Spatial-Temporal Features Based Sensor Network Partition in Dam Safety Monitoring System

Area of Science:

• Geotechnical Engineering
• Structural Health Monitoring
• Data Science

Background:

• Massive sensor data from dam structures poses challenges for real-time operational status evaluation.
• Existing regional partitioning methods often neglect sensor spatial distribution and time-series variations, hindering accurate analysis.
• Effective dam safety monitoring requires methods that capture spatial-temporal features and data correlations.

Purpose of the Study:

• To develop a network partitioning algorithm (NPA) that accurately represents spatial and temporal features of dam monitoring data.
• To improve the real-time evaluation of dam safety by integrating spatial-temporal data analysis.
• To enhance the understanding of dynamic changes in dam working conditions.

Main Methods:

• Utilized a time-series denoising autoencoder (TSDA) to compress high-dimensional monitoring data and extract spatial-temporal features.
• Proposed a network partitioning algorithm (NPA) leveraging TSDA-derived spatial-temporal features.
• Introduced an auxiliary objective variable to optimize the NPA's objective function for improved partition quality.

Main Results:

• The NPA demonstrated superior performance compared to TSDA+K-Means and TSDA+GMM on public and real arch dam datasets.
• NPA improved the silhouette coefficient by 45.1% and 58.4% over TSDA+K-Means and TSDA+GMM, respectively.
• NPA increased the Calinski-Harabaz Index by 30.8% and 61.6% compared to the other methods.

Conclusions:

• The proposed NPA effectively captures essential spatial-temporal features for dam monitoring data partitioning.
• NPA significantly enhances the accuracy and reliability of dam safety evaluation and real-time monitoring.
• The algorithm provides a robust foundation for analyzing dynamic changes in dam structures.