Abstract: This paper proposes a condition early warning and auxiliary decision-making method based on the substation centralized monitoring system. First, an AI early warning architecture for oil chromatography under the substation centralized monitoring system is designed. Second, the Isomap algorithm is applied to perform nonlinear dimensionality reduction on the collected oil chromatography fault sample dataset, and additive Gaussian noise is introduced for fault sample data augmentation to enhance the robustness of the AI early warning model. Then, the processed data are fed into the XGBoost algorithm for model training: the XGBoost algorithm identifies fault types, while the Guangming Electric Power large model provides semantic interpretation, cause analysis and corresponding operation and maintenance (O&M) suggestions. Finally, the proposed method is launched for pilot operation on a new generation of centralized control system of a certain power supply company. Experimental results show that the method achieves reliable fault identification performance, with a precision of 93.9%, a recall of 93.2%, and an F₁ score of 0.935. The F₁ score is 2.9, 17.2, 14.6, and 4.7 percentage points higher than that of XGBoost, SVM, KNN, and RF methods, respectively. Moreover, the semantic interpretation of Guangming Power's large model is clear, and the operation maintenance suggestions are feasible, which can be further promoted and applied in related fields.

Key words: substation centralized monitoring system, XGBoost algorithm, guangming electric power model, dissolved gases in oil, Isomap algorithm