Wavelet Analysis And Neural Network Theory And Their Application To Gas-Solid Fluidized Beds

In this paper, the two -phase flow behaviors of gas-solid circulating fluidized bed have been first studied by analysing experimental data. Then, based on fractal theory wavelet package analysis(WPA) de-noising method was used to denoise solids concentration and pressure fluctuation signals. The solids concentration signals have also been investigated by wavelet multi-resolution analysis and wavelet package decompose, contributed to disclose the non-uniform flow structures of gas-solids phase. Finally, BP neural network recognition model of particulate and aggregative fluidization and RBF neural network prediction model for chaotic time series of circulating fluidized bed have been set up, which provides new methods for on-line recognition of fluidization state and control and prediction of circulating fluidized bed systems.The experimental analysis results show that the gas and solids phases local flow structure of CFB riser are composed of dilute phase(continuous phase) and dense phase( discontinuous phase, cluster). The solids holdup is lower in the center than in the wall region at the radial positions, and the solids holdup is higher in the lower section than in the upper section of the riser. It is also found that the flow developments in all radial locations are not synchronized, but it extended gradually from the core region to the wall, until finally a fully developed flow is achieved across the whole cross-section of the riser. The flow development at the annulusregion near the wall, as a control process of the whole flow development in the riser, is much slower, and more sensitive to the change of operation conditions than that at the core region.The denoising results about solids holdup and pressure fluctuation signals of gas-solids CFB riser indicate that WPA denoising method based on fractal theory can efficiently differentiate the noises from the solids holdup signals, and the denoised effect is much better than that by filter method based on the Fourier transform. The wavelet multi-analysis is adopted to further understand the microstructure of solids holdup fluctuation, The vivid fingermark images show that it is a fractal and dissipative structure.The low and frequency signals reflect the dilute phase fluctuation behavior and dense phase(cluster) fluctuation behavior, respectively. To deep understand the multi-scales characteristic of gas-solids circulating fluidization, the WP decompose methods for obtaining the eigenvalue of gas-solid fluidized bed has been developed. The results show that eigenvalues of different scales are effective for identification of non-uniform and dynamic structure of gas-solid fluidized bed.To effectively identify the particulate and aggregative fluidization of fluidized bed, BP neural network model has been developed. The results show that BP neural network model is more efficient than the traditional method. A hybrid method for prediction of solids holdup in gas-solid circulating fluidized bed is proposed based on chaos phase reconstruction and wavelet package as well as neural networks. Experimental results show that the model provides good predictions and has promising applications.The above research results have developed the research method for gas-solid fluidized bed, deep understood the gas-solid fluidization systems and contributed to a further grasp of the flow behavior of the gas and solids and actual application of fluidized bed.