Abstract: If a fault occurs in the cable of the coal mining machine during its reciprocating movement, it may lead to equipment shutdown or safety hazards. Using optical fiber sensing technology, embedding fiber optic units in traditional cables can timely acquire their curvature and shape information, making it the preferred solution for monitoring the state of coal mining machine cables. This paper proposes a shape restoration algorithm for optical fiber units based on Physics-Informed Neural Networks (PINN). By constructing the ResNet-SEAttention-PINN model, we predict the curvature components κ1 and κ2 of the optical fiber units and calculate the centerline equation based on a parallel transport framework to restore the shape of the optical fiber units. This is a key process to ensure the accuracy of monitoring the state of coal mining machine cables. Experimental results show that the shape restoration algorithm for optical fiber units achieves high accuracy in predicting curvature components, with a Mean Squared Error (MSE) of 1.061×10-8 and a Mean Absolute Error (MAE) of 8.715×10-5, and a coefficient of determination (R2) of 0.981. The average absolute position error (Mean APE) and maximum absolute position error (Max APE) for restoring the shape of the optical fiber unit using the parallel transport framework are 0.2697 mm and 0.7768 mm, respectively. The research findings can provide reliable technical support for real-time monitoring of cable oper-ating status and intelligent safe production in coal mines.