In order to improve the speed and accuracy of seismic response prediction of aqueduct structure, Jiehe aqueduct was studied, and Midas Civil-2021 was used to construct the finite element model. On the basis of verifying the reliability of the finite element model, the sample data were obtained, and the machine learning model was constructed by using the long short-term memory (LSTM) algorithm and the time series transformation (TSTF) algorithm to predict the nonlinear seismic response of the aqueduct, and the prediction results were optimized by adjusting the time window size and sampling period. The prediction results of the displacement response at the top of the pier show that the average accuracy of the LSTM model and the TSTF model is 76.22% and 88.30%, respectively. Compared with the prediction speed of the finite element model, that of the LSTM model and the TSTF model is improved by 128.54% and 47.90%, respectively. The analysis results of the vulnerability of the aqueduct structure show that the damage exceedance probability of the pier gradually increases with the rise of the water level.