Research on an Agent-Driven Knowledge Management System for Power Grid Infrastructure Projects

With the increasing complexity of power grid projects and the growing scale of data, traditional knowledge management approaches are no longer sufficient to meet the collaborative and intelligent decision-making needs of transmission infrastructure projects. This paper proposes a construction method for a knowledge management system based on an agent-driven mechanism. Engineering participants are modeled as functionally heterogeneous agents, and task-driven multi-agent collaboration is achieved using the Belief-Desire-Intention (B-D-I) model and state transition diagrams. Engineering knowledge is automatically extracted and graphically structured using BERT-CRF and LSTM-Attention models. Knowledge representation and reasoning are further enhanced through graph neural networks, while a Dueling DQN algorithm is employed to optimize the knowledge recommendation strategy, improving the system's adaptability in dynamic environments. Experiments on real-world transmission project datasets demonstrate that the proposed system significantly outperforms existing approaches in terms of knowledge extraction accuracy, graph reasoning performance, recommendation hit rate, and task collaboration efficiency. This research provides a cognitively and inferentially capable knowledge management solution for complex power engineering projects.