Abstract: A novel sparse strategy was proposed for the least square support vector machines regression,and the dynamic process identification of aircraft engine was analyzed in particular.In the process of solving the original sample,the special nonlinear mapping matrix was recursively decomposed by using the modified Gram-Schmidt method,and the excellent sample was selected when the square of remaining output reached the threshold value;thus,the new sample quantity was reduced and the least square support vector machines(LS-SVM) regression model training speed was enhanced by saving memory.The simulation result shows that,the algorithm has better sparseness,less training time and equivalent precision;and the aircraft engine dynamic process simulation shows that the predicting error of high rotor's relative rotational speed is less than 0.2%,the predicting error of the low rotor's relative rotational speed is less than 0.35%,and the predicting error of the temperature at the exit of high-pressure turbine is less than 3.5 ℃.The model is suitable for aircraft engine's control and system simulation.