Intelligent HVDC Control Strategies for Improving AC Network Stability Using Neural Network Models

Abstract

The stability of modern alternating current (AC) power systems is increasingly challenged by the integration of high-voltage direct current (HVDC) links, distributed generation, and renewable energy sources. Traditional HVDC control schemes, while effective under nominal operating conditions, are often constrained by linearized assumptions and fixed parameters, rendering them inadequate for mitigating complex dynamic disturbances such as sudden load changes, line faults, and generator tripping. This paper presents a comprehensive investigation into intelligent HVDC control strategies that leverage artificial neural network (ANN) models to enhance AC network stability. By employing ANN-based predictive controllers in conjunction with conventional HVDC supplementary controls, the proposed approach enables real-time adaptive modulation of active and reactive power flows, thereby improving voltage profiles, damping electromechanical oscillations, and maintaining system frequency within stringent limits. A detailed multi-machine AC system model is developed, incorporating synchronous generator dynamics, excitation and governor systems, and bipolar HVDC links, allowing high-fidelity simulation of transient and small-signal disturbances. The ANN controller is trained using historical system responses to map complex non-linear interactions to optimal control actions, facilitating robust and adaptive performance across varying scenarios. Simulation results demonstrate substantial improvements in key stability metrics, including a reduction in maximum voltage deviations by over 50%, enhanced rotor angle damping by 35%, and frequency deviation mitigation by 57%, compared to conventional HVDC control strategies. Critical analysis reveals that the hybrid ANN-conventional HVDC framework provides both predictive adaptability and operational reliability, addressing limitations of purely linear controllers while maintaining practical feasibility for real-world deployment. The study also examines theoretical and practical implications, including computational requirements, controller interpretability, and training data dependence, and situates the findings within the broader context of intelligent power system control. The proposed methodology offers a scalable, adaptive, and analytically grounded solution for maintaining AC system stability in increasingly complex and interconnected power networks, providing a foundation for future research on reinforcement learning integration, explainable neural network models, and large-scale smart grid applications.