A new diagnosis method based on wavelet packet and neural network is proposed for the fault diagnosis of submarine cables.By this method

the fiber Brillouin frequency shifts in the cables are converted into temperature and strain data.The monitoring signals are first denoised by using the six-layer wavelet decomposition and the faulty data are decomposed and reconstructed by three-layer wavelet packet

and then such structural eigenvectors as energy

standard deviation and Shannon entropy are extracted and inputted to the Back Propagation（BP）neural network for training and testing.The experimental analysis shows that inputting the standard deviation as the eigenvector to the BP neural network is the optimum method for the fault diagnosis in submarine cables.