[Objective] Intelligent operation and maintenance （O&M） of high voltage direct current （HVDC） systems face critical challenges， including insufficient accuracy in state assessment， limited predictive foresight in risk assessment， and poor real-time performance in fault diagnosis. Traditional methods are constrained by limitations in multi-source data fusion， small-sample generalization， and decision interpretability. This paper systematically reviews the integration pathways of knowledge graphs and pre-trained models， aiming to provide theoretical references for building a new generation of HVDC intelligent O&M systems. [Methods] First， the technical bottlenecks in three core scenarios—state assessment， risk prediction， and fault diagnosis—are analyzed. Subsequently， the integration fundamentals and synergistic mechanisms of knowledge graphs and pre-trained models are elaborated. Finally， key advancements and solutions applied to these three scenarios are summarized. [Results] The deep integration of knowledge graphs and pre-trained models effectively addresses pain points such as data silos， weak generalization with small samples， and the “black-box” nature of decision-making. Notable breakthroughs have been achieved in multi-modal feature alignment， temporal causal reasoning， and cloud-edge collaborative diagnosis. [Conclusions] However， challenges remain regarding automatic knowledge updating， the integration of physical mechanisms， and real-time assurance. Future research should focus on automated knowledge engineering， physics-informed pre-trained models， and trustworthy evaluation systems.