Research On Prediction Of Sound Absorption Performance Of Acoustic Covering Layer Based On Neural Networ

With the development of industry,noise processing requirements are getting higher and higher,and various models are becoming more and more complex.Acoustic theoretical models are usually modeled and solved by theoretical methods.However,for special models in complex environments,it is difficult to construct models,and there are errors in the calculation results.The model constructed by the simulation software overcomes this problem.Through the simulation software,the complex model can be solved and the more accurate solution of the model can be obtained,but the modeling process also needs continuous debugging.In some special models,simulation calculation will require a lot of time and complex models.In some ordinary cases,different models need to be constructed when the simulation model is used to select specific acoustic structures with low requirements for the sound absorption performance of materials or when different materials need to be compared.In this case modeling adjustments will take a lot of time and require constant debugging and modification.How to quickly and accurately select a specific acoustic structure model for a specific acoustic environment is the main research direction of this paper.With the construction of big data and the development of artificial intelligence,a large number of algorithms and neural networks for data prediction have emerged in machine learning,providing new methods to overcome complex physical models.The machine learning methods commonly used in data prediction include BP neural network and random forest.In this paper,these two models and improved methods are used to predict the sound absorption performance of the acoustic cavity coating layer.The main work of this paper is as follows.The simulation software is used to construct the model.The structural parameters of the model(the parameter value of the cavity structure)and the parameters of the model material(the Young ’ s modulus,density,Poisson ’ s ratio and loss factor of the rubber)are used to construct the input parameter set by using the method of random range.The acoustic reflection coefficient of the model is used as the output set,and the sample set is constructed by using the input parameter set and the output parameter set.The sample set created in this paper can be subdivided into uniform parameter creation and random parameter creation.At the same time,random sample sets are created for oblique incidence and vertical incidence of acoustic waves.In terms of machine learning,this paper proposes three prediction models based on BP neural network,BP neural network optimized by genetic algorithm and random forest regression prediction model to predict the acoustic reflection coefficient of acoustic cavity absorption cover layer.Through the analysis results,we can get :Firstly,GA-BP neural network has the advantages of faster speed and higher accuracy than BP neural network in the case of small sample set.Secondly,in the case of small sample set,the prediction effect of uniform parameter sample set is worse than that of random parameter sample set.Random sample set can characterize the characteristics of acoustic results,and the prediction results are more consistent with the simulation results.Third,in the case of large sample set,BP neural network and GA-BP neural network are not very different.The prediction effect of random forest prediction model is better,which is basically consistent with the analytical solution.Fourthly,in the case of acoustic vertical incidence and acoustic oblique incidence,the random forest is used for prediction.The prediction results obtained show no significant difference,indicating that the prediction model is universal for the prediction of acoustic coating.Fifth: the overall prediction results are more than 95 %,indicating that the prediction results for curve fitting effect is good,has initially reached the expected standard of the study.It is proved that the prediction model is feasible for sample set prediction.