Abstract: With the development of deep learning technologies, neural network methods have been increasingly applied to tsunami wave prediction. Current research predominantly relies on collecting field observation data or using numerical simulations to build databases for prediction. However, due to the simplifications inherent in models and uncertainties in parameter settings, numerical simulation results may have certain limitations. While observational data can provide direct records of real tsunamis, obtaining such data is challenging, with limited observational coverage, making it difficult to effectively evaluate the predictive performance and generalization ability of neural network models. To address these challenges, this study conducted physical model experiments of tsunami-like wave propagation under two terrain conditions with varying incident wave heights and still water depths. A spatiotemporal evolution database of tsunami-like waves was constructed, and two predictive approaches: temporal sequence prediction and observation point prediction, were employed using one-dimensional convolutional neural networks (1D-CNN) to predict tsunami-like waves. This paper initially employed the measurement point prediction method and carried out four sets of prediction experiments in a flat-bottomed water tank. The average value of the root mean square error of the prediction results was 1.393, the average value of the average absolute error was 1.150, and the average value of the coefficient of determination was 0.997. On this basis, this paper further designed eight sets of prediction experiments in a complex terrain water tank. The average value of the root mean square error of the prediction results was 2.431, the average value of the average absolute error was 1.354, and the average value of the coefficient of determination was 0.975. Additionally, in the time series prediction task, the average prediction error of the neural network for the wave crest of tsunami-like waves was 0.137%. The above results indicate that the model has a strong generalization ability and prediction reliability in the prediction of tsunami-like waves, providing a scientific basis for verifying the application value of neural networks in tsunami wave prediction.