Hierarchical AI Control and Protection for Renewable Multi-Terminal HVDC Grids

General Background: The transition toward large-scale wind and solar integration requires transmission infrastructures capable of stable long-distance power delivery. Specific Background: Multi-terminal VSC-HVDC grids provide this capability but introduce complex coordination challenges in control and ultra-fast DC fault protection. Knowledge Gap: Existing studies treat advanced control and protection as separate problems, limiting system-level coordination during disturbances and contingencies. Aims: This study develops a unified hierarchical AI-driven architecture that integrates deep reinforcement learning (DRL) for primary converter control, an AI-based secondary coordinator for system-wide optimization, and a CNN-assisted hybrid protection scheme for rapid fault management. Results: Co-simulation on a modified CIGRE B4 MTDC benchmark with wind and PV sources shows 62.4% faster settling time, 62.2% lower voltage overshoot, 98.93% fault classification accuracy with 0.61 ms detection latency, and 44.4% faster recovery under N-1 contingencies. Novelty: The work presents a single AI framework that jointly governs normal operation and emergency fault response across hierarchical control layers. Implications: The approach supports resilient, self-adaptive, and protection-aware MTDC grids for renewable-dominated power systems. Highlights: Unified AI architecture for control and protection in MTDC systems. Millisecond-level fault detection with high classification accuracy. Significant improvements in dynamic stability and contingency recovery. Keywords: VSC-HVDC, Deep Reinforcement Learning, Fault Protection, MTDC Grid, Renewable Integration