Abstract: To address the difficulty in automatic verification caused by inconsistent field naming in protection setting verification of distribution terminals, an intelligent verification method based on knowledge graph semantic matching and deep Q-network (DQN) continual learning is proposed. A protection setting knowledge graph is first constructed, in which setting sheet fields and terminal data fields are defined as standard entities and alias entities, respectively. Semantic similarity between fields is computed using a feature lexicon and the term frequency-inverse document frequency (TF-IDF) method combined with cosine similarity, enabling intelligent field matching. A DQN-based continual learning mechanism is further introduced, allowing the system to learn intrinsic associations among field features from manual review records and to provide reasonable recommendations for unseen yet feature-similar fields, while dynamically updating the knowledge graph. A graphical software tool is developed based on the proposed method. Finally, an experiment is conducted on-site for the verification of protection setting values of distribution terminals in a certain power supply company. The results demonstrate that the proposed technology achieves accurate semantic matching under both standard and variant data conditions, and the average verification time of the intelligent tool is significantly reduced compared to manual verification, greatly improving the efficiency of protection setting verification for distribution terminals.

Key words: distribution terminal, protection setting, knowledge graph, semantic matching, term frequency and inverse document frequency, cosine similarity, deep Q-network