Аңдатпа
Aiming at the challenges of knowledge fragmentation, inefficient retrieval and empirical decision-making in the operation and maintenance of modern data centers, this paper designs an intelligent decision support system (DSS) which integrates dynamic knowledge graph (KG) and semantic retrieval. The system constructs an operation and maintenance field KG covering the four-dimensional relationship among equipment, services, faults and solutions, adopts BERT+BiLSTM model to extract the entity relationship of multi-source heterogeneous data, and introduces a dynamic updating mechanism based on incremental learning, so that the F1 value reaches 0.88, which is 20 percentage points higher than that of the traditional rule method. Semantic-enhanced hybrid retrieval engine integrates BERT semantic coding and feature extraction of time-series convolutional network (TCN) to achieve full-link reasoning from symptom to fault chain, with an average retrieval accuracy of 83%, which is 56.6% higher than that of keyword retrieval. The risk-aware decision-making optimization model uses Monte Carlo simulation to evaluate the system-level risk of the restoration scheme, and combines reinforcement learning (RL) to generate the optimal operation sequence, which improves the fault diagnosis accuracy to 92% and shortens the average diagnosis time to 3.5 minutes. The experimental results show that the system significantly improves the efficiency of fault location and the quality of decision-making in real data center operation and maintenance scenarios, and provides an interpretable and adaptive closed-loop solution for intelligent operation and maintenance of complex distributed systems.