基于离散PSO的软测量辅助变量选择算法

Selection of Secondary Variables Based on Discrete PSO in Soft-sensing

工业对象的复杂化带来了可测变量的增多,这些变量集合中大量冗余的信息会降低软测量建模的精度。针对这个问题,提出了基于离散PSO的软测量辅助变量选择算法。算法将传统PSO连续的优化过程通过对粒子位置的隶属度计算,将其离散成0或1。0、1分别表示某变量未被选中和被选中,每个粒子就代表一种变量选取情况。将PLS回归用于适应度函数的计算,有利于克服多元回归中多重共线问题。最后,将该算法用在了丙烯精馏塔塔顶丙烯浓度的软测量实验中,实验结果表明该方法有效,并提高了模型的预测精度。

The complication of the industrial objects leads to a large number of measurable variables. The redundant information in the variables set may reduce the precision of soft-sensing model. To solve this problem, an algorithm based on discrete PSO was proposed to select secondary variables. Traditional optimization process of continuous PSO was divided into 0 or 1 via the membership degree calculation of the particle. The value of 0 or 1 denotes whether a variable was selected. Each particle position would represent a kind of variables selection. The PLS used in the fitness function could overcome the problem of collinearity in multivariable regression. Finally, the proposed algorithm was applied in soft-sensing modeling to predict the propylene concentration. The experiment results indicate that the proposed algorithm can improve the precision of prediction.