Secondary Operation Risk Assessment Method Integrating Graph Convolutional Networks and Semantic Embeddings

Area of Science:

• Electrical Engineering
• Computer Science
• Risk Management

Background:

• Secondary operation risk assessment is crucial for power industry safety.
• Traditional methods struggle with unstructured data and complex equipment relationships.
• Expert judgment limitations hinder efficient and accurate risk assessment.

Purpose of the Study:

• To develop a hybrid model integrating graph convolutional networks (GCNs) and semantic embedding techniques for power industry secondary operation risk assessment.
• To overcome the limitations of traditional methods in handling unstructured data and complex equipment relationships.
• To improve the efficiency and accuracy of risk assessment in operational scenarios.

Main Methods:

• Construction of a domain-specific knowledge graph for the power industry.
• Utilizing GCNs for extracting structural information from the knowledge graph.
• Fine-tuning the RoBERTa pre-trained model for generating semantic embeddings of textual data.
• Implementing a hybrid similarity measurement combining K-means clustering and multi-node weighted evaluation.

Main Results:

• The proposed hybrid model significantly outperforms traditional methods in risk assessment.
• Key performance metrics including accuracy, recall, and F1 score show substantial improvement.
• The model demonstrates practical application value in secondary operation scenarios within the power industry.

Conclusions:

• The hybrid GCN and semantic embedding model offers a superior approach to secondary operation risk assessment in the power industry.
• This method effectively addresses challenges posed by unstructured data and complex equipment interdependencies.
• The validated performance confirms the model's utility for enhancing operational safety.