Adaptive PID Tuning Using Kolmogorov-Arnold Networks

The paper proposes an implementation of Kolmogorov-Arnold networks (KANs) for the purpose of dynamic proportional-integral-derivative (PID) control tuning in first- and second-order linear systems under noisy and time-varying reference conditions. Toy da-tasets, based on instantaneous system error, output and reference trajectory, are used for training the networks and comparing KANs results over a performance of: i) a PID with fixed coefficients, taken from MATLAB’s Simulink PID Autotune; ii) an MLP based neural network (NN), trained on the same datasets; iii) a traditional adaptive PID scheme with gain scheduling; iv) an LMS-based online tuning approach. Results show that KANs out-perform MLPs and LMS even with less optimized datasets under noisy and quick-changing conditions and perform on par with methods, such as gain scheduling, while allowing for more flexibility and easier setup.