Research On Intelligent Active Vehicle Passing Noise Of City Control

With the increase in demand for urban transport,the noise pollution caused by cars has become more and more serious,especially the noise of cars passing through the streets has brought a great impact on the work and life of residents along the streets.The energy passing through the vehicle is often predominantly low frequency,characterized by nonsmoothness,short duration and accompanied by pulse signal.At present,the most common means of controlling vehicle pass-by noise is to set up noise barriers on both sides of the road,which is a Passive Noise Control(PNC)measure and is not ideal for controlling noise in the low and medium frequency bands.Therefore,in this paper,the following research has been carried out to address the problem of vehicle pass-by noise cancellation based on the advantages of Active Noise Control(ANC)for low-frequency noise control,and has achieved good results.(1)To solve the problem of non-smoothness of the vehicle pass-by noise signal,the Complementary Ensemble Empirical Mode Decomposition(CEEMD)method,which has strong processing capability for non-smooth signals,is introduced in this study.The method can reduce the non-smoothness of the signal by adding positive and negative Gaussian white noise to the decomposition.In addition,the decomposition of the timeseries signal into multiple bands of Intrinsic Mode Functions(IMF)makes it easier to extract the data implicit information and to fully extract the characteristics of the data series,thus reducing the difficulty of finding the optimal filter weight coefficients in this study.In addition,this study introduces the frame-splitting plus windowing method in speech signal processing,which decomposes a section of a car through a noisy signal into several shorter length segments and performs CEEMD decomposition in turn,avoiding the long decomposition time due to the long noisy signal,which makes the signal input to the filter delayed.At the same time,the pseudo-components in the multiple components of the CEEMD decomposition are removed by correlation analysis,to avoid the pseudocomponents affecting the system noise reduction.The simulation results show that as the number of decompositions increases,the frequency of each order component obtained from the decomposition gradually decreases,and the smoothness of the noise signal gradually improves.(2)Traditional filter adaptive algorithms have problems such as slow convergence of the filter weight coefficients,low system robustness,and even scattering of the control system when the input signal is complex,in the process of processing time-varying signals.To address these problems,this study proposes an improved Filtered-X Least Mean Square(FXLMS)algorithm based on the Squirrel Search Algorithm(SSA,Squirrel Search Algorithm).In the proposed algorithm,SSA is used to replace the gradient descent method in the iterative process of the traditional FXLMS algorithm to improve the global search capability of the filter power coefficient finding,while effectively avoiding the tendency to fall into local optima during the finding process,speeding up the convergence speed and reducing the number of iterations.The simulation results show that the proposed SSA-FXLMS algorithm outperforms the FXLMS algorithm and the Filtered-S Least Mean Square(FSLMS)algorithm in terms of noise reduction effect and convergence speed.(3)In this study,a road noise prediction model based on CEEMD and GRU neural network is developed,making full use of the information mining capability of CEEMD and the learning capability of GRU neural network,and based on the prediction results,the Person feasibility analysis process is completed before the adaptive control session is executed and the number of filters required and the number of component orders of input adaptive control are determined,thus effectively reducing the computational This reduces the computational stress.The simulation results show that the parameters determined from the predicted values of the model proposed in this paper are the same as those determined using the actual values and are superior to other prediction models.