Traditional control ideas for power converters are, to most extent, physical model-based, which starts from time-domain or frequency-domain modeling progress. For model construction, the main procedures, first principle-based mathematical deduction and manual experiments, are considerably time-consuming and cost-expensive, which is not generic and restarts facing different systems. After model construction, a particular control scheme is designed according to the trade-off of model property, cost, work requirements, and so on. One of the model-based evergreen tree schemes, transfer function-based proportion integration differentiation (PID) control, occupies industry for a long time. With the fast-developing computing ability of industrial chips, real-time optimization and nonlinear control becomes more and more promising, such as the recently popular model predictive control (MPC) which is a representative of model-based control method and mainly discussed in this thesis. By the constructed math models, future-horizontal behaviors of the power electronics system could be predicted, which is used for the selection of optimal switch state according to work requirements. However, while facing much longer prediction horizon demand or complex topologies, like the popular multi-level modular converters (MMC), even the most advanced field programmable gate arrays (FPGA) or digital signal processors (DSP) meet computing resource limitations and cost problems. To generalize the advantages of MPC in more applications, inspiration comes from artificial intelligence (AI). After the introduction of MPC and fusion examples of AI-assisted power electronics like surrogate models, as the ratio of data-driven or AI components of control loops increases, we show gradient-free hyper-parameters adjustment, imitator concept of developed controllers, and sparse-prior-knowledge reinforcement learning-based model-free control without effort paid for modeling. Like the coupled DNA spiral structure, on the interactive roads of model-based and model-free motivation, the first and rough step towards generic, systematical, physics-informed, and data-driven control design is embarked for the power electronics system in this dissertation.