Abstract: A spatio-temporal prediction model for dam deformation is proposed, which incorporates the wavelet theory and the Sparrow Search Algorithm (SSA) to optimize the extreme learning machine (ELM). This model addresses the challenge of accurately describing the overall response characteristics of dam deformation using single measurement point prediction models which fail to account for the spatial relations among measure points. Additionally, it overcomes the limitations of statistical models based on regression analysis, which struggle to uncover the complex nonlinear mapping relationship between environmental variables and the magnitude of deformation, often resulting in poor prediction accuracy. To validate the feasibility of the proposed approach, an actual dam project is taken as an illustrative example. The approach begins with wavelet analysis to eliminate noise from the original displacement measurements of the dam. Subsequently, the influence of coordinate changes in the measurement points on displacement is considered. SSA-ELM is employed to establish non-linear models for independent and dependent variables, constructing a spatio-temporal prediction model for dam deformation based on wavelet analysis. Application of the proposed model to a real-world example demonstrates its ability to accurately predict deformation across non-arranged measuring points. The model exhibits impressive performance indicators, including a highly significant complex correlation coefficient of 0.996 8, a root mean square error of 0.340 4, and an average absolute error of 0.275 4, which exceed those achieved by both the ELM model and statistical model. By integrating both temporal and spatial dimensions, the proposed model achieves high prediction accuracy and holds significant value as a reference for the analysis and evaluation of dam safety.

Key words: dam deformation prediction, wavelet analysis, sparrow search algorithm, extreme learning machine, spatiotemporal distribution model