Abstract: Deformation is an important effect reflecting the dynamic evolution of a dam. In order to improve prediction precision of the statistic model, with the advantage of the extreme learning machine (ELM) to deal with the nonlinear problems, data mining for dam displacements residuals of the statistic model is conducted. Because ELM is short of the chaotic dynamic characteristics, in order to solve this problem, the chaotic dynamic characteristics of the dam displacements residuals of the statistic model are analyzed by the chaos theory, the results reveal its chaotic characteristics, and then the phase space is reconstructed, thus it can provide priori knowledge for the chaos-optimized ELM. Based on the statistic model, combined with the advantages of ELM, a combined model combining a statistic model with the chaos-optimized extreme learning machine(ELM) is developed. The combined model is applied to the case histories of practical engineering. The analysis results show that the combined model is reasonable, and the prediction precision is higher than the statistical model and the combined model combining the statistical model with the chaos-optimized BP neural network, which will be of application value to researchers in dam deformation monitoring.