In order to solve the issues of noise interference in the prototype monitoring data of concrete dams and the difficulty in optimizing the numerous hyperparameters of intelligent algorithm used for deformation prediction, a concrete dam deformation prediction model is proposed based on the ensemble empirical mode decomposition (EEMD), artificial fish swarm algorithm (AFSA) and convolutional neural network (CNN). This model uses EEMD to decompose the original dam deformation data to obtain the intrinsic mode function (IMF), and utilizes the wavelet threshold denoising method to denoise the noisy IMF components and reconstruct the components. The hyperparameters of the CNN model are optimized using the AFSA, and the reconstructed data is trained with the optimized CNN model. The trained model is subsequently used for prediction. The validation results from a case study of a high arch dam show that, compared with models such as CNN, extreme learning machine (ELM) and back propagation (BP) neural network, the proposed model exhibits higher accuracy and stronger stability in concrete dam deformation prediction.