Adaptive Control Design for Power Electronics Converters Using Kolmogorov-Arnold Networks

The design of control algorithm parameters has a significant impact on the reference tracking accuracy and efficiency of power electronic converters. Therefore, it is important to relate the control parameter design to the converter's performance metrics and operating conditions in order to achieve optimized performance. In conventional implementations, these parameters are precomputed and remain nearly constant throughout the converter's operating range. This can lead to sub-optimal performance when the metrics are highly dependent on the converter's loading conditions. Moreover, the relationship between control algorithm parameters, operating conditions, and performance metrics can be complex and difficult to model. In this paper, the Kolmogorov-Arnold Network (KAN) is proposed to derive an analytical expression that maps control parameters to converter performance under varying load conditions. The approach is demonstrated through the design of weighting factors in a finite-set model predictive control (FS-MPC) algorithm. Experimental validation confirms that the resulting analytical expression effectively controls the average switching frequency under different loading conditions. The proposed method offers high accuracy with low computational complexity.