Online soft sensor design using local partial least squares models with adaptive process state partition

Abstract

We propose a soft sensing method using local partial least squares models with adaptive process state partition, referring to as the LPLS-APSP, which is capable of effectively handling time-varying characteristics and nonlinearities of processes, the two major adverse effects of common industrial processes that cause low-performance of soft sensors. In our proposed approach, statistical hypothesis testing is employed to adaptively partition the process state into the unique local model regions each consisting of certain number of consecutive-time data samples, and partial least squares is adopted to construct local models. Advantages of this adaptive strategy are that the number of local models does not need to be pre-defined and the local model set can be augmented online without retraining from scratch. Moreover, to improve the prediction accuracy, a novel online model adaptation criterion is proposed, which not only takes the current process dynamics into account, but also enables mining the information contained in the neighborhood of the query sample. The guidelines for tuning the model parameters are also presented. The LPLS-APSP scheme is applied to develop the dynamic soft sensors for a simulated continuous stirred tank reactor and a real industrial debutanizer column, and the results obtained demonstrate the effectiveness of this proposed approach, in comparison to several existing state-of-the-art methods, for online soft sensor design.

Optimization of time‐variable‐parameter model for data‐based soft sensor of industrial debutanizer

“…It indicates well performance of sensor based on the top temperature and the one-step delayed of output. Equations (10) and (11) are applied for TF modeling. The following quasilinear equation shows the minimum error and maximum prediction accuracy and more appropriate performance indexes compared to others for debutanizer…”

“…[5][6][7][8] The statistical methods or soft computing have been applied in different data-based soft sensors. The most popular of them are multivariate statistical regression techniques include multiple linear regressions (MLRs), 9 partial least squares (PLS), [10][11][12] principal component analysis (PCA) model, [13][14][15] genetic fuzzy model, 16 support vector machine method, 17 artificial neural networks (ANN), 18 a combination with PCA model and ANN, 9,19,20 a PLS-radial basis function neural network-based model, 21 and a combination with linear regressions (LRs) model and ANN. 22 The data-based model has gained the reputation by extending the availability of the recorded data in the process industries and computational power on it.…”

Online estimation for catalyst activity of acetylene hydrogenation reactor

“…In industrial processes, it is necessary to obtain some key variables that cannot be measured through online estimation for better control and optimization, [19][20][21] so the method of soft sensor is developed. In recent years, the data-driven statistical regression soft sensor has been applied in chemical processes, such as principal component analysis, [22][23][24] partial least square, [25][26][27] artificial neural network, 28,29 support vector machine, 30,31 and just-in-time learning. [32][33][34] Considering the nonlinearities and time-varying characteristics, a soft sensor development based on supervised ensemble learning with improved process state partition is proposed.…”

Small moving window calibration models for soft sensing processes with limited history

“…In this work, we present and compare the predictive accuracy of five simple soft sensing strategies for regression tasks on multivariate chemical process data, using regression models developed on small moving windows. The widely used mean moving window [16], moving window partial least squares (PLS) [17], and moving window updated recursive partial least squares (RPLS) [12], were compared with moving window random forest regression (RF), and a novel, random forest-partial least squares regression ensemble (RF-PLS) method. These modeling approaches can be used in real-time applications for one or several-time-step ahead predictions for soft sensing with little a priori knowledge and with minimal assumptions about the data.…”