This thesis focuses on subspace based data-driven modeling and control techniques for batch and continuous processes. Motivated by the increasing amount of process data, data-driven modeling approaches have become more popular. These approaches are better in comparison to first-principles models due to their ability to capture true process dynamics. However, data-driven models rely solely on mathematical correlations and are subject to overfitting. As such, applying first-principles based constraints to the subspace model can lead to better predictions and subsequently better control. This thesis demonstrates that the addition of process gain constraints leads to a more accurate constrained model. In addition, this thesis also shows that using the constrained model in a model predictive control (MPC) algorithm allows the system to reach desired setpoints faster. The novel MPC algorithm described in this thesis is specially designed as a quadratic program to include a feedthrough matrix. This is traditionally ignored in industry however this thesis portrays that its inclusion leads to more accurate process control. Given the importance of accurate process data during model identification, the missing data problem is another area that needs improvement. There are two main scenarios with missing data: infrequent sampling/ sensor errors and quality variables. In the infrequent sampling case, data points are missing in set intervals and so correlating between different batches is not possible as the data is missing in the same place everywhere. The quality variable case is different in that quality measurements require additional expensive test making them unavailable for over 90\% of the observations at the regular sampling frequency. This thesis presents a novel subspace approach using partial least squares and principal component analysis to identify a subspace model. This algorithm is used to solve each case of missing data in both simulation (polymethyl methacrylate) and industrial (bioreactor) processes with improved performance. / Dissertation / Doctor of Philosophy (PhD) / An important consideration of chemical processes is the maximization of production and product quality. To that end developing an accurate controller is necessary to avoid wasting resources and off-spec products. All advance process control approaches rely on the accuracy of the process model, therefore, it is important to identify the best model. This thesis presents two novel subspace based modeling approaches the first using first principles based constraints and the second handling missing data approaches. These models are then applied to a modified state space model with a predictive control strategy to show that the improved models lead to improved control. The approaches in this work are tested on both simulation (polymethyl methacrylate) and industrial (bioreactor) processes.

No description available.

This thesis is concerned with subspace identification and its applications for controller performance assessment and process modeling from closed-loop data. A joint input-output closed-loop subspace identification method is developed which provides consistent estimation of the subspace matrices and the noise covariance matrix required for the LQG benchmark curve estimation. Subspace LQG benchmark is also used for performance assessment of the cascade supervisory-regulatory control systems. Three possible scenarios for LQG control design and performance improvement are discussed for this structure. A closed-loop subspace identification method is also provided for estimation of the subspace matrices necessary for performance assessment. A method of direct step model estimation from closed-loop data is provided using subspace identification. The variance calculation required for this purpose can be performed using the proposed method. The variances are used for weighted averaging on the estimated Markov parameters to attenuate the noise influence on the final step response estimation. / Process Control

<p>Inom processindustrin finns en etablerad reglerhierarki där basreglering sker med PID-regulatorer och där avancerad, multivariabel styrning sköts av MPC-programvara. Steget mellan dessa två nivåer kan upplevas som stort. För mindre och snabba multivariabla processer undvikes helst en multivariabel ansats, med försämrad reglering som följd. På Preem AB har detta upplevts som ett problem. Syftet med examensarbetet har varit att utveckla en alternativ, multivariabel styrstrategi för en process med ett mindre antal reglerstorheter som interagerar. Detta har gjorts genom en utveckling av en LQG-regulator i styrsystemet DeltaV.</p><p>För att implementera en regulator i ett styrsystem måste hänsyn tas till en rad faktorer, såsom hantering av olika körlägen, bortfall av signaler, integratoruppvridning, kommunikation med slavregulatorer och inte minst operatörernas gränssnitt för hantering av regulatorn. Att sedan utveckla en regulator för en process kräver bland annat stegförsök, analys och anpassning av stegtestdata, modellidentifiering, framtagning av trimningskonstanter, testning av styrstrategi i simulerad miljö och idrifttagning. Den typen av frågeställningar addresseras i rapporten.</p><p>Examensarbetet visar att det finns en plats för LQG-regulatorn i processindustrin för en viss typ av problem. Den utvecklade regulatorn har implementerats på en avsvavlingsprocess på Preems oljeraffenaderi i Lysekil med lyckat resultat. Oscillationer i processen, som tidvis påverkat produktionen av propen, har kunnat reduceras.</p> / <p>Process control in process industry is done in a hierarchy in which PID controllers are used for basic control and MPC software is used for advanced, multivariable process control. The implementation of multivariable control using MPC software is a major undertaking and development of such controllers for small and fast multivariable processes is therefore avoided. To achieve better control for such processes, a simpler approach to multivariable control is often sought. The purpose of this masters thesis is to develop an alternative, multivariable control strategy for processes with a smaller number of interacting control variables. This is achieved by developing an LQG-controller in the DCS DeltaV at Preem AB.</p><p>Implementation of such a controller in a DCS requires that consideration is given to a number of factors, including handling of different modes, loss of signals, reset windup, communication with slave controllers and construction of operator interface. To develop a controller for a specific process also requires step testing, model identification, tuning of the controller parameters, simulation of the control strategy and commissioning. Solutions to such issues are addressed in this report.</p><p>The thesis shows that  LQG-controllers can be useful in process industry for some niche applications. The LQG-controller has successfully been applied to a desulphurisation process at Preem's oil refinery in Lysekil, where oscillations affecting the production of propylene have been reduced.</p>

In system identification, the choice of model structure is important and it is sometimes desirable to use a flexible model structure that is able to approximate a wide range of systems. One such model structure is the Wiener class of systems, that is, systems where the input enters a linear time-invariant subsystem followed by a time-invariant nonlinearity. Given a sequence of input and output pairs, the system identification problem is often formulated as the minimization of the mean-square prediction error. Here, the prediction error has a nonlinear dependence on the parameters of the linear subsystem and the nonlinearity. Unfortunately, this formulation of the estimation problem is often nonconvex, with several local minima, and it is therefore difficult to guarantee that a local search algorithm will be able to find the global optimum. In the first part of this thesis, we consider the application of dimension reduction methods to the problem of estimating the impulse response of the linear part of a system in the Wiener class. For example, by applying the inverse regression approach to dimension reduction, the impulse response estimation problem can be cast as a principal components problem, where the reformulation is based on simple nonparametric estimates of certain conditional moments. The inverse regression approach can be shown to be consistent under restrictions on the distribution of the input signal provided that the true linear subsystem has a finite impulse response. Furthermore, a forward approach to dimension reduction is also considered, where the time-invariant nonlinearity is approximated by a local linear model. In this setting, the impulse response estimation problem can be posed as a rank-reduced linear least-squares problem and a convex relaxation can be derived. Thereafter, we consider the extension of the subspace identification approach to include linear time-invariant rational models. It turns out that only minor structural modifications are needed and already available implementations can be used. Furthermore, other a priori information regarding the structure of the system can incorporated, including a certain class of linear gray-box structures. The proposed extension is not restricted to the discrete-time case and can be used to estimate continuous-time models. The final topic in this thesis is the estimation of discrete-time models containing polynomial nonlinearities. In the continuous-time case, a constructive algorithm based on differential algebra has previously been used to prove that such model structures are globally identifiable if and only if they can be written as a linear regression model. Thus, if we are able to transform the nonlinear model structure into a linear regression model, the parameter estimation problem can be solved with standard methods. Motivated by the above and the fact that most system identification problems involve sampled data, a discrete-time version of the algorithm is developed. This algorithm is closely related to the continuous-time version and enables the handling of noise signals without differentiations.

Inom processindustrin finns en etablerad reglerhierarki där basreglering sker med PID-regulatorer och där avancerad, multivariabel styrning sköts av MPC-programvara. Steget mellan dessa två nivåer kan upplevas som stort. För mindre och snabba multivariabla processer undvikes helst en multivariabel ansats, med försämrad reglering som följd. På Preem AB har detta upplevts som ett problem. Syftet med examensarbetet har varit att utveckla en alternativ, multivariabel styrstrategi för en process med ett mindre antal reglerstorheter som interagerar. Detta har gjorts genom en utveckling av en LQG-regulator i styrsystemet DeltaV. För att implementera en regulator i ett styrsystem måste hänsyn tas till en rad faktorer, såsom hantering av olika körlägen, bortfall av signaler, integratoruppvridning, kommunikation med slavregulatorer och inte minst operatörernas gränssnitt för hantering av regulatorn. Att sedan utveckla en regulator för en process kräver bland annat stegförsök, analys och anpassning av stegtestdata, modellidentifiering, framtagning av trimningskonstanter, testning av styrstrategi i simulerad miljö och idrifttagning. Den typen av frågeställningar addresseras i rapporten. Examensarbetet visar att det finns en plats för LQG-regulatorn i processindustrin för en viss typ av problem. Den utvecklade regulatorn har implementerats på en avsvavlingsprocess på Preems oljeraffenaderi i Lysekil med lyckat resultat. Oscillationer i processen, som tidvis påverkat produktionen av propen, har kunnat reduceras. / Process control in process industry is done in a hierarchy in which PID controllers are used for basic control and MPC software is used for advanced, multivariable process control. The implementation of multivariable control using MPC software is a major undertaking and development of such controllers for small and fast multivariable processes is therefore avoided. To achieve better control for such processes, a simpler approach to multivariable control is often sought. The purpose of this masters thesis is to develop an alternative, multivariable control strategy for processes with a smaller number of interacting control variables. This is achieved by developing an LQG-controller in the DCS DeltaV at Preem AB. Implementation of such a controller in a DCS requires that consideration is given to a number of factors, including handling of different modes, loss of signals, reset windup, communication with slave controllers and construction of operator interface. To develop a controller for a specific process also requires step testing, model identification, tuning of the controller parameters, simulation of the control strategy and commissioning. Solutions to such issues are addressed in this report. The thesis shows that  LQG-controllers can be useful in process industry for some niche applications. The LQG-controller has successfully been applied to a desulphurisation process at Preem's oil refinery in Lysekil, where oscillations affecting the production of propylene have been reduced.

The prevalence of batch and batch-like operations, in conjunction with the continued resurgence of artificial intelligence techniques for clustering and classification applications, has increasingly motivated the exploration of the applicability of deep learning for modeling and feedback control of batch and batch-like processes. To this end, the present study seeks to evaluate the viability of artificial intelligence in general, and neural networks in particular, toward process modeling and control via a case study. Nonlinear autoregressive with exogeneous input (NARX) networks are evaluated in comparison with subspace models within the framework of model-based control. A batch polymethyl methacrylate (PMMA) polymerization process is chosen as a simulation test-bed. Subspace-based state-space models and NARX networks identified for the process are first compared for their predictive power. The identified models are then implemented in model predictive control (MPC) to compare the control performance for both modeling approaches. The comparative analysis reveals that the state-space models performed better than NARX networks in predictive power and control performance. Moreover, the NARX networks were found to be less versatile than state-space models in adapting to new process operation. The results of the study indicate that further research is needed before neural networks may become readily applicable for the feedback control of batch processes. / Thesis / Master of Applied Science (MASc)

No description available.

One of the key challenges in the Oil & Gas industry is to best manage reservoirs under different conditions, constrained by production rates based on various economic scenarios, in order to meet energy demands and maximize profit. To address the energy demand challenges, a transformation in the paradigm of the utilization of "real-time" data has to be brought to bear, as one changes from a static decision making to a dynamical and data-driven management of production in conjunction with real-time risk assessment. The use of modern methods of computational modeling and simulation may be the only means to account for the two major tasks involved in this paradigm shift: (1) large-scale computations; and (2) efficient utilization of the deluge of data streams. Recently, history matching and optimization were brought together in the oil industry into an integrated and more structured approach called optimal closed-loop reservoir management. Closed-loop control algorithms have already been applied extensively in other engineering fields, including aerospace, mechanical, electrical and chemical engineering. However, their applications to porous media flow, such as - in the current practices and improvements in oil and gas recovery, in aquifer management, in bio-landfill optimization, and in CO2 sequestration have been minimal due to the large-scale nature of existing problems that generate complex models for controller design and real-time implementation. Their applicability to a realistic field is also an open topic because of the large-scale nature of existing problems that generate complex models for controller design and real-time implementation, hindering its applicability. Basically, three sources of high-dimensionality can be identified from the underlying reservoir models: size of parameter space, size of state space, and the number of scenarios or realizations necessary to account for uncertainty. In this paper we will address type problem of high dimensionality by focusing on the mitigation of the size of the state-space models by means of model-order reduction techniques in a systems framework. We will show how one can obtain accurate reduced order models which are amenable to fast implementations in the closed-loop framework .The research will focus on System Identification (System-ID) (Jansen, 2009) and Model Predictive Control (MPC) (Gildin, 2008) to serve this purpose. A mathematical treatment of System-ID and MPC as applied to reservoir simulation will be presented. Linear MPC would be studied on two specific reservoir models after generating low-order reservoir models using System-ID methods. All the comparisons are provided from a set of realistic simulations using the commercial reservoir simulator called Eclipse. With the improvements in oil recovery and reductions in water production effectively for both the cases that were considered, we could reinforce our stance in proposing the implementation of MPC and System-ID towards the ultimate goal of "real-time" production optimization.

No description available.