基于小波分析和神经网络的聚丙烯产品质量预测

Melt Index and Isotactic Index Prediction of Polypropylene by Neural Networks Combined with Wavelet Analysis

建立了基于小波变换和BP神经网络的温度和压力信号模型,用于聚丙烯产品质量预测模型。聚合前期的升温过程对最终产品质量影响大,第一个模型运用db4小波对前期聚合温度(压力)信号进行变换,将所得的部分系数作为神经网络输入参数,研究了聚合前期温度(压力)与熔融指数的关系。第二个模型以分解信号的功率频谱作为部分输入,预测聚丙烯的等规度。两个模型仿真结果表明,用温度变换结果作为输入比压力作为输入具有优势,熔融指数的相对误差为2.91%,等规度预测值的相对误差最小为1.0%,效果比较令人满意。

The temperature and pressure signal models basing the wavelet transform and BP neural networks were established to predict the quality of Polypropylene （PP）. The process of rising temperature was a great important role for the quality of PP. The polymerization temperature series or pressure series were decomposed by db4 wavelet to different frequency components as some inputs of the BP neural network to study the correlation of melt index （MI） with the temperature or pressure in the first model. The FFT power spectrum analysis of decomposited signals were used as some inputs to predicte isotactic index （liP） in the second model. The simulation of two models showed the temperature transform results had advance as input than the pressure transform results. The relative error of melt index was 2.91% and the relative error of isotactic index was 1.0 %.