基于稳健关联向量回归的黑液波美度软测量

针对造纸工业碱回收蒸发工段黑液波美度不易在线实时测量的现状,提出一种基于稳健关联向量回归的软测量方法,建立黑液波美度的软测量模型,实现黑液波美度的在线测量。稳健关联向量回归方法通过最大化一个建立在稳健子集上的似然函数得到模型的参数,同时保证回归的稳健性。构造了迭代算法来计算稳健关联向量回归,在优化超参数的同时,寻找稳健子集。实验结果表明,用该方法建立波美度软测量模型不仅是可行的和有效的,而且能够克服异常点的影响。

Soft sensor for measuring Baume degree of black liquor based on robust relevance vector regression

Real-time online measurement of the Baume degree of black liquor of in the alkali recover process in the paper making industry is very difficult.A soft sensor method for Baume degree was developed based on the robust relevance vector regression(RRVR) method for online measurements.The method calculates the model weights and ensures the robustness of the regression algorithm by maximizing the likelihood function defined over a robust subset which is assumed to include the principal data without outliers.The iterative regression algorithm optimizes the hyperparameters and simultaneously selects the robust subset.Test results show that the soft sensor for measuring the Baume degree of black liquor is effective but can resist outliers.