Abstract: A pipeline corrosion stage prediction method based on K-means clustering algorithm and long short term memory (LSTM) neural network was proposed to use pipeline corrosion signals obtained by acoustic emission detection. First, the corrosion signals were classified by K-means algorithm, and then the LSTM neural network model was constructed. The model parameters were optimized by using acoustic emission waveform as the starting point through unsupervised learning. Finally, the pipeline corrosion stage was predicted and the model was evaluated according to evaluation indicators. The results show that LSTM neural network model could be more stable and more robust by adding hidden layers appropriately. Compared with existing fault diagnosis models, LSTM neural network model had higher accuracy.