Abstract: Embedding fiber optic units into traditional shearer cables and using fiber optic sensing technology to obtain curvature and shape information in real time is an optimal solution for the condition monitoring of shearer cables. Existing shape reconstruction methods of fiber optic units based on mathematical modeling have low computational efficiency and cannot meet the requirements of real-time monitoring, while data-driven methods have weak extrapolation capability and are difficult to adapt to complex and dynamic underground working conditions. To address this issue, a shape reconstruction algorithm of fiber optic unit in shearer cable based on Physics-Informed Neural Network (PINN) prediction was proposed. In the PINN model, a Residual Network (ResNet) integrated with an SEAttention mechanism was introduced to construct a ResNet-SEAttention-PINN model to predict the curvature components of the fiber optic unit. Based on the strain-curvature mapping relationship, the parallel transport frame was used to solve the centerline curve equation of the fiber optic unit to reconstruct its shape. Simulation results showed that the Mean Absolute Position Error (Mean APE) and the Maximum Absolute Position Error (Max APE) of the proposed algorithm based on the ResNet-SEAttention-PINN model were 0.269 5 and 0.776 7 mm, respectively, which were significantly better than those of comparison algorithms based on convolutional neural network, recurrent neural network, PINN, ResNet-PINN, and SEAttention-PINN models. A physical experiment was carried out using an adjustable-radius calibration frame and an RP3000 dynamic distributed fiber optic strain testing system, and the results showed that under practical working conditions, the algorithm still maintained high reconstruction accuracy, with a Mean APE of 0.312 5 mm.