Combining Data-Driven and Event-Driven for Online Learning Predictive Control in Power Converters

The combination of data-driven and event-driven opens up the possibility of alleviating the long-standing research challenges for power converters in classical finite control-set model predictive control, i.e., model parametric uncertainties and unnecessary switching loss. Inspire by this, we will launch a major study on the problem of designing online learning predictive controller. Unlike most prior works in this area, it can be accomplished by an integrated data-driven and event-driven design framework. To be more precise, the design procedures rely on a combination of developing a data-driven model-free adaptive predictive control, introducing an online reinforcement learning technique, and leveraging an event-driven mechanism. Furthermore, we also provide extensions to robust model-free predictive control design based on input–output data and against unknown uncertainties under low switching frequency operation, while avoiding a priori knowledge of model information and weighting factors. Finally, we illustrate our approach and highlight its advantages on a numerical example, and the results presented are promising and motivate further research in this field.

I. Introduction

Conventionally, power electronic converters are controlled by making use of linear control techniques in conjunction with proportional-integral controllers based on pulsewidth modulation (PWM) stage. This allows simple and clear performance quantification of the control scheme. The traditional linear control design may result, however, in inherently limited response characteristics. Hence, it is slowly being substituted by more advanced nonlinear control solutions in control engineering. Recently, model predictive control (MPC) is a well-established and widely used solution to control complex nonlinear dynamical systems [1]. In MPC, an optimization problem is solved at each time step to produce a control action. This approach not only optimizes some predefined performance criteria, but also guarantees an adequate level of performance in the design of the nonlinear controller.