| For active noise control, currently the more used control algorithm was that based onthe linear adaptive filtering theory, such as the minimum mean square LMS algorithm.Although it was simple, with little calculations and easy to realize, the linear algorithmwould’t work when the system was nonlinear. It required the adoption of nonlinearalgorithm. BP neural network algorithm was used in place of the widely used LMSalgorithm to adjust the parameters of the adaptive filter online.BP network was the most common kind of neural network, as to nonlinear problems,it had a significant advantage and show a strong practical ability. However with its furtherresearch and application, some of its defects began to emerge, for instance, slowconvergence, longer learning time, easy to be trapped in local minimum, lake of unifiedprinciples on the building of networking, extremely sensitive to the initial setting ofweights, poor generalization ability of the network and so on. The ant colony algorithmwhich had the characteristics of global optimization, positive feedback and distributedcomputing was applied to determine the initial weights and threshold value of BP network.The results show, the proposed improved BP algorithm which based on Adaptive Max-MinAnt System had greater capacity to dealing with practical problems than the basic BPalgorithm.The characteristics of secondary channel had a great effect on the adaptive activecontrol system. The active noise control model of engine exhaust system was established,neural network adaptive inverse method was used to identify the secondary channeltransfer function offline, then it was cascaded before the secondary channel to offset theimpact on active noise control system generated by the secondary channel. On the basis ofanalysis of the generation mechanism and frequency distribution of engine exhaust,conducted a simulation experiment by MATLAB/SIMULINK, and compared the resultwith which used LMS algorithm. The results show, for nonlinear system, neural networkwas better than LMS in terms of convergence speed and convergence precision. The power spectrum of the simulation results was plotted in MATLAB, it could be seen that becauseof neural network algorithm noise signal got an offset over a wide frequency range.Because of its advantages of simple structure, good stability and so on, feed forwardsystem was the most commonly used and most mature structure. But it required thereference input of the noise, and it was often difficult to obtain noise reference input signalor the signal to noise ratio was too low in actual application, so it was still hard to apply itin practice. Therefore, it was necessary to try it in engine noise active control system.Neural network Internal Model Control system was applied on the elimination of engineexhaust noise and the particle swarm optimization algorithm was adopted to regulateonline feedback filter parameters. Still simulated in MATLAB/SIMULINK environment,the results suggested that feedback structure of active noise control could also achievegood effects.A measurement of the exhaust noise of a car was conducted, and the feed forward andfeedback model were used to deal with the measured noise, and achieved the same effect.Furth suggested that the active control method based on neural network algorithm couldbetter eliminate automotive exhaust noise, it had a certain practical ability, and had apositive role on improving the active control to automotive exhaust noise. |
PDF Full Text Request