Abstract: Deformation is a comprehensive reflection of the safety state of a dam; To ensure its long-term service, a significant task is to develop a reliable prediction model between its deformation and environmental variables. Previous models are easily affected by the noises in data sets or structural parameters, and often fall into local extremum or overfitting. To improve the accuracy and generalization ability of the model, this paper presents a back propagation (BP) neural network method based on the singular spectrum analysis (SSA) and an improved whale optimization algorithm (IWOA). The method uses SSA to filter out noises in the raw data, and extracts feature components from the dam deformation time series. Then, an IWOA-optimized BP neural network is used to explore the complicated nonlinear relationship between the denoised data and environmental variables. Practical applications to the Bailianya arch dam show that in comparison with the traditional optimization algorithm, SSA can eliminate the outliers effectively from the raw data, and the BP neural network optimized by IWOA is much better in accuracy and stability, both applicable to the analysis and prediction of dam deformation monitoring data.

Key words: dam deformation prediction model, singular spectrum analysis, whale optimization algorithm, enhanced optimization strategy