To eliminate the influence of environmental load on the guided wave-based structural health monitoring （SHM）， a load compensation method based on deep learning neural network（DLNN） is proposed based on the load influence experiments carried out on carbon fiber composite structures. This method utilizes compensation standards for the establishment of the DLNN with the optimized network structure and network parameters. As a result， the storage of reference signals is effectively reduced and a wide range of load can be compensated by the proposed method. The effectiveness of the method is verified by experimental data. Experimental results show that the compensation accuracy reaches -20 dB in the load range of 0~90 MPa. Finally， the compensation method combined with the damage imaging method is demonstrated on the damage monitoring of the carbon fiber composite structures. Results show that the damage localization error is less than 0.9 cm and the reliability of damage diagnosis under the influence of load is effectively improved.