It is difficult to select the appropriate ventilation mode in clinical mechanical ventilation. This paper presents a nonlinear multi-compartment lung model to solve the difficulty. The purpose is to optimize respiratory airflow patterns and get the minimum of the work of inspiratory phrase and lung volume acceleration, minimum of the elastic potential energy and rapidity of airflow rate changes of expiratory phrase. Sigmoidal function is used to smooth the respiratory function of nonlinear equations. The equations are established to solve nonlinear boundary conditions BVP, and finally the problem was solved with gradient descent method. Experimental results showed that lung volume and the rate of airflow after optimization had good sensitivity and convergence speed. The results provide a theoretical basis for the development of multivariable controller monitoring critically ill mechanically ventilated patients.