A recent study co-authored by Ercan Atam, Atakan Zeybek, and Şaziye Betül Özateş makes classical PID control smarter through a deep learning–based gain-scheduling approach that brings advanced control performance closer to standard industrial PID structures.
The recent paper co-authored by Ercan Atam, Atakan Zeybek, and Şaziye Betül Özateş introduces a deep learning–based gain-scheduled PID framework that boosts performance without replacing the familiar PID architecture. The approach learns time-varying PID gains from scheduling signals generated by advanced supervisory controllers, such as an LPV controller or a reinforcement learning agent. Validated on two challenging case studies, the method reports up to 71.6% performance improvement over an optimally tuned fixed-gain PID controller, while closely approaching the supervisory controller’s performance.
The work appears in IEEE Transactions on Emerging Topics in Computational Intelligence, an IEEE Computational Intelligence Society journal widely regarded as a leading venue for emerging computational intelligence research. For details, please refer to:
https://ieeexplore.ieee.org/document/11434543
