Recognition Of Polyethylene's Agglomeration In Fluidized-bed Reactor Based On Unbalanced Data

In order to detect the polymer agglomeration in fluidized bed reactor(FBR)and reduce the economic costs caused by the process of industrial production polyethylene agglomerate.A method of real-time monitoring of agglomeration in fluidized bed polyolefin reactor based on unbalanced sample data and voiceprint feature recognition is developed.First,the acoustic emission detection technology is applied to measure the acoustic signal generated by the air and material particles,particles and the reactor wall,air and collision between particles and fluidized bed reactor wall friction on the inner wall of FBR.Then,the voiceprint features of the collected acoustic signal from three different states(Micro agglomerate,mild agglomerate,severe agglomerate)are extracted with the Eight dimensions Mel Frequency Cepstrum Coefficients(MFCC)and the Linear Prediction Cepstrum Coefficients(LPCC).And then the signal of the 8 dimensions wavelet packet energy was analyzed,Relief algorithm and PCA algorithm were applied to the selection of feature and reduction of dimension,the data is as simple as possible under the condition of retain the most amount of information.To classify the extracted voiceprint features,an improved Adaboost algorithm is proposed to build the real-time agglomeration classification model.The improved algorithm according to the cost of samples and the base classifier discriminates results determine the degree of chaos to the importance of the samples.Due to the combination of cost factor and Gini index based decision-making calculation with the Adaboost algorithm,the proposed improved Adaboost algorithm can classify unbalanced small samples with better accuracy and F-score index than the traditional Adaboost algorithm.The comparative experiment results in a fluidized bed pilot plant have verified the effectiveness and feasibility of the proposed method.And laid a solid foundation for the industrial production of polyethylene agglomeration in the process of monitoring.