Abstract: Due to the protection of surface thermal insulation layer, the internal temperature of concrete dam in cold region lags behind the air temperature. When internal thermometers are lacking, HST (hydrostatic-seasonal-time) model generates large error in predictions; even when internal thermometers are present, MR (multiple regression) model can not refelct the nonlinear relation between temperature and deformation. To address the problem of inaccurate deformation prediction in alpine regions, we propose to use a support vector regression (SVR) weighted by the RReliefF algorithm and DLM(Distribution Lag Model) temperature factors to predict dam displacement. The necessary hyperparameters are optimized using an improved whale swarm approach with OBL (Opposition-based Learning). By comparing the performance of the proposed model with MR and other machine learning algorithms, we found that the proposed model has higher prediction accuracy and better reflects the working characteristics of an insulated concrete dam.

Key words: Regression ReliefF Factor Weighted Support Vector Machine Regression, Opposition-based Learning Whale Optimization Algorithm, DLM, dam deformation prediction, alpine regions