE IECON. J. Gippsland. IEEE PESC. 15–19. Pontt. S. “Three-phase inverter with output LC filter using predictive control for UPS applications. 6. pp. J. 15–19. Ammann.” in Proc. 2007. 772–778. Lee. “Predictive current control of a voltage source inverter. Power Electron. 2007.” IEEE Trans. pp. no. Cortes. S. 3. no. Rodriguez. Nov. 2008. 2004. I. C. 11th Int.” in Proc. no. pp.” in Proc. Moon. [58] L. J. M. Holtz. Rodríguez.” IEEE Trans. vol. R. 2001. and A.. and S. Rees. Mossoba and P. Power Electron. Empringham. Ind. Amman. Fi˘ ser. 12–16. Zhang. New Orleans. and J. J. vol. 15–19. [53] M. 2004.-J. Ammann. Rodriguez. 2006. “Reduction of switching losses and increase in efficiency of power converters using predictive control. 2001. “Cost functionbased predictive control for power converters. C. U. [34] S. Betz.. Feb. P. 35. C. Power Electron. and P. R. Lett. Perantzakis. “Practical issues with predictive current controllers. Jan. F. 613–620.

controlled ac drives for shovels. His main research interests include predictive control. high-power conveyors. degrees in electronics engineering from the Universidad Técnica Federico Santa María (UTFSM). he was the Director of the Electronics Engineering Department. such as cycloconverter-fed synchronous motors for SAG mills.D. 2007. 2009 at 10:13 from IEEE Xplore. and Ph. Since 2009. VOL. His research interests include electric drives. His main research interests are power electronics. Valparaíso. adjustable-speed drives. Erlangen. Chile. He has directed over 40 R&D projects in the field of industrial electronics. He received the IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS Best Paper Award in 2007 from the IEEE Industrial Electronics Society. Universität Stuttgart. Rodríguez is an active Associate Editor of the IEEE Power Electronics and IEEE Industrial Electronics Societies since 2002. Since 2002. René Vargas (S’05) received the Engineer and M. Chile. ON. and predictive control. Valparaíso. in 1978. degree in electrical engineering from the University of Erlangen. Germany. he has been Rector. Chile. in recognition of his ability to organize and direct moral conditions and technical preparation. degree from the Universidad Técnica Federico Santa María (UTFSM). respectively. Downloaded on June 5. and 2008. Modern Rectifiers (2005). During his sabbatical leave in 1996. new converter topologies. Chile. in 1985. where. José Rodríguez (M’81–SM’99) received the Engineer degree in electrical engineering from the Universidad Técnica Federico Santa María (UTFSM). Valparaíso. he was a Visiting Scholar with the Institute of Control and Industrial Electronics. and.D. in 2002. Multilevel Inverters (2002). 56. His main research interests include multilevel inverters. he has been with the Electronics Engineering Department. he has coauthored over 250 journal and conference papers and has contributed one book chapter. degree in the field of discrete-time modulation schemes. he was the Vice Rector of academic affairs.Sc. His research group has been recognized as one of the two centers of excellence in engineering in Chile in the years 2005 and 2006. Authorized licensed use limited to: Ryerson University Library. He was with the Institute of Power Electronics and Electrical Drives. he has been with the Institute of Power Electronics and Control Engineering. Universität Stuttgart. in 2004 and 2008. Valparaíso. UTFSM where he is currently a Research Associate. where he was responsible for the mining division. high-power current sources. Warsaw University of Technology.D. from 2004 to 2005.-Ing. Stuttgart. and automotive power electronics. in 1977. UTFSM.Sc. He received the M. he was with Siemens Corporation. He has extensive consulting experience in the mining industry. He joined the Electronics Engineering Department. matrix converters. and new control techniques applied to power conversion and drives. Ulrich Ammann (M’06) received the Dipl.1838 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS. Dr. Germany. he has been a Professor with the UTFSM. and renewable energy power systems. Chile. NO. and adjustable-speed drives. (with honors) degrees in electronics engineering from the Universidad Técnica Federico Santa María (UTFSM). Canada. degree in electrical engineering from the Universität Stuttgart.-Ing. particularly in the application of large drives. Kouro was distinguished by the President of the Republic as the youngest researcher of the Chilean National Fund of Scientific and Technological Development (FONDECYT) in 2004. and Predictive Control of Power Electronic Drives (2008)]. in 2005. degrees in electronics engineering and the Ph. where he is currently working toward the Ph.D. Patricio Cortés (S’05–M’08) received the Engineer and M. In 2007. Chile. Toronto. since 2005. Since 1977. from 2001 to 2004. Santiago. Prof. Stuttgart. as a Research Assistant in 2004 and became an Associate Researcher in 2008. His main research interests include power converters. JUNE 2009 Samir Kouro (S’04–M’08) was born in Valdivia. He has served as the Guest Editor of the IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS in five opportunities [Special Sections on Matrix Converters (2002). degree in the Electronics Engineering Department. and powerquality issues. 6. during scientific stays in 2006. . Since 2003. Restrictions apply. and the Dr. High-Power Drives (2007). Warsaw. including predictive techniques. Germany. as a Research Assistant. adjustable-speed drives. Poland. He was also the recipient of the “Ismael Valdés Award” from the Institute of Engineers of Chile in 2005. where he is currently working toward the Ph. he has been a Postdoctoral Fellow at Ryerson University. in 2004 and 2008 respectively.Sc.