水处理论文笔记（2）

题目：Explicit and interpretable nonlinear soft sensor models for inflfluent surveillance at a full-scale wastewater treatment plant

作者：Xiaodong Wang 

发表时间：2019

期刊：Journal of Process Control   SCI二区

摘要：

In wastewater treatment plants, the most adopted sensors are those with the properties of low cost and

fast response. Soft sensors are alternative solutions to the hardware sensor for online monitoring of hard

to-measure variables, such as chemical oxygen demand (COD) and total phosphorus (TP). The purpose

of this study is to obtain a modelling approach which is able to identify the nonlinearity of inflfluent and

explain the correlation of inputs-outputs. Thus, the variation of inflfluent characteristics was investigated

at the fifirst stage, which provided the basis to build global and local multiple linear regression models.

Secondly, a nonlinear modelling tool multivariate adaptive regression splines (MARS) was applied for

inflfluent COD and TP prediction. Satisfactory prediction accuracy was obtained in terms of root

mean square error (RMSE) and R2. Unlike other machine learning techniques which are

¨ black box ¨models, MARS provided interpretable models which explained the nonlinearity and correlation of inputs-outputs. The

MARS models can be used not only for prediction, but also to provide insight of inflfluent variation.

 

关键词：Multiple linear regression  Multivariate adaptive regression splines  MARS   Nonlinear model  Soft sensorWastewater treatment plant

方法：利用多元自适应回归算法预测COD和TP值

结果：

The online monitoring of inflfluent wastewater characteristics is

essential for wastewater treatment process surveillance and con

trol. Soft sensor is an alternative solution for online measurement

of COD and total phosphorus (TP) in an economic manner. In this

study, we investigated the possibility of using easy-to-measure

variables as predictors to construct both global and local soft sensor

models for COD and TP prediction. The goal is to build interpretable

nonlinear models to serve as soft sensors for the surveillance of

wastewater treatment process.

 

The global MLR models performed similar to the MARS models

in terms of R 2 for COD and TP prediction. However, the RMSEs of

MARS models were smaller than that of the corresponding MLRs.

MARS has the advantage of capturing nonlinearity in flfluctuating

situations.

 

Compared with other “black box” modeling techniques, such as

neural network, useful information and knowledge can be retrieved

from MARS models. The MARS models indicated the points where

signifificant changes happened. Moreover, splines may also suggest

the number of groups for pre-classifification of the dataset.