论文部分内容阅读
聚丙烯装置熔融指数软测量中存在非线性和多工况切换操作的问题,针对普通的自适应仿射传播聚类查找最优聚类耗时较长的问题,提出一种改进的自适应仿射传播聚类,通过计算偏向参数范围缩小搜索空间,提高聚类速度和精度。多模型建模方法通常比单一模型建模方法适用范围更广、效果更佳,小波核函数不仅具有非线性映射的特征而且也具有小波分析对非平稳信号的逐级精细描述的特征,本文提出将小波核函数与正交最小二乘法相结合的方法分别对数据子集建立模型,粒子群算法实现对参数的选择,能够以较高的精度逼近函数,并通过开关切换方式根据当前工作点所属子类模型进行预测输出。通过对聚丙烯熔融指数的软测量建模研究表明,本文提出的方法具有良好的回归精度和较好的泛化性能。
The problem of non-linearity and multi-conditions switching operation in the melt index soft measurement of polypropylene plant exists. Aiming at the problem that the average length of the adaptive clustering algorithm is longer than that of the conventional adaptive affine propagation clustering, an improved adaptive imitation Spread propagation clustering, reduce the search space by calculating the bias parameter range, improve the clustering speed and accuracy. Multi-model modeling method is usually broader than the single model modeling method, the effect is better, the wavelet kernel not only has the characteristics of non-linear mapping, but also has the characteristics of wavelet analysis of non-stationary signals described in detail step by step, this paper presents The wavelet kernel function and the orthogonal least square method are combined to establish a model for the data subset respectively. Particle swarm optimization algorithm can select the parameters, and can approximate the function with high accuracy. Based on the switching mode, Subclass model to predict output. Through the soft measurement of polypropylene melt index modeling studies show that the proposed method has good regression accuracy and good generalization performance.