基于EMD与BP神经网络的汽车关门声品质预测

Sound Quality Prediction for Vehicle Door-Slamming Noise Based on Empirical Mode Decomposition and Back Propagation Neural Network

本文中提出一种基于经验模态分解(EMD)与BP神经网络相结合的汽车关门声品质预测方法。该方法首先基于EMD提取关门声音信号特征;然后,对样车的关门声品质进行主观评价试验,得出主观评价值;最后将得到声音信号特征作为输入,主观评价值作为输出,利用神经网络进行训练得到基于EMD的汽车关门声品质的预测模型。同时,以心理学参数值作为神经网络输入,建立了基于心理学参数的声品质的另一预测模型。分别用两个模型对30辆车关门声样本进行预测的结果表明,基于EMD模型的预测数据,更接近主观评价的结果,即具有更高的声品质预测精度。

A sound quality prediction method for vehicle door slamming noise is proposed in this paper based on the combination of empirical mode decomposition （EMD） and back propagation （BP） neural network. With the method, the features of door-slamming sound signal are extracted first based on EMD ; then, a subjective evaluation test on the door-slamming sound quality of sample vehicles is conducted with subjective evaluation values obtained, and finally by using neural network for training with the features of sound signal as input and the subjective evaluation values as output, an EMD-based vehicle door-slamming sound quality prediction model is obtained. Meanwhile another prediction model for sound quality based on psychology parameters is also built with psychology parameters as neural network input. The results of predictions on the door-slamming sound samples of 30 vehicles with two models respectively show that compared with psychology parameters-based model, the EMD-based model gets a predicted data much closer to the results of subjective evaluation, meaning it has higher accuracy in predicting vehicle door-slamming sound quality.