Intelligent reflector shift optimization method and system based on unsupervised learning

The invention discloses an intelligent reflector shift optimization method and system based on unsupervised learning, and belongs to the technical field of communication. Comprising a base station, a user and an intelligent reflecting surface, a direct connection link between the base station and the user is blocked, input data is indirect cascade channel state information, and data preprocessing comprises the following steps: firstly, converting a one-dimensional channel state information vector into a two-dimensional complex matrix, splitting a real part and an imaginary part of the two-dimensional complex matrix, and additionally adding an absolute value part; forming three-channel data as the input of the neural network; the whole network is composed of two convolution-attention blocks, a flattening layer and a plurality of full-connection layers, a Squeeze-and-exccitation network is selected as an attention mechanism, a label-free unsupervised learning method is adopted as a model, a loss function is def