Predictive Control in Matrix Converters Marie Curie ECON2 Summer School University of Nottingham, England July 9-11, 2008 Marco Esteban Rivera Abarca Universidad Técnica Federico Santa María Department of Electronics Engineering Valparaíso, Chile.

-2- Marie Curie ECON 2 Summer School, July Introduction 2.Power Circuit and Basic Concepts 3.Control Strategy: Predictive Direct Torque Control (PDTC) 4.Models used to Obtain Predictions 4.1 Matrix Converter 4.2 Induction Machine 4.3 Input Filter 5.Results 6.Improvements 7.Future Work 8.Conclusion Outline Introduction Power Circuit Control Strategy: PDTC Models Results Improvements Future Work Conclusion Ready for the next

-3- Marie Curie ECON 2 Summer School, July 2008 Introduction Matrix Converter is a single-stage power converter and represents an alternative to back-to-back converters in cases where size and the absence of large capacitors or inductances are relevant issues. Model Predictive Control has been used in applications related to power converters, generally with modulation techniques. In this work is presented a control strategy to control input PF, torque and flux on an IM, based on Predictive Control: without linear controllers without hysteresis without modulators (PWM) Ready for the next Outline Introduction Power Circuit Control Strategy: PDTC Models Results Improvements Future Work Conclusion

-4- Marie Curie ECON 2 Summer School, July 2008 Power Circuit and Basic Concepts Matrix Converter Input Filter Load : Induction Machine Ready for the next Bidirectional Switch Outline Introduction Power Circuit Control Strategy: PDTC Models Results Improvements Future Work Conclusion

-5- Marie Curie ECON 2 Summer School, July 2008 Control Strategy: Predictive Direct Torque Control (PDTC) An intuitive approach Ready for the next Power Supply Matrix Converter Induction Machine Switching State (k) Digital Controller Time (k+1) Switching State (k+1) Time (k+1) Outline Introduction Power Circuit Control Strategy: PDTC Models Results Improvements Future Work Conclusion

-6- Marie Curie ECON 2 Summer School, July 2008 Block diagram of the Predictive strategy Measurements are acquired. 1 Predictions of Flux and Torque are computed for each of the 27 switching states by means of a model. 2 The torque reference is generated by a PI controller. 3 The reactive input power is also predicted for each state. 2 A quality function g is evaluated for each prediction. 4 Ready for the next Control Strategy: Predictive Direct Torque Control (PDTC) Outline Introduction Power Circuit Control Strategy: PDTC Models Results Improvements Future Work Conclusion The switching state that minimizes g is selected to be applied during the next sampling interval. 5

-7- Marie Curie ECON 2 Summer School, July 2008 Control Strategy: Predictive Torque Control (PTC) Quality Function g : The Evaluation Criterion Must reflect the desired objectives, in order to determine the best state. Considering both objectives (adding): Objectives related to the load: Minimize the error on the electric torque and flux magnitude. Objectives related to input variables: Controllable input Power Factor (most cases unity PF, no reactive power). The versatility of the method allows to include other objectives simply by adding terms to the quality function. Ready for the next Outline Introduction Power Circuit Control Strategy: PDTC Models Results Improvements Future Work Conclusion

-8- Marie Curie ECON 2 Summer School, July 2008 Models Input Filter Matrix Converter Induction Machine Ready for the next Outline Introduction Power Circuit Control Strategy: PDTC Models Results Improvements Future Work Conclusion

-9- Marie Curie ECON 2 Summer School, July 2008 Models Input Filter Matrix Converter Induction Machine Ready for the next Outline Introduction Power Circuit Control Strategy: PDTC Models Results Improvements Future Work Conclusion

-10- Marie Curie ECON 2 Summer School, July 2008 Simulation Results Parameters B=0 No PF or reactive input power control. Quality Function: B=0 No PF or reactive input power control. High distortion in the input current. THD=68.5% Low ripple and fast dynamic response. Sinusoidal output currents, smooth freq. transition. B=146·10 -6 Controlled PF/reactive input power. Low distortion in the input current. THD=4.7%. PF=1. Practically identical output variables B=146·10 -6 Controlled PF or reactive input power. Ready for the next Outline Introduction Power Circuit Control Strategy: PDTC Models Results Improvements Future Work Conclusion

-11- Marie Curie ECON 2 Summer School, July 2008 Improvements Ready for the next How to reduce the processing time? What happens when I don´t have a correct model of the load? Perfect Prediction? Outline Introduction Power Circuit Control Strategy: PDTC Models Results Improvements Future Work Conclusion

-12- Marie Curie ECON 2 Summer School, July 2008 Improvements Improve the code of the algorithm. Ready for the next Other techniques of predictive control like DMC and GPC. High computational cost. Improve the predictive models: - Kalman Filter to flux estimator. - Load parameters estimation using LS and RLS methods. - Correction of Input Currents: Study Input Filter. - Correction of Input Currents: AC Supply Unbalanced. Outline Introduction Power Circuit Control Strategy: PDTC Models Results Improvements Future Work Conclusion

-13- Marie Curie ECON 2 Summer School, July 2008 Kalman Filter in flux estimator Stator Flux [Wb] Rotor Flux [Wb] Outline Introduction Power Circuit Control Strategy: PDTC Models Results Improvements Future Work Conclusion Improvements

-14- Marie Curie ECON 2 Summer School, July 2008 Load parameters estimation using LS and RLS methods Load Parameters Real & Estimated Signal Outline Introduction Power Circuit Control Strategy: PDTC Models Results Improvements Future Work Conclusion Improvements

-15- Marie Curie ECON 2 Summer School, July 2008 Correction of Input Currents: Study Input Filter Outline Introduction Power Circuit Control Strategy: PDTC Models Results Improvements Future Work Conclusion Improvements

-16- Marie Curie ECON 2 Summer School, July 2008 Correction of Input Currents in presence of unbalances Minimization negative sequence of input currents. To generate a reference of input currents. Outline Introduction Power Circuit Control Strategy: PDTC Models Results Improvements Future Work Conclusion Improvements

-17- Marie Curie ECON 2 Summer School, July 2008 Future Work Experimental Implementation Predictive DTC. Experimental Implementation of Improvements studied in UCC. Publications of results respective. Experimental Strategy using an Indirect Matrix Converter. Outline Introduction Power Circuit Control Strategy: PDTC Models Results Improvements Future Work Conclusion

-18- Marie Curie ECON 2 Summer School, July 2008 Conclusion Simple and effective control for a matrix converter based induction motor drive. Controls together input and output variables (PF and motor). Discrete-time switching - semiconductors switch only at predefined and equidistant instants (No PWM). This discrete approach match with the discrete nature of the matrix converters switching states and digital control platforms. The versatility of the method allows to include additional objectives. The topic is still open for research. Outline Introduction Power Circuit Control Strategy: PDTC Models Results Improvements Future Work Conclusion

I appreciate your attention Marco Rivera

-20- Marie Curie ECON 2 Summer School, July 2008 Selection of the Weighting Factors Outline Introduction Power Circuit Control Strategy: PTC Models Results Benefits? Conclusion Extra Ready for the next PF Control

-21- Marie Curie ECON 2 Summer School, July 2008 Selection of the Weighting Factors Outline Introduction Power Circuit Control Strategy: PTC Models Results Benefits? Conclusion Extra A Reactive Power Torque Error A Flux Error A High value Low value Always B=1 1 Ready for the next