基于RBF网络的有源噪声控制
|
摘要
噪声对人体健康的影响已得到了广泛的关注。管道消声器已经被提出来应用于工厂、办公室等地方。被动阻尼技术是一种传统的消声技术,但是它只对高频段的噪声有较好的降噪效果,很难对低频段噪声产生作用。有源噪声控制技术正是基于此产生的一种有效的噪声控制方法。
有源噪声控制是指人为的产生次级噪声去抵消原有噪声的一种方法。有源噪声控制的基本原理是根据两列同频率、相位差固定的声波会发生相消性干涉。AANC是采用自适应方式完成次级声源控制的有源消声。AANC系统的核心是自适应滤波器和相应的自适应算法。自适应滤波器可以按某种事先设定的准则自动调节其本身的传递函数以达到所需要的输出。设计自适应滤波器时可以不必预先先知道其输入的统计特性,而且,在滤波过程中输入的统计特性如随时间作慢变化它也能自动适应。这些突出的优点使它顺理成章的被有源噪声控制研究所接纳和发展。
本文利用神经网络理论对控制器进行了优化设计,在对控制器进行设计时,选择恰当的网络模型及扩散常数,将对函数的逼近能力产生积极的影响。采用RBF(Radial Basis Function)设计的新控制器不仅克服了BP网络设计时需要预设初值的问题,而且进一步提高了一维输入一维输出的训练速度。文中利用RBF网络的任意精度逼近函数的能力对控制器的传递函数进行逼近,对输入样本进行训练从而尽可能达到理想传递函数应有的输出量。针对RBF网络的特点,用RBF网络对控制器的传递函数进行了逼近,可以取的较理想的输出效果,从而达到提高系统鲁棒性的目的。仿真结果表明,利用RBF网络对控制器进行设计,使有源噪声控制系统的鲁棒性得到了明显的改善。
Noise impact on human health has been a widespread concern. Pipeline muffler has been put forward to the factory, office, and other places. Passive damping technology is a traditional muffler technology, but only on the high frequency noise is good noise effects, it is very difficult to produce an effect of low frequency noise. Active Noise Control Technology based on this emerged as an effective noise control methods.
Active Noise Control refers to the formation of sub-human noise to offset a way of the original noise. Active Noise Control in accordance with the basic principles of the same two frequencies, the acoustic phase will be fixed in cancellation of interference. AANC is accomplished by adaptive secondary source of the active noise control. AANC the heart of the system is adaptive filter and the corresponding adaptive algorithm. Adaptive filter can be set according to certain criteria in advance automatically adjust its own transfer function to achieve the required output. Adaptive filter design can not know in advance the statistical characteristics of the input, but also in the process of filtering the input statistical characteristics such as intercropping slow change at any time it can automatically adapt to. These highlight the advantages of it was logical to accept the Active Noise Control Institute and development.
This paper uses the theory of neural network controller has been optimized design, the controller design, select the appropriate network model and the proliferation of constant, function approximation ability will have a positive impact. By RBF (Radial Basis Function), the new controller not only overcome the BP network design needs default initial value problems, and has further enhanced the one-dimensional input output training speed. RBF network, the use of arbitrary precision approximating function of the ability of a controller transfer function approximation, the importation of samples for training to the extent possible, should be the ideal transfer function output. RBF for the characteristics of the network, the controller with RBF network for the transfer function approximation, can take the better output, thereby raising the robustness of the system purposes. The simulation results show that, using the controller RBF network design, the active noise control system robustness have been significantly improved.
有源噪声控制是指人为的产生次级噪声去抵消原有噪声的一种方法。有源噪声控制的基本原理是根据两列同频率、相位差固定的声波会发生相消性干涉。AANC是采用自适应方式完成次级声源控制的有源消声。AANC系统的核心是自适应滤波器和相应的自适应算法。自适应滤波器可以按某种事先设定的准则自动调节其本身的传递函数以达到所需要的输出。设计自适应滤波器时可以不必预先先知道其输入的统计特性,而且,在滤波过程中输入的统计特性如随时间作慢变化它也能自动适应。这些突出的优点使它顺理成章的被有源噪声控制研究所接纳和发展。
本文利用神经网络理论对控制器进行了优化设计,在对控制器进行设计时,选择恰当的网络模型及扩散常数,将对函数的逼近能力产生积极的影响。采用RBF(Radial Basis Function)设计的新控制器不仅克服了BP网络设计时需要预设初值的问题,而且进一步提高了一维输入一维输出的训练速度。文中利用RBF网络的任意精度逼近函数的能力对控制器的传递函数进行逼近,对输入样本进行训练从而尽可能达到理想传递函数应有的输出量。针对RBF网络的特点,用RBF网络对控制器的传递函数进行了逼近,可以取的较理想的输出效果,从而达到提高系统鲁棒性的目的。仿真结果表明,利用RBF网络对控制器进行设计,使有源噪声控制系统的鲁棒性得到了明显的改善。
Noise impact on human health has been a widespread concern. Pipeline muffler has been put forward to the factory, office, and other places. Passive damping technology is a traditional muffler technology, but only on the high frequency noise is good noise effects, it is very difficult to produce an effect of low frequency noise. Active Noise Control Technology based on this emerged as an effective noise control methods.
Active Noise Control refers to the formation of sub-human noise to offset a way of the original noise. Active Noise Control in accordance with the basic principles of the same two frequencies, the acoustic phase will be fixed in cancellation of interference. AANC is accomplished by adaptive secondary source of the active noise control. AANC the heart of the system is adaptive filter and the corresponding adaptive algorithm. Adaptive filter can be set according to certain criteria in advance automatically adjust its own transfer function to achieve the required output. Adaptive filter design can not know in advance the statistical characteristics of the input, but also in the process of filtering the input statistical characteristics such as intercropping slow change at any time it can automatically adapt to. These highlight the advantages of it was logical to accept the Active Noise Control Institute and development.
This paper uses the theory of neural network controller has been optimized design, the controller design, select the appropriate network model and the proliferation of constant, function approximation ability will have a positive impact. By RBF (Radial Basis Function), the new controller not only overcome the BP network design needs default initial value problems, and has further enhanced the one-dimensional input output training speed. RBF network, the use of arbitrary precision approximating function of the ability of a controller transfer function approximation, the importation of samples for training to the extent possible, should be the ideal transfer function output. RBF for the characteristics of the network, the controller with RBF network for the transfer function approximation, can take the better output, thereby raising the robustness of the system purposes. The simulation results show that, using the controller RBF network design, the active noise control system robustness have been significantly improved.
引文
[1] P.Leug,P rocess of silencing sound oscilation,U. S.Patent 2 043 416,1936.
[2] S .J. Elliott and P. A.Neison,Active control of sound,IEEE Signal Processing Mag- azine,1993,10 :12 -35.
[3]陈克安,马远良著,“自适应有源噪声控制一一原理、算法及实现”,西北工业大学出版社,1993。
[4] P A .Neison,A .R .D .Curtis,S. J. Elliott,A .J. Bullmore,The active minimization of harmonic enclosed sound fields,J.Sound and Vib.,1987,Vol.117(l):1- 5.
[5] J. W .Parldns,S. D. Sommerfeldt,J. Tichy,Narrowband and broadband active contr- ol in an enclosure using the acoustic energy density,Journal of the Acoustical Society of America,2000,Vol.108(1):192- 203.
[6] H.F .Olson,E.G. May,Electronic sound absorber,J. Acoust.Soc.Am .,1953,Vol.6: 1130-1136.
[7] W.B. Conover,Fighting noise with noise,Noise Control,1956:78 -82.
[8] J.C.Burgess,Activea daptive sound control in a duct:a computer simulation,Jo- urnal of the Acoustical Society of America,1981,Vo l.70 (3):715-726.
[9] S.M .Kuo,I. Panahi,K .M .Chung,T .Horner,M .Nadesld,J. Chyan,Design of active noise control systems with the TMS320Family,Application Report of T I:DigitalSignal Processing Solutions,1996 .
[10] P.D .Wheeler,The role of noise cancelation techniques in aircrew voice commun- ication systems,Proc. R yal Aeronautial Society Symposium on Helmets and Helmet -Mounted Devices,1987.
[11] L. J. Erriksson, M. C. Allie, Use of random noise for on-line transducer modelling in an adaptive active attenuations ystem,Journal of the Acoustical Society of American, 1989,Vol.85 :797-802.
[12] C.H .Gerhold,Active control of fan-generated tone noise,AIAA Journal,1997,Vol. 35 (1):17 -22.
[13] W .Neise,G .H .Koopman,Active sources in the cutoff of centrifugal fans to reduce the blade tones at higher–order duct mode frequencies,Journal of Vibration and Ac- oustics,1991,Vol.113:123-131.
[14] G .H .Koopmann,WNeise,W .Chen,Active noise control to reduce the blade to ne noise of centrifugal fans, Journal of Vi bration,Acoustics,Stress,and Reliability in De-sign,1988,vol.110:377-383.
[15] S. J. Elliott, P. A. Nelson, I. M. Stothers, et al., In-flight experiments on the active control of propeler–induced cabin noise,Journal of Sound andVibration,1990,Vol.140: 219-238.
[16] S .J. Elliott,I. M .Stothers,P. A .Neison,et al ,The active control of engine noise in- side cars,Proc.Inter- Noise′88,1988:987-990.
[17]陈克安。有源噪声控制,北京:国防工业出版社,2003。
[18]沙家正,有源消声器近年来的发展,声学技术,1988,7(2):5-7。
[19]何振亚.自适应信号处理,科学出版社,2003,5.21-30。
[20]陈新海,李言俊,周军,自适应控制及应用,第一版,西安,西北工业大学,1998.12。
[21] SimonHaykin,郑宝译.自适应滤波器原理(第四版)[M],北京:电子工业出版社,2003。
[22] Paulo S.R.Diniz著,刘郁林等译,自适应滤波算法与实现,电子工业出版社, 2004。
[23] S.J.ELLIOTT.Signal Processing for Active Control[M].ACADEMIC PRESS.2001.
[24] Kuo, Sen M. Wu, Hsien-Tsai , Active noise control systems with adaptive nonl- inear filters , Proceedings of the IEEE International Conference on Control Applic- ations, v 2, Proceedings of the 2004 IEEE International Conference on Control Appli- cations, 2004, p 1330-1335.
[25]张翔,李传光,自适应有源噪声控制算法的研究及实现,北京理工大学学报, 2002 .2,22( 1)。
[26] BOUCHARD M. Multichannel fast afine projection algorithm for a ctive noise control[C].IEEE International Conference on Acoustics,Speech and Signal Processing. ICASSP ,Orlando,Florida,2002,2:1873-1876.
[27] SICURANZA G L, CARINI A. Filtered-X affine projection algorithm form ultich- annel active noise control using second-orde rVolterra filters[J].IEEE Signal Processing Leters, 2004,11(11);853-857.
[28]陈克安,尹雪飞,应用于多通道有源控制的自适应组合逆算法[J],信号处理, 2006, 22( 3) : 366- 369。
[29]刘金琨,智能控制,北京:电子工业出版社,2005。
[30]孙增忻,张再兴,邓志东,智能控制理论与技术.北京:清华大学出版社,2003。
[31]何玉彬,李新忠,神经网络控制技术及其应用,科学出版社,2000。
[32]魏海坤,神经网络结构设计的理论与方法,国防工业出版社,2005,2。
[33]徐丽娜,神经网络控制,电子工业出版社,2005。
[34]阎平凡,张长水,人工神经网络与模拟进化计算算,北京:清华大学出版社,2000。
[35]王洪元,史国栋,人工神经网络技术及其应用,中国石化出版社,2002,1。
[36]武妍,张立明,神经网络的泛化能力与结构化算法研究,计算机应用研究,2002,6:21-25。
[37] Lightbody G,Irwin G W ,Direct neural model reference adaptive control,IEE Proc.Control Theory Appl., 1995,142(1):31 -43.
[38] Lauret P, Fock E, and Mara TA .A Node Pruning Algorithm Based on a Fourier Amplitude Sensitivity Test Method IEEE TRW. on neural networks, 2006,17 (2):273 -293.
[39] M .Bouchard,B .Pailard,C .T. L. Dinh,Improved training of neural networks for the nonlinear active control of sound and vibration,IEEE Tans.Neural Networks,19 99,vol.10 (2):391-401.
[40]张奇志,贾永乐,周雅莉,噪声有源控制的人工神经网络方法,电声技术,2000(7):24-26。
[41]范文兵,陶振麟,张素贞.基于递推正交最小二乘的RBF网络结构优化,华东理工大学学报,2001, 27(5):503-506。
[42] Meng J F .Wu SH Q Lu JW .Face recognition with radial basis function(RBF) neural networks.IEEE Trans.Neural Networks,2002,13(3):697-710.
[43] Rubanov, N.S. The layer-wise method and the backpropagation hybrid approach to learning a feedforward neural network. IEEE Trans. Neural Networks,2000,11(2):295 -305.
[44]张兴兰,曹长修,梅彬,一种新型径向基函数神经网络学习算法,重庆大学学报,2002,25(10):56-60。
[45] Li, Haiying ,Chen, Jie Chen, Kean, Tan, Yushan, Optimal Arrangement of Error Sensors in AANC System Based on the Genetic Algorithm, Proceedings of the 3rd Wo- rld Congress on Intelligent Control and Automation,2000,2:561-563.。
[46]张菊香,邱阳,基于RBF网络的有源噪声控制,应用力学学报,2003,20(1):24 -26。
[47]郑丕谔,马艳华,RBF神经网络的递阶遗传训练新方法,控制与决策,2000 , 15 (2) : 165-168。
[48] M. O. Tokhi , R. , Wood , Active control of noise using neural net-works , Proc- eedings of the Institute of Acoustics , 1995 , 17 ( Part4) :209～216.
[49] M. O Tokhi , R. , Wood , Active noise control using multi-layered perceptron ne-ural networks , Journal of Low Frequency Noise , Vi-bration and Active Control , 1997 ,16 (2) :109～142.
[50] Muhammad Tahir Akhtar, Masahide Abe, Member, andMasayuki Kawamata, On Active Noise Control Systems With OnlineAcoustic Feedback Path Modeling,IEEE Trans .Actions on Audio Speech, and Language Procesing, 2007, 15(2):593-600.
[51] Muhammad Tahir Akhtar, Masahide Abe , and Masayuki Kawamata, A New Vari- able Step Size LMS Algorithm-Based Method for Improved Online Secondary Path Modeling in Active Noise Control Systems, IEEE Trans. Actions on Audio Speech, and Language Procesing, 2006,14(2):720-726.
[52]楼顺天,胡昌华,张伟,基于Matlab的系统分析与设计,西安:西安电子科技大学出版社,2001。
[53]刘金琨,先进PID控制MATLAB仿真,北京:电子工业出版社,2004,9。
[54]王正林,王胜开,陈国顺,MATLAB/Simulink与控制系统仿真,北京:电子工业出版社,2005,5。
[2] S .J. Elliott and P. A.Neison,Active control of sound,IEEE Signal Processing Mag- azine,1993,10 :12 -35.
[3]陈克安,马远良著,“自适应有源噪声控制一一原理、算法及实现”,西北工业大学出版社,1993。
[4] P A .Neison,A .R .D .Curtis,S. J. Elliott,A .J. Bullmore,The active minimization of harmonic enclosed sound fields,J.Sound and Vib.,1987,Vol.117(l):1- 5.
[5] J. W .Parldns,S. D. Sommerfeldt,J. Tichy,Narrowband and broadband active contr- ol in an enclosure using the acoustic energy density,Journal of the Acoustical Society of America,2000,Vol.108(1):192- 203.
[6] H.F .Olson,E.G. May,Electronic sound absorber,J. Acoust.Soc.Am .,1953,Vol.6: 1130-1136.
[7] W.B. Conover,Fighting noise with noise,Noise Control,1956:78 -82.
[8] J.C.Burgess,Activea daptive sound control in a duct:a computer simulation,Jo- urnal of the Acoustical Society of America,1981,Vo l.70 (3):715-726.
[9] S.M .Kuo,I. Panahi,K .M .Chung,T .Horner,M .Nadesld,J. Chyan,Design of active noise control systems with the TMS320Family,Application Report of T I:DigitalSignal Processing Solutions,1996 .
[10] P.D .Wheeler,The role of noise cancelation techniques in aircrew voice commun- ication systems,Proc. R yal Aeronautial Society Symposium on Helmets and Helmet -Mounted Devices,1987.
[11] L. J. Erriksson, M. C. Allie, Use of random noise for on-line transducer modelling in an adaptive active attenuations ystem,Journal of the Acoustical Society of American, 1989,Vol.85 :797-802.
[12] C.H .Gerhold,Active control of fan-generated tone noise,AIAA Journal,1997,Vol. 35 (1):17 -22.
[13] W .Neise,G .H .Koopman,Active sources in the cutoff of centrifugal fans to reduce the blade tones at higher–order duct mode frequencies,Journal of Vibration and Ac- oustics,1991,Vol.113:123-131.
[14] G .H .Koopmann,WNeise,W .Chen,Active noise control to reduce the blade to ne noise of centrifugal fans, Journal of Vi bration,Acoustics,Stress,and Reliability in De-sign,1988,vol.110:377-383.
[15] S. J. Elliott, P. A. Nelson, I. M. Stothers, et al., In-flight experiments on the active control of propeler–induced cabin noise,Journal of Sound andVibration,1990,Vol.140: 219-238.
[16] S .J. Elliott,I. M .Stothers,P. A .Neison,et al ,The active control of engine noise in- side cars,Proc.Inter- Noise′88,1988:987-990.
[17]陈克安。有源噪声控制,北京:国防工业出版社,2003。
[18]沙家正,有源消声器近年来的发展,声学技术,1988,7(2):5-7。
[19]何振亚.自适应信号处理,科学出版社,2003,5.21-30。
[20]陈新海,李言俊,周军,自适应控制及应用,第一版,西安,西北工业大学,1998.12。
[21] SimonHaykin,郑宝译.自适应滤波器原理(第四版)[M],北京:电子工业出版社,2003。
[22] Paulo S.R.Diniz著,刘郁林等译,自适应滤波算法与实现,电子工业出版社, 2004。
[23] S.J.ELLIOTT.Signal Processing for Active Control[M].ACADEMIC PRESS.2001.
[24] Kuo, Sen M. Wu, Hsien-Tsai , Active noise control systems with adaptive nonl- inear filters , Proceedings of the IEEE International Conference on Control Applic- ations, v 2, Proceedings of the 2004 IEEE International Conference on Control Appli- cations, 2004, p 1330-1335.
[25]张翔,李传光,自适应有源噪声控制算法的研究及实现,北京理工大学学报, 2002 .2,22( 1)。
[26] BOUCHARD M. Multichannel fast afine projection algorithm for a ctive noise control[C].IEEE International Conference on Acoustics,Speech and Signal Processing. ICASSP ,Orlando,Florida,2002,2:1873-1876.
[27] SICURANZA G L, CARINI A. Filtered-X affine projection algorithm form ultich- annel active noise control using second-orde rVolterra filters[J].IEEE Signal Processing Leters, 2004,11(11);853-857.
[28]陈克安,尹雪飞,应用于多通道有源控制的自适应组合逆算法[J],信号处理, 2006, 22( 3) : 366- 369。
[29]刘金琨,智能控制,北京:电子工业出版社,2005。
[30]孙增忻,张再兴,邓志东,智能控制理论与技术.北京:清华大学出版社,2003。
[31]何玉彬,李新忠,神经网络控制技术及其应用,科学出版社,2000。
[32]魏海坤,神经网络结构设计的理论与方法,国防工业出版社,2005,2。
[33]徐丽娜,神经网络控制,电子工业出版社,2005。
[34]阎平凡,张长水,人工神经网络与模拟进化计算算,北京:清华大学出版社,2000。
[35]王洪元,史国栋,人工神经网络技术及其应用,中国石化出版社,2002,1。
[36]武妍,张立明,神经网络的泛化能力与结构化算法研究,计算机应用研究,2002,6:21-25。
[37] Lightbody G,Irwin G W ,Direct neural model reference adaptive control,IEE Proc.Control Theory Appl., 1995,142(1):31 -43.
[38] Lauret P, Fock E, and Mara TA .A Node Pruning Algorithm Based on a Fourier Amplitude Sensitivity Test Method IEEE TRW. on neural networks, 2006,17 (2):273 -293.
[39] M .Bouchard,B .Pailard,C .T. L. Dinh,Improved training of neural networks for the nonlinear active control of sound and vibration,IEEE Tans.Neural Networks,19 99,vol.10 (2):391-401.
[40]张奇志,贾永乐,周雅莉,噪声有源控制的人工神经网络方法,电声技术,2000(7):24-26。
[41]范文兵,陶振麟,张素贞.基于递推正交最小二乘的RBF网络结构优化,华东理工大学学报,2001, 27(5):503-506。
[42] Meng J F .Wu SH Q Lu JW .Face recognition with radial basis function(RBF) neural networks.IEEE Trans.Neural Networks,2002,13(3):697-710.
[43] Rubanov, N.S. The layer-wise method and the backpropagation hybrid approach to learning a feedforward neural network. IEEE Trans. Neural Networks,2000,11(2):295 -305.
[44]张兴兰,曹长修,梅彬,一种新型径向基函数神经网络学习算法,重庆大学学报,2002,25(10):56-60。
[45] Li, Haiying ,Chen, Jie Chen, Kean, Tan, Yushan, Optimal Arrangement of Error Sensors in AANC System Based on the Genetic Algorithm, Proceedings of the 3rd Wo- rld Congress on Intelligent Control and Automation,2000,2:561-563.。
[46]张菊香,邱阳,基于RBF网络的有源噪声控制,应用力学学报,2003,20(1):24 -26。
[47]郑丕谔,马艳华,RBF神经网络的递阶遗传训练新方法,控制与决策,2000 , 15 (2) : 165-168。
[48] M. O. Tokhi , R. , Wood , Active control of noise using neural net-works , Proc- eedings of the Institute of Acoustics , 1995 , 17 ( Part4) :209～216.
[49] M. O Tokhi , R. , Wood , Active noise control using multi-layered perceptron ne-ural networks , Journal of Low Frequency Noise , Vi-bration and Active Control , 1997 ,16 (2) :109～142.
[50] Muhammad Tahir Akhtar, Masahide Abe, Member, andMasayuki Kawamata, On Active Noise Control Systems With OnlineAcoustic Feedback Path Modeling,IEEE Trans .Actions on Audio Speech, and Language Procesing, 2007, 15(2):593-600.
[51] Muhammad Tahir Akhtar, Masahide Abe , and Masayuki Kawamata, A New Vari- able Step Size LMS Algorithm-Based Method for Improved Online Secondary Path Modeling in Active Noise Control Systems, IEEE Trans. Actions on Audio Speech, and Language Procesing, 2006,14(2):720-726.
[52]楼顺天,胡昌华,张伟,基于Matlab的系统分析与设计,西安:西安电子科技大学出版社,2001。
[53]刘金琨,先进PID控制MATLAB仿真,北京:电子工业出版社,2004,9。
[54]王正林,王胜开,陈国顺,MATLAB/Simulink与控制系统仿真,北京:电子工业出版社,2005,5。