To solve the problems associated with complex network structures and low model accuracy when using the local linear model tree (LOLIMOT) to identify nonlinear engine systems, we propose a modified LOLIMOT structure that combines a nonlinear autoregressive moving average model (NARMAX) with a LOLIMOT network. Based on the concepts underlying NARMAX, we replace the linear function of the local sub-model with a nonlinear polynomial function. Using a forward selection method based on the AIC saliency criterion, we then select the appropriate nonlinear terms according to their importance, and use the simplified nonlinear function to construct the local sub-model of LOLIMOT, which yields a modified LOLIMOT. We then use the modified LOLIMOT to identify a nonlinear transient process for an aero-engine. The results show that the modified LOLIMOT reduces the complexity of the original model by 46% and increases the prediction accuracy by at least 50%. These results verify that the modified model has a more efficient network structure in the aero-engine control field, low model complexity, and the ability to meet the demand for high accuracy.