Substation settlement monitoring based on wireless sensing technology and attention-LSTM model

As the core hub of the power grid system, the substation site settlement directly threatens the safe operation of key power equipment and the overall stability of the grid. However, in practical engineering, settlement data often show strong nonlinearity and nonstationarity and are easily affected by multiple environmental factors. Prediction models usually face difficulties in maintaining sufficient accuracy when handling high-frequency disturbances and identifying sudden abnormal points. To enhance the ability of the model to identify abnormal settlement segments in substation sites, this paper develops a W-ALSTM framework by coupling a wireless sensing network with a task-oriented attention-LSTM predictor. By integrating spatial weighting, wavelet denoising, periodic embedding, positional encoding, and dual-branch spatiotemporal fusion, the method performs multi-scale modeling and feature enhancement on the settlement sequence. The experimental results show that the proposed method achieves a mean absolute error of 1.19 mm, a root mean square error within 1.49 mm, and a coefficient of determination of 0.962 at multiple sensor points. The residuals show extremely low autocorrelation, and the model responds well to sudden trend changes with strong stability in overall error distribution. The results verify the effectiveness of the attention mechanism and periodic embedding structure in improving time-series modeling performance. The proposed model demonstrates good stability, generalization, and deployment adaptability, providing a high-precision and low-latency intelligent prediction solution for substation settlement monitoring.