In this paper, according to the definition of power cable channel damage and the actual situation, the vibration signals of mechanical excavation, crushing hammer construction, manual excavation and heavy vehicles passing by are selected as the initial data for the prediction of preventive damage of power cable channels. After completing the data collection, Variable Difference Modal Decomposition (VMD) is applied to reduce noise and obtain the IMF components. The eigenvalues such as energy entropy, cliff factor, waveform factor, center of gravity frequency and frequency standard deviation of each IMF component are extracted to form an eigenvector set, thus forming the data set for the study and analysis. The data set is divided into a training set and a test set according to the ratio of 8:2, and the long and short-term memory neural network is used to study the power cable channel damage prediction and diagnosis. The MAE in the single-step prediction of LSTM is 1.08, the MRE is 2.69%, and the RMSE is 1.39, and the prediction model in this paper is much better than the control prediction model, which indicates that the LSTM network can well predict the vibration of the damage of the power cable channel. It shows that the LSTM network can accurately predict and diagnose the signal trends and fluctuations caused by damaged power cable channel vibration.