Ivan Mohler

SOFT SENSORS FOR SPLITTER PRODUCT PROPERTY ESTIMATION IN CDU

Soft sensor application for properties estimation of splitter bottom product in a crude distillation unit (CDU) is investigated. Based on continuous temperature, pressure, and flow measurements, two soft sensors are developed as estimators of the initial boiling point and end boiling point of splitter product. Soft sensor models are developed using multiple regression techniques and neural networks. After performing multiple linear regression analysis, it was concluded that linear models are not sufficiently accurate for the implementation in the real plant. Within multilayer perceptron (MLP) and radial basis function (RBF) neural networks, different learning algorithms are used (back propagation with variations of learning rate and momentum, conjugate gradient descent, Levenberg-Marquardt) as well as pruning and Weigend regularization techniques. Statistics and sensitivity analysis are provided for both models. Two developed soft sensors will be used as on-line estimators of heavy naphtha properties and for control purposes.

Development of Soft Sensors for Crude Distillation Unit Control

Abstract Soft sensors for distillation end point (D95) on-line estimation in crude distillation unit (CDU) are developed. Experimental data are acquired from the refinery distributed control system (DCS) and include on-line available continuously measured variables and laboratory assays. Soft sensors are developed using different linear and nonlinear identification methods. Additional laboratory data for model identification are generated by Multivariate Adaptive Regression Splines (MARSplines). The models are evaluated based on Route Mean Square (RMS), Absolute Error (AE), FIT and Final Prediction Error (FPE) criteria. The best results are achieved with Box Jenkins (BJ), Output Error (OE) and Hammerstein–Wiener (HW) model. Based on developed soft sensors it is possible to estimate fuel properties in continuous manner and apply inferential control. By real plant application of developed soft sensors considerable savings could be expected, as well as compliance with strict law regulations for product quality specifications.

Soft sensors model optimization and application for the refinery real-time prediction of toluene content

ABSTRACT Industrial facilities nowadays show an increasing need for continuous measurements, monitoring and controlling many process variables. The on-line process analyzers, being the key indicators of process and product quality, are often unavailable or malfunction. This paper describes development of soft sensor models based on the real plant data that could replace an on-line analyzer when it is unavailable, or to monitor and diagnose an analyzer’s performance. Soft sensors for continuous toluene content estimation based on the real aromatic plant data are developed. The autoregressive model with exogenous inputs, output error, the nonlinear autoregressive model consisted of exogenous inputs and Hammerstein–Wiener models were developed. In case of complex real-plant processes a large number of model regressors and coefficients need to be optimized. To overcome an exhaustive trial-and-error procedure of optimal model regressor order determination, differential evolution optimization method is applied. In general, the proposed approach could be, of interest for the development of dynamic polynomial identification models. The performance of the models are validated on the real-plant data.

Development of inferential models for fractionation reformate unit

Industrial facilities show an increasing need for continuous measurement and monitoring a large number of process variables due to strict product quality requirements, environmental laws and for advanced process control application. On-line analyzers typically suffer from long measurement delays not desirable in continuous control. Suitable alternative are soft sensors and inferential control. In this paper the development of soft sensor models for the estimation of light reformate benzene content is carried out. Linear dynamical autoregressive model with external inputs (ARX), autoregressive moving average model with exogenous inputs (ARMAX) and Box-Jenkins (BJ) models are developed. For the regression vector optimization usually performed by trial and error, Genetic Algorithm (GA) and Simulated Annealing (SA) methods have been applied. The results indicate that the GA and SA as global optimization methods are suitable for the regressor order estimation of linear dynamical models with multiple inputs. Based on developed soft sensors, it is possible to apply advanced process control schemes.

Primjena naprednog vođenja i optimiranja regulacije u industrijskim postrojenjima

This paper describes application of the new method and tool for system identification and PID tuning/advanced process control (APC) optimization using the new 3G (geometric, gradient, gravity) optimization method. It helps to design and implement control schemes directly inside the distributed control system (DCS) or programmable logic controller (PLC). Also, the algorithm helps to identify process dynamics in closed-loop mode, optimizes controller parameters, and helps to develop adaptive control and model-based control (MBC). Application of the new 3G algorithm for designing and implementing APC schemes is presented. Optimization of primary and advanced control schemes stabilizes the process and allows the plant to run closer to process, equipment and economic constraints. This increases production rates, minimizes operating costs and improves product quality.