Hybrid PID-Neural Network Control Strategy for Load Frequency Control in Smart Power Systems

Abstract

Load frequency control remains one of the most pressing operational challenges in modern smart power systems, particularly as intermittent renewable energy sources displace conventional synchronous generators and erode system inertia. Conventional proportional-integral-derivative controllers, long relied upon for frequency regulation, exhibit performance degradation under nonlinear, time-varying conditions that characterise contemporary grid operations. In this paper, a hybrid control strategy is proposed by combining proportional-integral-derivative controller with a feedforward neural network to implement load frequency control for two-area interconnected power system. The proportional-integral-derivative term corrects the steady-state error and the base frequency response; while the neural network provides an adaptive corrective signal that reduces residual transient error and improves frequency recovery under stochastic and time-varying load perturbations within the linearised plant framework. The neural network is then trained offline with the Levenberg-Marquardt backpropagation via simulated area control error trajectories, and it subsequently runs online without any retraining of its weights. Simulation experiments on a two -area thermal power system (TAPS) show that for step load perturbations, stochastic load variations and renewable energy penetration, the proposed strategy reduces the integral squared error by 78% and the integral time absolute error by 77.4% compared to a classical proportional-integral-derivative controller. Peak frequency deviation is curtailed from 0.021 Hz to 0.0081 Hz and settling time dropped significantly, from over 9.3 seconds on average to only 4.2 seconds Closed-loop stability remains consistent under robustness evaluation under parametric uncertainties of up to 20% Our results prove that the proposed hybrid approach is a practical and computationally efficient strategy for frequency regulation in renewable-integrated smart girds.