Abstract: Dam deformation is nonlinearly correlated with water level, temperature, aging and many other factors. Support vector machine (SVM) is of great superiority in dam safety monitoring as it accommodates small sample, nonlinear and high dimensional learning problems. In this article, the principle of SVM is expounded, the procedures of building an SVM-based deformation monitoring model are summarized, and parameter optimization method is introduced as well. The prediction ability of the SVM-based monitoring model for concrete dam deformation is analyzed through a case study. Results demonstrate that the short term prediction ability of the model is better than its long term prediction ability; the prediction ability is affected by the number of prediction sets rather than by algorithm optimization. The results indicate that selecting an appropriate number of prediction sets is important to the validity of the model.

Key words: concrete dam, deformation monitoring, Support Vector Machine, particle swarm optimization, prediction ability, case analysis