Predictive Control of a Current Source Rectifier with Imposed Sinusoidal Input Currents P. Zavala ∗ , M. Rivera † , S. Kouro ∗ , J. Rodriguez ∗ , B. Wu ‡ , V. Yaramasu ‡ , C. Baier † , J. Mu˜ noz † , J. Espinoza § , P. Melin § ∗ Department of Electronics Engineering, Universidad T´ ecnica Federico Santa Mar´ ıa, Valpara´ ıso, CHILE, Email: jrp@usm.cl † Department of Industrial Technologies, Universidad de Talca, Curico, CHILE, Email: marcoriv@utalca.cl ‡ Department of Electrical and Computer Engineering, Ryerson University, Toronto, Canada Email: bwu@ee.ryerson.ca, vyaramas@ee.ryerson.ca § Department of Electrical Engineering, Universidad de Concepci´ on, Concepci´ on, CHILE Email: Jose.Espinoza@udec.cl, pemelin@udec.cl Abstract—A new predictive control strategy for current source rectifiers which allows an effective control of source and load currents is presented in this paper. This method uses the commutation states of the converter in the subsequent sampling time according to an optimization algorithm given by a cost function and the discrete system model. The two control goals are: (a) regulation of dc-link current according to an arbitrary reference, and (b) a good tracking of the source current to its sinusoidal reference. The feasibility of the proposed method is verified by MATLAB/Simulink software. NOMENCLATURE i s Source current vector [      ]  v s Source voltage vector [      ]  i i Input current vector [      ]  v i Input voltage vector [      ]    DC-side voltage   DC-side current   Filter capacitor   Filter inductor   Filter resistor   Load inductor   Load resistor I. I NTRODUCTION Current source converters (CSC) are commonly used in medium-voltage, high-power drives in the megawatt level such as pumps, fans, compressors, conveyors and ship propulsion [1]. The CSC is traditionally controlled with classic cascaded linear control loops (usually PI controllers), rotating frame coordinate transformations and a modulation stage [2]. The modulation methods used in practice for CSC are the trape- zoidal pulse width modulation (TPWM) [3], off-line calculated pulse patterns with selective harmonic elimination (SHE) [4], and current space vector modulation (SVM) [5]. In order to keep lower switching frequencies, hybrid modulations com- bining TPWM and SHE are also used in practice, depending on the fundamental frequency (TPWM for lower frequencies and SHE for higher frequencies) [2]. Some of the challenges for the modulation stage of the CSC include contending with restrictions on some switching states; overcoming the trade-off of low switching frequency operation (to improve efficiency); and avoiding lower order harmonics to prevent resonance issues with the output filter and load (or input filter and grid for grid-tied CSC). In addition, neither TPWM nor SHE control the amplitude of the fundamental component generated by the CSC. Only the phase angle and fundamental frequency are controlled by the modulation; the rectifier controls the amplitude by adjusting the dc-current amplitude. This leads to lower dynamic performance since the large dc-choke causes slow dc-current regulation. On the other hand, the phase, frequency and amplitude of the fundamental can be controlled and the dc-current can be fixed when using current-SVM; this, however, results in a slightly higher THD (particularly low frequency harmonics), which could affect resonance issues of the converter. Finite control set model predictive control (FCS-MPC) [6], [7] has been demonstrated to be particularly useful for power converter topologies with diverse and complex control chal- lenges and restrictions. The FCS-MPC has been introduced for matrix converters [8], [9], active front ends [10], and two- level and multilevel inverters [11], [12]. The use of FCS- MPC features power factor control and high-quality sinusoidal input currents [8], [9]. The FCS-MPC is inherently suitable to the limited number of switching states (control actions, or control set) of power converters and the discrete nature of digital implementation platforms such as Microprocessors, DSPs and FPGAs. A discrete-time predictive model of the system is used to predict future values of the variables for each possible switching state of the power converter. The predicted values are then used to evaluate the control goals such as: reference tracking, special system requirements (efficiency, harmonics, common-mode voltages, etc), constraints (satu- ration, forbidden switching states, etc.) and compensations (dead time compensation, voltage imbalance compensation, etc.) which are included in a cost function. The switching state that leads to the lowest cost and hence meets control goals best is then generated. Predictions can be done over two or more sample periods to achieve better performance [13], which ＹＷＸＭＱＭＴＷＹＹＭＰＲＲＴＭＸＯＱＳＯＤＳＱＮＰＰ＠ﾩＲＰＱＳ＠ｉｅｅｅ ＵＸＴＲ