Aiming at improving the sound quality of the interior noise of special vehicles， the XGBoost algorithm is used to establish a sound quality prediction model. The average relative error between the predicted value of the model and the actual subjective evaluation value is 2.43%. At the same time， according to the nonlinear and non-stationary characteristics of interior noise， an algorithm based on Empirical Mode Decomposition （EMD） and Filtered-x Least Mean Square （FxLMS） algorithm is proposed.Based on the active control method， the results of the prediction model show that subjective irritability is optimized by 2.11， with an improvement rate of 26.6%. This method has a good control effect on the nonlinearity and non-stationarity of special vehicle interior noise， and can effectively improve the interior sound quality.