基于时差处理的自适应多层次软测量建模方法

Self-Adaptive Multilevel Soft Sensor Modeling Method Based on Time Difference Processing

在污水处理过程控制中,软测量是针对难以测量变量进行有效测量的一种手段。然而,建模输入的使用限制,使一些与预测目标相关但不易获取的变量不宜作为输入信息,阻碍了建模。对此,提出一种多层次软测量建模方法。首先,利用支持向量回归建立多个软测量子模型,输出与最终目标变量相关但不易获取的子目标;然后,利用预测的子目标与原始辅助变量构造主模型的输入变量集,增加预测所需的输入信息,从而提高预测效果;同时,在建模中引入一种时差处理方法,增强模型的自适应能力,应对因外部干扰而导致的性能退化问题;最后,通过仿真案例对本文所提方法的有效性进行验证。结果表明:本文所提方法相对于单模型SVR(缺少CODe信息)和单模型SVR(具有完整CODe信息),有更好的预测表现,RMSE为0.0398,r为0.9987。

In the process control of wastewater treatment,soft sensing is an effective mean to measure the difficult variables.However,the limitation of modeling input makes some variables that are related to the target but not easy to obtain unsuitable as input information,hindering modeling.Therefore,a novel multiple-level of soft sensor modeling method is proposed.First,multiple soft sensor sub-models are built using support vector regression to output sub-objectives which are related to the final objective variable but are hard to acquire.Then,the sub-objectives and original secondary variables are constructed as the inputs of the primary model to increase the information needed for the prediction,thus improve the final prediction effect.Meanwhile,a time difference modeling method is introduced in the modeling to deal with the performance degradation caused by external interference.The proposed method is validated through a case study of simulation and real application.The results show that Compared with single model SVR(lack of coder information)and single model SVR(with complete coder information),the method proposed in this paper has better prediction performance.RMSE is 0.0398,r is 0.9987.