A self-attention enhanced GRU network for predicting life cycle cost of substation GIS equipment

The challenges of limited sample size and anomalies in the life cycle cost (LCC) data of substation GIS equipment make it difficult to achieve accurate LCC estimation. To address these issues, this paper introduces similar substation GIS equipment LCC data and incorporates a self-attention mechanism to expand the sample size. Additionally, a GRU algorithm is applied to predict time series data and mitigate the impact of anomalies on the prediction. A novel SA-GRU-based model for substation GIS equipment LCC data prediction is proposed. Considering the discrepancies in data characteristics between similar substation GIS equipment LCC data and existing sample data, the self-attention mechanism is utilized to enhance data features. The influence weights between similar and sample data are quantified based on these enhanced features to ensure consistency in data characteristics. Furthermore, the GRU algorithm is employed to alleviate gradient vanishing and exploding issues during training, ultimately enabling LCC data prediction over the entire life cycle. Validation using LCC data from a 110 kV GIS equipment substation in Henan Province demonstrates that the proposed method significantly improves prediction accuracy compared to other prediction approaches.