Research on Identification of Gas Flow Distribution with Wavelet Feature Extraction

A method of wavelet probabilistic neural network was established for identifying the gas flow distribution in blast furnace. It was based on the characteristics of the wavelet transform in multi-scale which can decompose the temperature signal and extracted energy feature vectors from it. Because the correlation between cross temperature in the upper part of the furnace and gas flow distribution is close, the gas flow change could be reflected indirectly by the temperature change in cross temperature in order to gain the furnace situation and get the next operation for the furnace. With MATLAB, the blast furnace cross temperature data was firstly preprocessed, and then trained and simulated by the neural network. The neural network with wavelet feature extraction has the advantages of less node numbers and simple net scale and been improved in recognition. The experiment results show that adding wavelet applied to the probabilistic neural network has a higher accuracy of recognition than the general one.