Investigation Of Artificial Neural Networks And Its Combination With Wavelet Transform For Oscillograghic Determination

In recent years, oscillographic analysis has obtained great progress on both the method principle and measure mode. Theories of oscillographic potentiometry and oscillographic chronopotentiometry (OCPY) are tending to maturity gradually. Oscillographic titration has become a new field of oscillographic analysis. At aspects of oscillographic chronopotentiometric determination (OCPD), the main research contents are to enhance signal-noise-ratio, to improve reproducibility and resolution of oscillograms as well as to actualize automation of OCPD for micro or trace components. In this paper, Neural Network and Combined Wavelet Transform with Neural Network for OCPD are studied systematically. It also eliminated the artificial influence in oscillographic determination, enhanced the automatization degree of OCPD. This thesis consists of four chapters, and the author's contributions are as follows: In recent years, oscillographic analysis has a quiet. The theories of oscillographic potentiometry and oscillographic chronopotentiometry have grown up gradually, oscillographic titration has become a novel area in titration analysis. In aspect of oscillographic determination, it has become the mostly matter in study how to enhance the signal-to-noise ratio (S/N) of oscillographic signal, improve the resolution and reproducibility, and realize the auto-determination of micro and trace component. In this paper, the neural network and its combination with wavelet using oscillographic determination have been studied systematically, eliminated the artificial influence in oscillographic determination, enhanced the automatization degree of oscillographic determination. This thesis consists of four chapters, and the author's contributions are as follows: 1. After noise and background in multi-order differential OCPY signals are deduced with wavelet transform, the error back propagation neural network (BPNN) is used in OCPD. 1． The characteristic of denoiseing and of wavelet transform were used in oscillographic chronopotentiometric signal processing, multi-order differential oscillographic chronopotentiometry determination. Take prediction of Pb2+ concentration in 0.5 moll/L NH4Ac as an example, the effect of the number of node of hidden layer, the learning rate factor and the momentum factor on the prediction result have been discussed. Result shows that errors caused by noise, background and handwork measure have been removed basically, and the analytical rate and the reliability of analytical results have been enhanced.2．Resilient propagation neural network (RPNN) is firstly applied in determination of trace chromium in the passivation solution of copper foil with oscillographic chronopotentiometry. The oscillographic specialty and determination condition of chromium have been studied, and the influence of the number of network layer﹑node of each layer and other network parameter have been discussed. The concentration of Cr(Ⅲ) in the passivation solution of copper foil was 7.383×10-3mol/L, the average recovery was 103.2%, and the method detection limit for Cr(Ⅲ) was 8.0×10-8mol/L . Compared with application of standard BP neural network, RPNN eliminated the over-fitting phenomenon in training, and enhanced the accuracy of prediction.3．Dyadic Wavelet Neural Network,WNN4. Five theses have been accomplished by author during the study of master's degree.