Abstract: Gear heat treatment quality affects the reliability of molded products, life, precision and other physical and chemical indicators. Long-term operation of the enterprise contains a large number of potentially valuable industrial production data, but it is cut and divided. Therefore, a data-driven method based on historical heat treatment process data was proposed, which was called LSTM-NSGAII. The method can optimize the quality of the gears by solving the heat treatment process parameters. By pre-processing the gear heat treatment process data, the process database was constructed; the decision variables were screened, and the LSTM neural network model of the gear carburising and quenching heat treatment process temperature, carbon potential, and time was established to achieve the prediction of the hardness and the depth of effective carburizing layer; the NSGAII algorithm was used to globally search for the optimum, and the optimal heat treatment process parameters were obtained via data-driven quality assessment. The experimental results show that the method proposed can improve the gear strength by 1.4% under the condition of meeting the standard, and the deviation of the effective carburized layer depth from the central target value is around 0.03 mm, providing a feasible optimization method for heat treatment process technology.