Research On Noise Prediction And Active Noise Reduction Algorithm Of Rail Vehicles Based On Neural Network

Rail vehicles are commonly used for long-distance travel in our country,noise generated during driving will not only affect the physical and mental health of drivers and passengers,but also one of the key indicators for evaluating the production,performance and comfort of rail vehicles.Compared with the noise collection in a small space such as a car,the noise collection of rail vehicles is more difficult.The reasons are: one is that there are many rail vehicle noise transmission paths and the rail vehicle noise sources are complex and coupled with each other;second is that the complex working conditions on site are inconvenient for noise collection.Therefore,noise prediction can be performed by using multiple non-acoustic parameters that are easy to implement,and serve as the basis for further active noise reduction.At the same time,although passive noise reduction technologies such as sound absorption and sound insulation materials can effectively suppress medium and high frequency noise,the main contribution frequency band of rail vehicle noise is low frequency noise,and its noise reduction effect is limited,so active noise reduction technology is introduced.Therefore,aiming at the control problem of complex multi-source nonlinear low-frequency rail vehicle noise,this thesis proposes a rail vehicle noise prediction and active noise reduction algorithm based on neural network.MRA-2LFNN is used to predict the rail vehicles noise,and based on the prediction results,a C-FNN active noise reduction algorithm is designed to suppress the noise of rail vehicles.Firstly,the principle of neural network and active noise reduction method is studied,and then the performance of conventional neural network model and active noise reduction algorithm is analyzed,which lays a theoretical foundation for the next step of noise prediction and active noise reduction of rail vehicles.Secondly,the MRA-2FNN noise prediction model is designed to estimate the fan noise of the rail vehicle electric traction system,the input variables of the model are determined by multiple regression,and the nonlinear relationship between the main factors affecting the fan noise and the actual noise value is mapped by the fuzzy neural network.After the prediction is completed,the results are analyzed to provide a strong basis for the optimization of the rail vehicle fan noise.Finally,an active noise reduction algorithm based on C-FNN without secondary path sensor is designed.The convolution is used to establish the transfer relationship between the noise sources at different positions of the rail vehicle and the noise at the target noise reduction point,and construct a virtual error signal as the feedback signal of the active noise reduction algorithm to improve the correlation between the reference signal and the real noise at the target noise reduction point,fuzzy layers achieve active noise reduction technology without secondary path sensor.The noise reduction effect of the algorithm is verified by numerical simulation,semi-physical simulation and noise reduction experiment.The research shows that the rail vehicle noise prediction and active noise reduction algorithm based on neural network can provide an important basis for the forward design of rail vehicle noise,eliminate the industry’s chronic problems of first manufacturing and then noise reduction.Besides,it provides a solution for rail vehicle noise active control without secondary path sensors.