An Optimized Inverse Neural Network Control Augmented with Feedback PI Controller for Time-Varying Systems

Linear proportional-integral-derivative (PID) controllers often fail to provide adequate control performance for nonlinear and time-varying systems due to their constant parametric nature. To address the challenges posed by nonlinear and time-varying systems in their regulation, intelligent controllers using neural networks (NN) and fuzzy logic have been vastly investigated. Out of these techniques, NN-based controllers are more popular for the design of indirect adaptive control schemes owing to their immense learning capability. Yet, these NN controllers may fail to perform well if their hyperparameters, typically learning rate and initial weights, are not designed properly. The control performance may worsen if these NNs fail to mimic or learn the system's dynamics. To address this issue, the current work explores the possibility of employing bio-inspired optimization to tune the essential hyperparameters of NN, along with the inclusion of a PI controller in the loop to compensate for the unidentified dynamics. Through the extensive simulation studies carried out in this work, it was found that the learning capability of the INN assisted by a feedback PI controller, i.e. (INN+PI) controller, could achieve superior performance compared to INN only. This paper also compares the ability of the designed INN+PI controller with only the PI controller in terms of set point tracking, the controller’s output on the basis of significant decrement in MSE and IAE, and elucidates the simulation results.