【摘要】 针对采用关联向量机进行软测量建模所存在的多输出建模问题,提出了一种鱼群优化算法(AFSA)—多输出关联向量机(MVRVM)软测量建模方法。通过加权组合全局性Poly核函数和局部性Gauss核函数,形成混合核函数多输出关联向量机模型,有效融合多特征数据信息;然后采用鱼群优化算法对多输出关联向量机模型的相关核参数进行优化,以进一步改善模型的输出精度和稳定性。将该建模方法应用于甲醇制烯烃生产过程(MTO)反应器出口乙烯和丙烯(简称双烯)收率软测量研究中,结果表明:采用该建模方法所建立的软测量模型能有效预测双烯收率变化,具有较高的预测精度和稳定性,这可为复杂化工过程多参数监测与控制提供有力的技术方法支持。更多还原

【Abstract】 Aiming at the multi-output problem existing in soft sensor model with relevance vector machine, a modeling method based on AFSA-MVRVM was proposed. Firstly, the hybrid kernel function was built by weighted combination of global Poly kernel function and local Gauss kernel function to realize the effective fusion of multi-feature data information. Secondly, the MVRVM based on the hybrid kernel function was used to establish soft sensor model. Finally, the parameters of MVRVM model were optimized by AFSA to further improve the model prediction accuracy and stability. The presented modeling method was applied to develop soft sensor of ethylene-and-propylene yield in the methanol to olefin(MTO) reaction process. The experimental results indicated that the soft sensor of ethylene-and-propylene yield based on the above modeling method had good prediction precision and stability, which provide technical support for the monitoring and control of multiple quality indexes in the complex chemical process.更多还原