Runda Jia

Real-time product quality control for batch processes based on stacked least-squares support vector regression models

Abstract A novel real-time final product quality control method for batch operations based on stacked least-squares support vector regression models (stacked LSSVR) is proposed. It combines midcourse correction (MCC) and batch-to-batch control. To enhance the model prediction accuracy and generalization capability, a stacked LSSVR approach is presented. Quality control is achieved by predicting the final product quality using stacked LSSVR models and adjusting process variables at some pre-specified decision points. Then a decision is made on whether or not control action is taken at every decision point. Once the control action is expected, the manipulated variable values are calculated and the control action is taken to bring the off-spec product quality back to the target. Then a batch-to-batch control is used to overcome the model plant mismatches and unmeasured disturbances. At last, the proposed modeling and quality control strategy is illustrated on a simulated batch reactor.

An improved correlation-based just-in-time modeling method using dynamic partial least squares and adaptive local domain partition

This paper proposes an improved correlation-based just-in-time modeling method, referring to as the ICoJIT, for improving the prediction accuracy and real-time performance of the conventional correlation-based just-in-time (CoJIT) modeling method. To achieve this objective, a novel adaptive local domain partition method has been developed based on the moving window technique and the fitting precision, which takes into account the input and output information simultaneously and has potentially the capabilities of obtaining the optimal local domain partition adaptively and capturing new process states by adding new local domains. Utilizing dynamic partial least squares and adaptive local domain partition method, multiple local domains and corresponding local models can be obtained during the offline operation stage. So online computation burden is reduced compared to CoJIT modeling method. In addition, the proposed ICoJIT modeling method can efficiently deal with nonlinearity and time-varying behavior of processes as well as the CoJIT modeling method. The effectiveness of the proposed method is demonstrated through a real industrial process dataset in sulfur recovery unit process.

Robust PLS model based product quality control strategy for solvent extraction process

A novel product quality control strategy is presented in this paper. The quality control is achieved by predicting the product quality using a data-driven model and adjusting the manipulated variables when disturbances occur in the measured variables. The data-driven model employs robust partial least squares algorithm to predict offline measured product quality, which can minimize the adverse effect of outliers in the training data set. Base on the robust regression model, the optimal control action are computed by solving a quadratic optimization problem under the constraint that the optimal projected solution must fall within the region of historical scores. The prediction and control performances are examined through a simulated solvent extraction process.

Adaptive multi-LSSVR based soft sensing for cobalt oxalate synthesis process

An adaptive multiple least squares support vector regression (multi-LSSVR) method to enhance the model prediction accuracy and generalization capability is presented. In the proposed approach, data for building single LSSVR models is re-sampled based on bootstrap techniques to form a number of sets of training and test data. For each data set, a LSSVR model is developed which are then aggregated through partial least squares (PLS). In order to identify the changes of process, an efficiently adaptive strategy based on batch-to-batch information is used. It efficiently updates a trained multi-LSSVR model by means of incremental updating and decremental pruning algorithms whenever a new batch sample is added to, or removed from the training set. The proposed method is demonstrated on a simulated batch process and utilized to develop a soft sensor model for cobalt oxalate synthesis process of hydrometallurgy. It is shown that the method can not only enhance model prediction accuracy, but also track the system drift by compared against single LSSVR method.

A nonlinear robust partial least squares method with application

This paper introduces a novel multivariate regression approach, nonlinear robust partial least squares (NRPLS), based on partial robust M-regression (PRM) with radial basis function networks (RBFNs). RBFNs are used to deal with the nonlinearity of the process. PRM is a promising linear robust regression method for tackling contaminated data, because it can efficiently eliminate the influence of outliers by appropriately chosen weighting scheme. Unlike other versions of robust PLS, NRPLS algorithm not only minimizes the adverse effect of outliers, but also characterizes the nonlinear feature. Simulation studies are performed for comparison with conventional nonlinear PLS methods. The NRPLS algorithm is also applied to cobalt hydrometallurgy extraction process. The results show superior performance compared to those methods of PLS, PRM and RBF-PLS.