Adaptive Fault-Tolerant Data-Driven Control of Multilevel PEC Inverter via Soft Reinforcement Learning Under Switching-Level Transition

In this paper, an intelligent data driven control framework is proposed for output stabilization of Packed E-Cell (PEC) inverter during transition from nine level to five level. The key idea of proposed controller is to develop the control scheme to the PEC inverter without the need to the model identification and remove complexities associated with fault dynamics. To achieve an adaptive fault-tolerant PEC inverter, a soft actor-critic (SAC) algorithm with deep neural networks is employed to generate a control policy for tuning the controller parameters without the explicit knowledge of the inverter dynamics. By training the actor and critical networks, the suggested tuner mechanism aims to adapt the established controller to varying the inverter conditions, i.e., switching from nine-level (9L) to five-level (5L). The suggested data-driven controller with SAC algorithm contributes to real-time applications, offering a more straightforward control design to model-based controllers, especially in practical scenarios of power electronic systems. The effectiveness of the suggested control framework has been demonstrated through experimental implementation which confirms its ability to achieve the desired control objectives.