Ship shaking can cause cargo PWP (pore water pressure) increasing, the prediction of PWP is helpful for navigation safety. This paper presents a machine learning method for PWP time series prediction learning. Source data of iron concentrate are collected from scaled model test, they are used to train a time series prediction model and test the accuracy and effective of our method. The proposed method is based on PSR (phase space reconstruction) and LSTM (Long Short-Term Memory) Network. A input matrix is constructed by phase space reconstruction technology and the prediction model can be learned by a specific Long Short-Term Memory Network. The single-step pore water pressure prediction model is achieved, when MSE loss function value is minimum, the R value 0.98 is maximum is better than the baseline R value 0.93. This result suggests that this PSR-LSTM is more effective than LSTM, it can be a complement for physical test and numerical simulation.