摘要
针对企业对丁苯橡胶聚合转化率需在线高精度预测的需求,考虑实际工况复杂性和偏最小二乘(PLS)算法非线性处理能力的不足,分别建立了引入单核和混合核函数的丁苯橡胶聚合转化率PLS预测模型。工业数据仿真结果表明:基于核的PLS模型均可满足企业生产对预测精度的要求,即聚合转化率预测绝对误差大于1.5的比例不大于样本总数的10%,尤其是混合核PLS因兼有局部和全局特性,表现出更优的性能。
Aiming at the enterprise demand that polymerization conversion rate needs to be online high precision predicted for styrene butadiene rubber （SBR） , considering the complexity of actual working condition and the disadvantages of partial least squares （PLS）algorithm for its nonlinear processing power, PLS models with single or mixed kernel function are created separately and used to forecast the SBR polymerization conversion rate. The simulation results show that Kernel PLS models can all meet the enterprise requirements, which mean the ratio of polymerization conversion rate prediction absolute error greater than 1.5 is not more than 10 % of the total samples. Especially, the mixed Kernel PLS model shows more excellent properties because of its both local and global characteristics.
出处
《传感器与微系统》
CSCD
北大核心
2012年第3期143-146,共4页
Transducer and Microsystem Technologies
基金
国家自然科学基金资助项目(60964003)
关键词
偏最小二乘
核偏最小二乘
丁苯橡胶
聚合转化率
partial least squares （ PLS ） kernel PLS styrene butadiene rubber （ SBR ） polymerizationconversion rate
