Design choices for the prediction and optimization stage of finite-set model based predictive control

Abstract

Abstract-The interest in applying model-based predictive control (MBPC) for power-electronic converters has grown tremendously in the past years. This is due to the fact that MBPC allows fast and accurate control of multiple controlled variables for hybrid systems such as a power electronic converter and its load. As MBPC is a family of possible controllers rather than one single controller, several design choices are to be made when implementing MBPC. In this paper several conceptual possibilities are considered and compared for two important parts of online Finite-Set MBPC (FS-MBPC) algorithm: the cost function in the optimizations step and the prediction model in the prediction step. These possibilities are studied for two different applications of FS-MBPC for power electronics. The cost function is studied in the application of output current and capacitor voltage control of a 3-level flyingcapacitor inverter. The aspect of the prediction model is studied for the stator flux and torque control of an induction machine with a 2-level inverter. The two different applications illustrate the versatility of FS-MBPC. In the study concerning the cost function firstly the comparison is made between quadratic and absolute value terms in the cost function. Comparable results are obtained, but a lower resource usage is obtained for the absolute value cost function. Secondly a capacitor voltage tracking control is compared to a control where the capacitor voltage may deviate without cost from the reference up to a certain voltage. The relaxed cost function results in better performance. For the prediction model both a classical, parametric machine model and a back propagation artificial neural network are applied. Both are shown to be capable of a good control quality, the neural network version is much more versatile but has a higher computational burden. However, the number of neurons in the hidden layer should be suffciently high. All studied aspects were verified with experimental results and these validate the simulation results. Even more important is the fact that these experiments prove the feasibility of implementing online finite-set MBPC in an FPGA for both applications.

Field Programmable Gate Array Applications—A Scientometric Review

“…For digital control we found that the most important control techniques are implemented in FPGAs, like fuzzy control [1499][1500][1501][1502], PID [69,1503,1504], model predictive control [73,76,1505], adaptive control [80,81,1506], sensorless control [1507][1508][1509] and distributed control [96,99,101]. These digital control techniques were implemented in FPGAs due the high speed response that a hardware digital control FPGA-based implementation can achieve and its reconfiguration capability, allowing control application to high speed system such as induction motors [1507,[1510][1511][1512], torque control [75,1513,1514], Brushless DC (Direct Current) motors [1515][1516][1517], current control [1518,1519], among others [65,[1520][1521][1522][1523].…”

Three Vectors Model Predictive Torque Control Without Weighting Factor Based on Electromagnetic Torque Feedback Compensation

“…The standard PI-MPTC method uses torque and stator flux to form a cost function by using weighting factors to determine the optimal voltage vector for the next sampling time [21,22]. The weighting factor is affected by the system parameters [23], so the weighting factor setting is generally more complicated. In addition, the weighting factors have a great impact on the performance of the controller, because they balance the torque and stator flux effects on the system.…”

A Simplified Finite-State Predictive Direct Torque Control for Induction Motor Drive

“…The weighting factors are used to tune the importance or cost of a particular target in relation to the other control targets. Different strategies using online and offline search procedures have been proposed in [20], [21], which are strongly dependent on system parameters and require a comprehensive mathematical analysis. Moreover, when the desired control objectives are more than two, trial and error methods are used running computer simulations, which are extremely time consuming [22]- [24].…”