Global optimization-based perturbation neural network optical fiber nonlinear equalization method

The invention belongs to the technical field of optical fiber communication, and particularly relates to a perturbation neural network optical fiber nonlinear equalization method based on global optimization, and the method comprises the steps: obtaining transmitting end data collected by a receiving end of an optical fiber communication system; utilizing a first-order perturbation theory to introduce a triple into the transmitting end data to carry out input feature reconstruction; performing feature extraction on the reconstructed input features by adopting a principal component analysis method to obtain dimensionality-reduced input features; based on an iterative pruning algorithm, compressing the full-connection neural network equalizer to obtain a sparse-connection neural network equalizer; training a sparse connection neural network equalizer by adopting a transfer learning algorithm, and enabling the trained neural network equalizer to be in an optimal convergence state so as to compensate nonlinear da