An Approach of GIS Condition Assessment Based on Historical Cases and Measurement Data

The condition assessment of Gas Insulated Switchgear (GIS) ensures the safety and stability of power grids. However, the traditional GIS condition assessment models only analyze the partial discharges and ignore the information recorded in historical case texts, making it difficult to improve the model's accuracy. Therefore, this paper proposes a GIS condition assessment model based on historical cases and measurement data. First, the partial discharge data in the past 5 days were analyzed using the convolutional neural network, and the pattern feature vectors were extracted. Second, the characters in the case text are converted into word embedding vectors using the word embedding model, and the case feature vectors are extracted using the Long Short-Term Memory (LSTM) network. Finally, the pattern feature vectors and the case feature vector are added, and LSTM is used for information analysis to obtain the insulation condition of GIS. Verified on the samples collected from substations, the proposed algorithm effectively evaluates the insulation condition, and the effect is improved by 19.39% compared with the traditional models.