Font Size: a A A

Wavelet Transform And PNN Fusion Model And Its Application In Signal Processing

Posted on:2009-07-22Degree:MasterType:Thesis
Country:ChinaCandidate:J Y LiuFull Text:PDF
GTID:2178360248453735Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
The signal to be collected and analyzed is often a noise signal synthesis that generated by system true signal and a number of noise signal sources in complex signal processing and application of practical engineering. To access useful composition of signals which reflect its features, in signal analysis and processing, the noise is removed according to the spectrum and time-scale characteristics of different signal characteristics. This paper combines theory of wavelet analysis with PNN, integrating the merit of wavelet transform with that of PNN, on one hand, using wavelet transform-frequency characteristics of local, and prominent useful signal characteristics and denoising. On the other hand, using the abilities of self-learning and adaptive for PNN processing time series data, new model and methods are provided in signal processing.The paper studied on the wavelet transform and PNN integration mechanism and new process neural networks models and methods which base on the multi-storey and multi-resolution of wavelet transform. This paper also established an integration model for pattern classification of process signal and adaptive wavelet PNN for signal prediction and continuous wavelet PNN for function approximation of process signals. The three models established have good approximation properties as wavelet has, and self-learning, adaptive and time-varying nature which PNN has. And deeply study the structure, algorithms and nature of the model.The constructed wavelet transform and PNN integrated model can obtain the family of wavelet function by wavelet decomposition to translation and telescopic transformation. Useful signals can be retained and unwanted signals can be eliminated by transforming the original sampling signals with the family of wavelet function, then the denoised signals can be disposed with PNN. In the model of adaptive wavelet PNN, combining effectively the structure model of PNN with multi-resolution analysis, a reliable theoretical basis on Wavelet Yuan and the determination of the entire network structure is offered, thereby avoiding the blindness of network structure design, At the same time we introduce translation factor and scale factor in network, and so the model structure is more freedom than wavelet decomposition. Linear distribution of network weights and convexity of learning objectives function can avoid nonlinear issues in the process of training network, such as local optimal. In the model of continuous wavelet PNN, using the nature of wavelet function it is fitted signal, thus expresses the mapping relational about system input/output signal, and realize approach of the nonlinear function.The paper establishes three kind of model structures, the high efficiency of the method is confirmed through examples. In view of the first kind of model structure, and it is applied flooding recognition in 7 oil wells of Daqing Oilfield. According to the actual data processing the test results are good.
Keywords/Search Tags:wavelet transform, PNN, wavelet process neural networks, learning algorithm, flooding recognition
PDF Full Text Request
Related items