Abstract: To address the issue that unknown environmental information affects control accuracy in force-controlled tasks such as grinding for industrial robots, an impedance control method based on a structurally improved physics-informed neural networks (PINN) is proposed. Firstly, different from the traditional PINN output position deviation, an improved impedance model of PINN is proposed. The network output is the second derivative of the position deviation, and the first derivative of the position deviation and the position deviation are obtained by recursive integration, which avoids the amplification of error interference noise and fluctuations when PINN must calculate the multi-order derivative. Secondly, based on the structure of impedance model, a neural network model for impedance parameter identification is designed, and it is connected with the LSTM-based PINN impedance model in series to construct the impedance model of PINN with improved structure. Finally, in order to further improve the control performance, the proportional fractional-order integral (PFI) operator based on proportional fractional-order integral is used to perform feedforward compensation for the set position. Simulation results show that compared with the traditional impedance control and the traditional PINN impedance control, the designed impedance control has better force control effects.