Machine learning-guided optimal power flow for weak hybrid AC-DC grids

This paper presents a novel machine learning-guided approach for optimal power flow in weak hybrid AC-DC grids. The methodology addresses two key challenges: voltage stability in weak grids and the integration of LCC-HVDC systems. First, a two-factor Taylor series approximation is developed to incorporate LCC-HVDC reactive power consumption into standard AC OPF models. Second, the paper demonstrates that using pre-trained machine learning models to predict bus voltages can achieve comparable reliability to conventional OPF methods while enabling more complex optimization scenarios. The approach is validated on Pakistan's transmission network which includes a 4 GW embedded LCC-HVDC link. Results show that the ML-guided quadratic convex programming (QCQP) method better reflects the physical reality of weak grids compared to interior point methods. The model successfully optimizes both AC and DC parameters while maintaining computational tractability, providing a practical solution for operations planning in weak grids with embedded HVDC links.