Abstract
Recently, an efficient multistep direct model predictive control (MPC) scheme for power converters has been proposed. It relies on the Sphere Decoding Algorithm (SDA) to solve the associated long-horizon optimal control problem. Since the SDA evaluates only a small number of candidate solutions to find the optimal one, a significant reduction in the average computational burden can be achieved compared to the basic exhaustive search approach. However, this is only true during steady-state operation. In fact, the SDA still requires a large execution time during transients. This paper shows that if not properly addressed, the dynamic performance of the system may be degraded, which clearly limits its practical application. To mitigate this issue, which particularly arises during transients, an efficient preconditioning approach for the SDA is proposed. This approach ensures that only a small number of candidate solutions are evaluated during both steady-state, and transients. This allows the multistep direct MPC to become a viable control alternative for power converters operating at low semiconductor switching frequencies, e.g., below 450 Hz. The proposal is validated using a grid-connected three-level converter as a case study. Both processor-in-the-loop simulations, and experimental results on a scaled-down 2.24 kVA laboratory setup are presented.
Highlights
Direct model predictive control (MPC) takes advantage of the power converter switching nature by directly considering the combinations of power switch states or output voltage levels as the manipulable system input
It was recently shown that the sphere decoding algorithm (SDA), as in [2], can be adopted to efficiently solve the optimal control problem associated with multistep direct MPC schemes [3]
Despite all the efforts put into reducing the computational burden of SDA, there is still a key problem that may affect the real-time implementation with long prediction horizons: the operation of multistep direct MPC during transients
Summary
Direct model predictive control (MPC) takes advantage of the power converter switching nature by directly considering the combinations of power switch states or output voltage levels as the manipulable system input. Using the basic ESA to solve the associated optimal control problem over a prediction horizon of more than one step (commonly known as multistep MPC) leads to a high computational burden since the total number of input combinations increases exponentially. Despite all the efforts put into reducing the computational burden of SDA, there is still a key problem that may affect the real-time implementation with long prediction horizons: the operation of multistep direct MPC during transients. The previously discussed idea of obtaining a new initial sphere center to address the computational burden of SDA during transients was originally proposed in the preliminary work [12] Afterwards, this idea has been adopted in [13] for a multistep MPC formulated as in [10], with simulations results on a variable speed drive system. The impact of transient operation on the computational burden is compared with the standard SDA initialization [3] and with the approach in which a limited number of computations in the SDA is allowed [11]
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
