Correlation cluster analysis of slope safety monitoring data in reservoir areas.

Abstract

Current predictive methods for dam failures in reservoirs remain limited, indicating that the underlying mechanisms of such failures are not yet fully understood. To further elucidate the interrelationships among safety monitoring data in the reservoir area, this study established 36 monitoring cross-sections distributed across upper, middle, and lower slope zones. Each cross-section was instrumented with eight different types of monitoring devices. A total of 4,320 samples were collected (432 samples per instrument type), resulting in an overall dataset of 34,560 measurements. The monitoring data were sequentially analyzed using: (1) descriptive statistics, (2) Welch/Brown-Forsythe post hoc One-way analysis of variance (ANOVA), and (3) cluster analysis. The results demonstrate that: (1) Significant correlations exist among monitoring variables, with the strongest positive correlation observed between loading and lean (r = 0.40), while the strongest negative correlation occurred between sedimentation and stress (r = -0.39). (2) Cluster analysis of the eight monitoring variables revealed two distinct clusters: soil displacement, stress, and water-level formed one cluster, while the remaining variables comprised the second cluster. In summary, variations in monitoring data and their correlations resulted from water-level and environmental changes in the reservoir area, with spatial differences across monitoring types. A thorough investigation of these variations and their causes will enable accurate safety assessments of the reservoir area and support tailored response strategies for different locations.