Raw Slurry Blending Process Back To The Chaos Analysis Of Time Series Of The Material Composition And The Svm Prediction

By taking off silicon and decomposition process, alumina returned material is one of compounding process's main raw materials which affect the slurrys'quality and yield. In order to guarantee the quality of alumina by optimizing the ingredients, components of mixing raw material is known as the prerequisite. As large fluctuation about components of Alumina returned material, seriously lagging in detection and only available to history information for testing, it is necessary to predict the time series in the components of alumina returned material.As to complex nonlinearity and no regularity existing in time series in the components of alumina returned material, this paper adopts G-P algorithm and Wolf algorithm of small amount data to respectively calculate correlation dimension and lyapunov index of time series about returned material's components. Non-integer correlation dimension and positive largest lyapunov index can recognize the chaotic characteristic of time series about silica residue composition.To make the LSSVM accurately predict the chaos time series, we improved the LS-SVM model from two respects:for one, considering that LS-SVM model's prediction effect is not ideal as to big scope and noise and we often pay more attention to the relative error in practice, so we can improve generalization function in existing LS-SVW model to get a new LSSVM forecasting model.For another, For precocious phenomena with the standard PSO algorithm and the defects of low efficiency becourse of only with experience or try computation selecting LS-SVM forecasting model parameters, first,by introducing averagegrain moment and the fitness function have improved PSO algorithm, which effectively avoid the standard PSO algorithm's the precocious problem. Then put forward two hybrid optimization algorithms based on improved PSO mixed in the Powell algorithm and HSS mixed in Powell algorithm.which reaches a better balance between computation efficiency and global optimization, it can search optimal objective function values to find the optimal parameter and realize the automatic optimization parameters.On the basis of the above studies, the discrete time series of silicon residue's components has been generalized reconstructed based on rough intensive theory and get a simplified and effective reconstructed phase space, which severs as the inputs of model and we adopts the proposed novel LS-SVM model to predict and simulate the components. The results illustrate the effectiveness of the model.