Abstract
In the power industry, secondary operation risk assessment is a critical step in ensuring operational safety. However, traditional assessment methods often rely on expert judgment, making it difficult to efficiently address the challenges posed by unstructured textual data and complex equipment relationships. To address this issue, this paper proposes a hybrid model that integrates graph convolutional networks (GCNs) with semantic embedding techniques. The model consists of two main components: the first constructs a domain-specific knowledge graph for the power industry and uses a GCN to extract structural information, while the second fine-tunes the RoBERTa pre-trained model to generate semantic embeddings for textual data. Finally, the model employs a hybrid similarity measurement mechanism that comprehensively considers both semantic and structural features, combining K-means clustering similarity search with a multi-node weighted evaluation method to achieve efficient and accurate risk assessment. The experimental results demonstrate that the proposed model significantly outperforms the traditional methods in key metrics, such as accuracy, recall, and F1 score, fully validating its practical application value in secondary operation scenarios within the power industry.