用于信號(hào)逼近的自適應(yīng)時(shí)延小波神經(jīng)網(wǎng)絡(luò)
AN ADAPTIVE TIME DELAY WAVELET NEURAL NETWORK FOR SIGNAL APPROXIMATION
-
摘要: 小波神經(jīng)網(wǎng)絡(luò)是一種強(qiáng)有力的函數(shù)逼近工具。本文結(jié)合時(shí)延神經(jīng)網(wǎng)絡(luò)和小波分析概念提出一種新的小波神經(jīng)網(wǎng)絡(luò)摸型自適應(yīng)時(shí)延小波神經(jīng)網(wǎng)絡(luò)(ATDWNN:adaptive timedelay wavelet neural network).ATDWNN可以對(duì)同一類存在不同時(shí)延的多個(gè)信號(hào)用同一個(gè)超小波(superwavelet)進(jìn)行逼近。為了訓(xùn)練ATDWNN,本文提出一種基于時(shí)間機(jī)理的競(jìng)爭(zhēng)學(xué)習(xí)算法。實(shí)驗(yàn)表明,ATDWNN不僅可以成功地對(duì)同一類存在不同時(shí)延的多個(gè)信號(hào)采用同一個(gè)超小波進(jìn)行逼近,而且可以用來(lái)估計(jì)各樣本信號(hào)的時(shí)延。Abstract: Wavelet neural networks (WNN) is a powerful tool for function approximation. In this paper a new model named adaptive time delay WNN(ATDWNN) is proposed which combines time delay neural network and wavelet decomposition. ATDWNN is used to approximate signals having different time delays in the same class. In order to train ATDWNN, time mechanism based competition learning is also proposed. It is shown through experiments that ATDWNN can not only approximate signals having different time delays by the same superwavelet, but also detect these time delays successfully.
-
Hornik K, Stinchcombe M, White H. Multilayer feedforward networks are universal approximar tors. Neural Networks, 1989, NN-2(2): 359-366.[2]Par J, Sandberg I W. Universal approximation using radial-based-function networks[J].Neural Computation.1991, 3:246-257[3]Zhang Q H, Benvenisete A. Wavelet networks. IEEE Trans. on Neural networks, 1992, NN-3(6): 889-989.[4]Zhang J, Walter G G, Miao Y B, Lee W N. Wavelet neural networks for function learning[J].IEEE Trans. on Signal Processing.1995, 43(6):1485-1496[5]Kreinovich V, Sirisaengtaksin V, Cabrea S. Wavelet neural networks are optimal approximators for functions of one variable. University of Texas at EL. Paso, Computer Science Department Technical Report, 1992, No. UTEP-cs-92-29.[6]Delyon B, Juditsky A, Benveniste B. Accuracy analysis for wavelet approximation. IEEE Trans. on Neural Networks, 1995, NN-6(2): 332-348.[7]Szu H H, Telfer B, Kadambe B. Neural network adaptive wavelets for signal representation and classification[J].Optical Engineering.1992, 31(9):1907-1916[8]Waibel A, Hanazawa T, Hinton G, Shikano K, Lana K. Phone recognition using time-delay neural networks. IEEE Trans On ASSP, 1989, ASSP-37(3): 328-339. -
計(jì)量
- 文章訪問(wèn)數(shù): 2218
- HTML全文瀏覽量: 134
- PDF下載量: 381
- 被引次數(shù): 0
下載:
