Abstract: Mine ventilation resistance coefficient is crucial to the safety management, diagnosis and intelligent development of mine ventilation system, and the airflow is usually used as the basis for the inversion of ventilation resistance coefficient, but the conventional nonlinear optimization method has the problem of non-uniqueness, which affects the accuracy of the inversion.A novel optimization method based on deep reinforcement learning(DRL)is used to invert the resistance coefficients, where the inversion is regarded as 1 Markov decision process, and the ventilation network solution model(VNSM)is embedded into the DRL environment, and the proximal strategy optimization is used to optimize the agent's strategy.Field tests show that MAE of the computed airflow versus the measured airflow is 0.354,MSE is 0.287,RMSE is 0.536,and MRE is 0.013.Compared with the standard genetic algorithm, differential evolutionary algorithm, and evolutionary strategy algorithms, DRL method reduces the values of MRE,MAE,RMSE and MSE by 23.5%,15.3%,14.1%,and 26.4%,respectively, and the sensitivity difference of DRL method to different roadways is small compared with other algorithms.