Adaptive soft sensor for online prediction based on moving window Gaussian process regression

Traditional soft sensors may be ill-suited for batch processes because they cannot efficiently handle process nonlinearity and/or time-varying changes as well as provide the prediction uncertainty. Therefore, a novel adaptive soft sensor, referred to as online ensemble Gaussian process regression (OEGPR), is proposed for nonlinear time-varying batch processes. The batch process is first divided into multiple local domains using a just-in-time localization procedure, which is equipped with a probabilistic analysis mechanism to detect and remove the redundant local domains. Then the localized GPR models and probabilistic data descriptors (PDD) are built for all isolated domains. Using Bayesian inference, the posterior probabilities of any test sample with respect to different local domains are estimated and set as the adaptive mixture weights of local predictions. Further, the overall mean and variance of the predictive distribution of the target variable are estimated via the finite mixture mechanism. Additionally, the OEGPR method performs adaptation at two levels to handle time-varying behavior: (i) local GPR and PDD models; and (ii) the mixture weights. The effectiveness of the OEGPR approach is demonstrated through a simulated fed-batch penicillin fermentation process as well as an industrial fed-batch chlortetracycline fermentation process.

show abstract

“…63 Correspondingly, the prediction mean also requires the reverse scaling to obtain the final estimation.…”

Section: 42mentioning

confidence: 99%

Adaptive Soft Sensor Development Based on Online Ensemble Gaussian Process Regression for Nonlinear Time-Varying Batch Processes

Jin

Chen

Wang

et al. 2015

Ind. Eng. Chem. Res.

View full text Add to dashboard Cite

show abstract

“…No matter the window of data slides or expands along the samples, the mean value of the output variable in the window should be updated in order to make the window be rescaled to zero mean again [18].…”

Section: Adaptive Gaussian Process Regressionmentioning

confidence: 99%

“…Ni et al proposed a GPR model with accumulated data (including old and new), which is adaptive by using forgetting factor and predictive offset smoother [17]. Grbić et al proposed a moving window (MW) GPR and recursively updated the covariance kernel function [18], and they also extended this method to the Gaussian mixture model (GMM) including combination weights and local GPR [19]. Zhou et al developed an adaptive GPR model and used the predictive error for quality monitoring in fed-batch processes [20].…”

Section: Introductionmentioning

confidence: 99%

Model Calibration Method for Soft Sensors Using Adaptive Gaussian Process Regression

Guo

Pan

et al. 2019

IEEE Access

View full text Add to dashboard Cite

The recursive Gaussian process regression (RGPR) is a popular calibrating method to make the developed soft sensor adapt to the new working condition. Most of existing RGPR models are on the assumption that hyperparameters in the covariance function are fixed during the model calibration. In order to improve the adaptive ability of the RGPR model, hyperparameters in covariance of Gaussian process regression (GPR) are adjusted in parallel by referencing the previous optimization. The matrix inversion formula is selectively used for updating the regression model. And a dynamic offset smoother is presented to further improve the reliability of the proposed method. Applications to a numerical simulation and the penicillin fermentation process evaluate the performance of the proposed method.

show abstract

“…In particular, it is difficult to build precise firstprinciple models that can explain why defects appear in products. This is a critical issue since product life cycles are getting shorter and the time available for improving of product quality and yield requires fast and adaptive solutions [12].…”

Section: Developmentmentioning

confidence: 99%

Data-driven development and maintenance of soft-sensors

Abonyi

Farsang

Kulcsár

2014

2014 IEEE 12th International Symposium on Applied Machine Intelligence and Informatics (SAMI)

View full text Add to dashboard Cite

Product quality related process variables have significant role in advanced process control (APC). Online analyzers and software sensors can provide accurate and timely information for APC systems. In this paper we give an overview of data based soft-sensor development. We show that soft-sensor models of APC require maintenance and demonstrate that statistical quality control (SQC) techniques can be effectively used to automatize the related fault detection tasks. I. INTRODUCTIONAdvanced process control systems should have the following functionalities to ensure and continuously improve product quality [1]:• prediction of product quality from operating conditions, • optimisation of operating conditions to improve product quality, and • detection of faults or malfunctions for preventing undesirable operation. These functionalities require timely and accurate information about process variables characterizing and influencing product quality. Offline laboratory tests mostly take more than two hours which time delay can cause control problems resulting economic loss. Online analysers have faster response time (1-4 minutes) and can eliminate the dependence on laboratory data by providing timely information for corrective action or real time control (see Fig. 1) [2]. However, due to the high instrumentation and maintenance costs and low reliability of online analyzers there is a need for an easily implementable, maintainable and robust alternative.

show abstract

Adaptive soft sensor for online prediction based on moving window Gaussian process regression

Cited by 9 publications

References 30 publications

Adaptive Soft Sensor Development Based on Online Ensemble Gaussian Process Regression for Nonlinear Time-Varying Batch Processes

Adaptive Soft Sensor Development Based on Online Ensemble Gaussian Process Regression for Nonlinear Time-Varying Batch Processes

Model Calibration Method for Soft Sensors Using Adaptive Gaussian Process Regression

Data-driven development and maintenance of soft-sensors

Contact Info

Product

Resources

About