KG-DNA: Knowledge graphs as network DNA for purpose-preserving intelligence in autonomous 6G networks

Autonomous network management in (Formula presented) must continuously adapt to dynamic conditions while preserving operator-defined service intent. Existing learning- and optimization-based approaches primarily optimize short-term performance. However, they do not explicitly control how repeated local decisions affect long-term service objectives. As a result, small intent violations may accumulate over time, leading to purpose drift. This paper proposes Knowledge Graph as Network DNA (KG-DNA), a bio-inspired management framework that encodes network structure, operational policies, and SLA/QoS objectives as heritable genetic components within a knowledge graph. Autonomous adaptation is realized through a three-stage process. In Stage I, service intent is formalized as regulatory genes that define immutable semantic constraints. In Stage II, operational genes are evolved using an intent-aware fitness function and constrained mutation, ensuring that each adaptation step remains intent-feasible. In Stage III, long-term purpose regulation is introduced by monitoring temporal intent violations, detecting persistent drift, and biasing future evolutionary updates toward intent-aligned policy regions through a bounded regulation factor. By combining step-level feasibility control with temporal regulation, KG-DNA enables performance optimization without sacrificing long-term service purpose. The evaluation scenario demonstrates improved SLA continuity, reduced purpose drift, and stable adaptation behavior compared to conventional autonomous management approaches.