A multipoint prediction model for nonlinear displacement of concrete dam

Measurement points are installed in concrete dams to monitor nonlinear displacement behavior of dam systems working in complex and aggressive environment. Most modeling works for predicting displacements of these measurement points are limited to single output in point-by-point modeling way, with little attention paid to partition-based multipoint combined modeling. Therefore, a novel multipoint prediction model (MPM) for concrete dam displacement is proposed whose novelty lies in, first, a partition method for measurement points of dam engineering that considers waveform used to identify dam deformation trend and chaos used to describe overall stability and inherent stochasticity in nonlinear displacement series of dam system, and second, a multi-output support vector machine optimized by an improved grey wolf optimizer with better hierarchy and information transmission. The monitoring data of a super-high concrete arch dam are used to verify the proposed model. Results indicate that the MPM shows feasibility and generalization capability regarding dam engineering and reaches the performance level of single-point benchmark models. Robustness, sensitivity, and reasonability of the model are also discussed by random noise, sliding window, and perturbations.