The complex nonlinear mapping relationship between the deformation of gravity dam and various environmental quantities makes the input independent variables of the deformation prediction model have high dimension, which affects the accuracy and generalization ability of the prediction model to some extent. To solve the problems, a combined prediction model is proposed to combine principal component analysis, the cuckoo search algorithm, and the nuclear limit learning machine network. The model uses the principal component analysis method to extract the main component information of the water level, temperature, and time-dependent influencing factors related to deformation, and optimize the input of variables of the network model; furthermore, it uses the cuckoo search algorithm, which exhibits better optimization performance, so as to determine the kernel parameters and regularization coefficients of the kernel extreme learning machine network. With the measured data of a gravity dam, the deformation displacement of the dam in the direction of the dam axis and in the upstream and downstream directions is predicted, compared with those of various models, and evaluated with different quantitative indicators. The analysis results reveal that the certainty coefficients R2
of the proposed model in the two different directions are 0.943 and 0.931, respectively, which are higher than those of the traditional neural network model and the stepwise regression model. In the upstream and downstream directions, deformation predictions of different measurement points, the accuracy and generalization ability of the model are better than those of the comparison model, thus verifying the feasibility and advantages of the model.