Search

• Overview of the current state of art of advanced process control and optimization technologies. • Summary of AI and ML-based approaches for closed loop control and optimization. • Application of rigorous model-based optimization, machine learning and hybrid approaches to optimization and control. • Refinery-wide and network-wide optimization. • Autonomous operation – future directions and challenges. Technologies based on Artificial Intelligence (AI) and Machine Learning (ML) concepts are advancing at a rapid pace. The new paradigms are challenging the status-quo of mature automation and control technologies in industry. Autonomous operation is a frequently stated goal of AI evangelists and technology providers. This white paper gives an overview of the current state of art of advanced process control and optimization technologies. It also provides a brief summary of the AI and ML-based approaches that address the closed-loop control and optimization space. Some results from industrial implementations are shared for both conventional and AI/ML-based approaches. Experience from four industrial applications is shared, covering rigorous model-based, machine learning and hybrid approaches to real-time optimization and control problems. The applications range from unit-based control & optimization to refinery & network wide optimization. A set of high-level requirements that need to be satisfied regardless of the underlying technology for closed-loop autonomous operations is reviewed. The article concludes with some future directions and perspectives highlighting areas where the emerging technologies may have significant impact in industry. [ABSTRACT FROM AUTHOR]

In this paper, a controller is proposed to automate the injection of propofol and remifentanil during general anesthesia using bispectral index (BIS) measurement. To handle the parameter uncertainties due to inter- and intra-patient variability, an extended estimator is used coupled with a Model Predictive Controller (MPC). Two methods are considered for the estimator: the first one is a multiple extended Kalman filter (MEKF), and the second is a moving horizon estimator (MHE). The state and parameter estimations are then used in the MPC to compute the next drug rates. The methods are compared with a PID from the literature. The robustness of the controller is evaluated using Monte-Carlo simulations on a wide population, introducing uncertainties in all parts of the model. Results both on the induction and maintenance phases of anesthesia show the potential interest in using this adaptive method to handle parameter uncertainties. • Two new control methods are proposed to dose propofol and remifentanil using BIS signal during general anesthesia. • The methods estimate both drug concentration and patient sensitivity before computing optimal drug rates. • Compared to a published PID, both methods improve the overall regulation of depth of hypnosis. [ABSTRACT FROM AUTHOR]

• A model is developed by integrating the Purdue model with a kMC algorithm. • Cell wall thickness is used to predict cellulose fiber collapse phenomena. • A soft sensing system is developed to estimate unmeasurable output variables. • A model-based controller is design to regulate wood chip properties. In the Kraft pulping process, the cell wall thickness (CWT) of wood chips significantly affects paper properties. This work proposes a novel multiscale model by combining the mass continuity and thermal energy balance equations adopted from a modified "extended Purdue model" with a kinetic Monte Carlo algorithm to describe the microscopic events such as the evolution of CWT and Kappa number; otherwise, these microscopic properties of wood chips are not accessible by existing measurement techniques. To reduce the model complexity, a data-driven reduced-order model is developed, which is then used for developing a soft sensing system (i.e., Kalman filter) to estimate the primary measurements (e.g., CWT, Kappa number) by utilizing the secondary measurements (e.g., active effective alkali and dissolved lignin concentrations in the free-liquor phase). Based on this soft sensing system, a model-based feedback control framework is developed to regulate both the CWT and Kappa number of wood chips. [ABSTRACT FROM AUTHOR]

This paper provides an introduction to Reinforcement Learning (RL) technology, summarizes recent developments in this area, and discusses their potential implications for the field of process control, and more generally, of operational decision-making. The paper begins with an introduction to RL that allows an agent to learn, through trial and error, the best way to accomplish a task. We then highlight new developments in RL that have led to the recent wave of applications and media interest. A comparison of the key features of RL and mathematical programming based methods (e.g., model predictive control) is then presented to clarify their similarities and differences. This is followed by an assessment of several ways that RL technology can potentially be used in process control and operational decision applications. A final section summarizes our conclusions and lists directions for future RL research that may improve its relevance for the process systems engineering field. [ABSTRACT FROM AUTHOR]

In this paper, the application of a robust nonlinear model predictive control (NMPC) framework to mammalian cell cultures is proposed, dealing with possible large kinetic parameter uncertainties. Industrial constraints formulated in view of good manufacturing practice and quality-by-design approach are also considered, namely the assurance that all state trajectories are contained within a corridor defined by lower and upper safety bounds. The latter are assimilated to the well-known tube-based paradigm which is used to formulate the corresponding robust NMPC problem. Both classical and tube-based NMPC performances are assessed in numerical simulations where specific key-species are regulated while dealing with an uncertain plant model. The capability of the tube-based method to reduce the impact of the parameter variations on the state trajectories and the violation of the constraints is highlighted, suggesting the transfer of the method on a real pharmaceutical process. • NMPC design for a 25-state data-driven mammalian cell culture model. • Qualitative model sensitivity analysis using confidence corridors. • Robust tube-based NMPC framework application to mammalian cell culture model. • Practical guidelines for experimental validation in industrial environment. [ABSTRACT FROM AUTHOR]

Optimal control of a CSTR represents a challenging task. The proposed paper discusses two issues. The first one addresses control of pH in a chemical vessel, where the reaction between sodium hydroxide and acetic acid occurs. The objective here is to improve control performance of a well tuned PI controller. It will be shown that this can be achieved by introducing a reference governor scheme. The second problem, elaborated in this paper, is the implementation of the reference governor paradigm. Concretely, we aim to design a fast and cheap MPC-based feedback controller. To achieve these goals, we exploit the region-less explicit technique, which efficiently reduces memory footprint issues of standard explicit MPC schemes. Such MPC-based reference governor was employed to control pH in the chemical vessel. Its control performance is compared with conventional PI controller. Finally, comparison of implementation requirements of region-less and region-based explicit techniques is investigated. [ABSTRACT FROM AUTHOR]

As manufacturing processes transition towards digitalization, data-driven process control is emerging as a key area of interest in future artificial intelligence technology. A crucial aspect in implementing data-driven process control is "What should we learn from the data?". In general, the data-driven control method can be categorized into two main approaches: Learning the model and learning the value. To assist in selecting the more suitable approach, this paper applies six different control methods, with three falling under each approach, to three distinct manufacturing process systems. The simulation results indicate that the model-learning approaches display higher data efficiency and exhibit lower variance in total cost. These methods prove to be particularly advantageous for addressing the regulation problems. Conversely, value-learning approaches show competitive potential in closed-loop identification and in managing economic cost problems. The remaining challenges associated with each technique are discussed, along with practical considerations for their implementation. • What should we learn, model or controller, with the collected data? • Quantitatively compare the model-based and model-free control approaches. • Model-based methods were superior in tackling regulation problems. • Model-free methods show competitive performance in optimizing economic cost. [ABSTRACT FROM AUTHOR]

• Feasible composition/temperature cascade control for a Kaibel column is studied. • The PI-MPC scheme uses PI controllers in the lower layer and MPC in the upper layer. • A composition estimation soft sensor is introduced to improve the PI-MPC performance. Kaibel dividing-wall distillation column (KDWDC) is a very effective distillation process intensification technology. However, the misgivings of controllability problems impede its industrialization. Therefore, three feasible control schemes are studied in this paper, including the composition/temperature cascade control scheme with PI controllers in the lower layer and MPC in the upper layer (PI-MPC), the PI-MPC scheme with a composition estimator (PI-MPC-CE), and the PI scheme with pure temperature controllers. In the PI-MPC scheme, the MPC can deal with the complicated dynamics of the KDWDC, while the PI controller can quickly stabilize the KDWDC. The PI-MPC-CE scheme can continuously predict the primary controlled variables from the easy-to-access auxiliary variables. Dynamic performances prove that the PI-MPC-CE scheme are better than the PI and PI-MPC scheme with lesser number of oscillations, shorter settling times, and reduced maximum and steady-state deviations. Besides, it is very promising because it disposes of the continuous real-time composition detection. [ABSTRACT FROM AUTHOR]

Buildings are known for being great consumers of energy and for operating under On/Off and linear control strategies in common commercial packages. In order to reduce their energy footprints and to improve thermal comfort, new methodologies are needed to obtain better input profiles. This paper presents a detailed study of the implementation of a nonlinear model predictive control (NMPC) approach for a heating, ventilation, and air conditioning (HVAC) system. The HVAC system is modeled by an index-1 differential-algebraic system of equations, obtained from rigorous material and energy balances. The proposed NMPC algorithm was implemented in GAMS and compared with another approach from the literature. Similarly, the objective function is defined using one or more of the following criteria: (1) tracking of temperature and relative humidity set points; (2) maximization of thermal comfort; (3) minimization of energy consumption. To better represent the situation in a real application, the simulations include model-plant mismatch in parameters such as air infiltration into the building, the coefficient of the thermal exchange between the building and the exterior, and occupancy level. Additionally, simulations with random measurement noise have been performed. The results have shown that the proposed approach, in which the model is not linearized at any step, is able to reduce the energy consumption and maintain the Predicted Mean Vote (PMV) close to the desired set point, while the computation burden is only increased by one second per iteration, which is negligible in comparison with a sampling time of 10 min. [ABSTRACT FROM AUTHOR]

Highlights • A novel data-driven model predictive controller for uni-axial rotomolding is proposed. • Subspace identification algorithm is used to identify the process model. • Proposed approach is implemented on an experimental setup. • Results demonstrate improved quality, variability rejection and ability to achieve product. Abstract This paper addresses the problem of achieving tight product consistency and enabling automated process changes to deliver user-selected criterion based product in a complex industrial batch process such as uni-axial rotational molding. To this end, a data driven state-space model is first identified. For a given trajectory of input moves (heater and compressed air profiles), this dynamic model is able to predict the evolution of the measured variable (internal product temperature). The dynamic model is augmented with a quality model, which relates its own terminal predictions to a selection of key quality variables (sinkhole area, ultrasonic spectra amplitude, impact test metric and viscosity). The dynamic and quality models are in turn utilized within a model predictive control (MPC) framework that enables specifying product quality requirements explicitly. Experimental results demonstrate the ability of the MPC not only in achieving tight quality control but also providing on-spec product for a new specification. [ABSTRACT FROM AUTHOR]

Highlights • A dynamic reference trajectory is calculated for MPC to improve the economic performance in transient processes. • The dynamics of the system is approximated by mesh points in the state space. These points construct a graph to describe the accessibility of the mesh points and economic costs of state transition. • The dynamic reference generation problem is converted to the shortest path problem in the obtained graph. • We extract a subgraph to accelerate the computational procedure at each sampling time. Abstract In this paper, we focus on generating the reference trajectory to improve the economic performance in transient processes for model predictive control (MPC). We convert the trajectory generation problem to the shortest path problem which is one common problem in graph theory. For this purpose we construct a graph, also called the state-input graph, to approximate the dynamics of the system. The nodes of the graph are states obtained by mesh discretization of the state space and the weighted edges represent the costs when driving the system from one state to another under an admissible control input. The admissible control inputs are predetermined by mesh discretization of the input space. The graph describes the accessibility of the mesh points in the state space under control inputs in one sampling time. We run Dijkstra's algorithm with the obtained graph to find a path from the current state to the optimal point at each sampling time. Then the path is sent to MPC as a reference. The procedure is repeated at the next sampling time and the reference is updated as well. Moreover, we extract a subgraph of the whole graph at each sampling time to accelerate the procedure. A numerical example and a benchmark chemical process are employed to demonstrate the effectiveness of the proposed approach. [ABSTRACT FROM AUTHOR]

For three decades, model predictive control (MPC) has been the flagship advanced control method in the chemical process industries. However, most implementations still use heuristic methods for designing MPC estimators, especially for offset-free MPC implementations. In this paper, we present a recently developed maximum likelihood-based method for the identification of linear augmented disturbance models for use in offset-free MPC. This method provides noise covariances that are used to derive Kalman filters and moving horizon estimators, forgoing the need for manual design and tuning of the estimator. The method is extended to handle closed-loop plant data. The proposed identification method and estimator design are evaluated in industrial-scale, real-world case study of a process at Eastman Chemical's Kingsport plant. Using this identified model, we reduced the mean stage cost by 38% compared to the performance of the existing, hand-tuned MPC model. • Maximum likelihood identification of disturbance models for offset-free linear MPC. • Method used to identify model from real-world, closed-loop chemical plant data. • Identified model used to design a replacement MPC for the chemical plant. • Tracking performance is improved by 38% over the old, hand-tuned MPC design. [ABSTRACT FROM AUTHOR]

In this paper, we propose an efficient data-driven predictive control approach for general nonlinear processes based on a reduced-order Koopman operator. A Kalman-based sparse identification of nonlinear dynamics method is employed to select lifting functions for Koopman identification. The selected lifting functions are used to project the original nonlinear state–space into a higher-dimensional linear function space, in which Koopman-based linear models can be constructed for the underlying nonlinear process. To curb the significant increase in the dimensionality of the resulting full-order Koopman models caused by the use of lifting functions, we propose a reduced-order Koopman modeling approach based on proper orthogonal decomposition. A computationally efficient linear robust predictive control scheme is established based on the reduced-order Koopman model. A case study on a benchmark chemical process is conducted to illustrate the effectiveness of the proposed method. Comprehensive comparisons are conducted to demonstrate the advantage of the proposed method. • An efficient data-driven predictive control for nonlinear processes is proposed. • A reduced-order Koopman identification approach is proposed for modeling. • A Kalman-based algorithm is leveraged to automate Koopman lifting function selection. • A robust linear MPC scheme is developed based on the reduced-order Koopman model. • Comparisons demonstrate the effectiveness and superiority of the proposed method. [ABSTRACT FROM AUTHOR]

Sequential stochastic model predictive control (MPC) is a control framework that sequentially employs two optimizers. The high-level optimizer generates a coarse long-term plan based on nominal uncertainty values, while the low-level optimizer refines the short-term plan by considering all possible realizations of these uncertainties. This paper highlights the advantages of the sequential stochastic MPC framework over the traditional multistage MPC approach. It offers faster computation times, improved feasibility of the optimal control problem (OCP), less conservative solutions, and enhanced flexibility in framework tuning. These advantages of sequential stochastic MPC are demonstrated through a case study on flood control in a hydropower station. Additionally, computational efficiency is further boosted with the simplified method which was proposed previously. The results reveal a remarkable performance improvement with an approximately 85 times faster computation time than the standard multistage MPC. These findings establish sequential stochastic MPC framework with the potential for practical implementation. • Proposing sequential stochastic MPC framework. • Uses two optimization steps sequentially to compute an optimal control input. • Demonstrate the better control solution than multistage MPC. • Offers more flexibilities in tuning. • Demonstrated Better feasibility and less computational demand than multistage MPC. [ABSTRACT FROM AUTHOR]

The kernel of industrial advanced process control (APC) lies in the formulation and solution of model predictive control (MPC) problems, which specify the controller moves according to the solution of an optimal control problem at each sampling time. A significant challenge is the online computation for large-scale industrial systems. As the state-of-the-art APC technology, the Shell–Yokogawa Platform for Advanced Control and Estimation (PACE) has adopted a systematic framework of handling dynamic optimization of large-scale systems, where an automatic decomposition procedure generates subsystems for distributed MPC. The decomposition is implemented on network representations of the MPC models that capture interactions among process variables, with community detection used to maximize the statistical significance of the subnetworks with preferred internal interconnections. This paper introduces the fundamentals of such a decomposition approach and this functionality in PACE, followed by a case study on a crude distillation process to showcase its industrial application. • An industrial implementation of auto-decomposition for distributed MPC is reported. • The platform for advanced control of Shell–Yokogawa, PACE, is introduced. • A directed network captures structural dynamic interactions among process variables. • Community detection algorithm is practically refined to yield desirable subsystems. • Industrial-scale plantwide applications showcase the computational acceleration. [ABSTRACT FROM AUTHOR]

Neural Ordinary Differential Equation (NODE) is a recently proposed family of deep learning models that can perform a continuous approximation of a linear/nonlinear dynamic system using time-series data by integrating the neural network model with classical ordinary differential equation solvers. Modeling nonlinear dynamic processes using data has historically been a critical challenge in the field of chemical engineering research. With the development of computer science and neural network technology, recurrent neural networks (RNN) have become a popular black-box approach to accomplish this task and have been utilized to design model predictive control (MPC) systems. However, as a discrete-time approximation model, RNN requires strictly uniform step sequence data to operate, which makes it less robust to an irregular sampling scenario, such as missing data points during operation due to sensor failure or other types of random errors. This paper aims to develop a NODE model and construct an MPC based on this novel continuous-time neural network model. In this context, closed-loop stability is established and robustness to noise is addressed using a variety of data analysis techniques. An example of a chemical process is utilized to evaluate the performance of NODE-based MPC. Furthermore, the performance of the NODE-based MPC under Gaussian and non-Gaussian noise is investigated, and the subsampling method is found to be effective in building models for MPC that are suitable for handling the presence of non-Gaussian noise in the data. • Overview of neural ODE modeling approach. • Model predictive control design using neural ODEs. • Handling noisy data in neural ODE modeling. • Application to a chemical process example. [ABSTRACT FROM AUTHOR]

Highlights • Development of a simple lumped battery model amenable for MPC implementation. • A novel MPC formulation that takes into account both charging speed and capacity fade to find optimal charging current profile that can replace constant current charging mode. • Interesting modifications in the MPC implementation to account for the variation of prediction horizon with variations in decision variable values. Abstract In this part of the paper, a control strategy for optimal charging is discussed. This work seeks to develop a capacity fade minimizing model predictive control (MPC) framework, which can help in identification and realization of optimum charge-discharge cycles in Lithium-ion (Li-ion) batteries. Although the model developed in the first part is a good representation for a battery, it has limitations for on-line applications due to its complexity. For on-line applications, it is important that the model is computationally fast, but at the same time incorporate the effects of various capacity fade mechanisms. Development of a simple lumped model to meet these requirements is also a part of this work. [ABSTRACT FROM AUTHOR]

Highlights • Utilizes data variety in subspace identification and control of batch processes. • The modeling approach is able to use data variety for batches of unequal lengths. • A novel model predictive control (MPC) with explicit model validity constraints, and the ability to specify target PSD is proposed. • The approach is demonstrated on batch crystallization process. Abstract This manuscript illustrates the use of big data for modeling and control of batch processes. A modeling and control framework is presented that utilizes data variety (temperature or concentration measurements along with size distribution) to achieve newer control objectives. For an illustrative crystallization process, an approach is proposed consisting of a subspace state-space model augmented with a linear quality model, able to model and predict, and therefore control the particle size distribution (PSD). The identified model is deployed in a linear model predictive control (MPC) with explicit model validity constraints. The paper presents two formulations: a) one that minimizes the volume of fines in the product by leveraging the variety of measurements and b) the other that directly controls the shape of the particle size distribution in the product. The former case is compared to traditional control practice while the latter's superior ability to achieve desired PSD shape is demonstrated. [ABSTRACT FROM AUTHOR]

Highlights • Simulation study to economics optimizing control and robust multi-stage optimizing control. • Discussion of numerical aspects of optimizing control as well as the software realization using high-fidelity high-dimensional process models of complex chemical processes. • Control problem formulation for the dynamic product changeover in a multi-product process. • Robustification of economics optimizing control by multi-stage nonlinear model predictive control in case of the presence of model uncertainty. Abstract By the use of online optimization, the profitability of chemical processes can be enhanced while meeting process-related, environmental and ecological constraints. Two well-known techniques to achieve an economically optimal operation of chemical processes are repetitive stationary optimization (RTO) combined with advanced control and direct economics optimizing control on a receding horizon (also called one-layer approach or dynamic RTO). Direct economics optimizing control can react faster to disturbances and is better suited for the optimization of transients between different products. On the other hand it is computationally demanding and requires sufficiently accurate dynamic models. In this paper, we discuss economics optimizing control of a very complex process, a multi-product transesterification reaction that is realized in a reactive distillation process. The process model, the specification of the economics optimizing controller, the implementation of the dynamic optimization, and the robustification of the controller by a multi-stage approach are discussed using simulation studies. [ABSTRACT FROM AUTHOR]

In this paper, we propose a conceptual framework for nonlinear systems to integrate real-time optimization (RTO) and model predictive control (MPC) under time-varying economic environments. In the RTO layer, we introduce a lookup table including a large number of steady-state points of the nonlinear system, which are predetermined offline. Once the parameters of the economic cost function are varied, we are able to take a quick online search on the lookup table to find a point for satisfied economic performance and then send it to the MPC layer as a temporary control target. The temporary target is also employed as the initial solution for solving the optimization of the RTO layer. When the optimal target is calculated by RTO, it will replace the temporary one as the new control target of MPC. Compared to the two-layer framework which suffers from long waiting time to get the optimal operating points, the lookup-table-based RTO (LT-RTO) framework provides a quick-produced suboptimal target for MPC. It avoids unnecessary economic losses if MPC is still tracking outdated target even parameters of the cost function have already changed. We demonstrate the effectiveness through a chemical process model that the LT-RTO framework makes an improvement of the economic performance. [ABSTRACT FROM AUTHOR]

This paper addresses both the design of an optimal variable setpoint and a setpoint-tracking control loop for the dissolved oxygen concentration in a biological wastewater treatment process. Although exact knowledge of influent changes during rain/storm events is unrealistic, we take advantage of the fact that during dry weather conditions the influent changes are periodic and thus predictable. Specifically, a nonlinear optimization procedure utilizes dry weather data to decide on a nominal fixed setpoint, or a weighting gain, or both; during weather events an algorithm uses the optimization solution(s) together with the ammonium predictions to adjust the setpoint dynamically (responding appropriately to significant changes in the influent). A constrained nonlinear neural-network model predictive control tracks the setpoint. Simulations with the BSM1 compare several variations of the proposed methods to a fixed-setpoint PI control, demonstrating improvement in effluent quality or reduction in energy use, or both. [ABSTRACT FROM AUTHOR]

With the potential to decrease operating costs and improve energy efficiency, model predictive control (MPC) has been proposed as a replacement for traditional heuristic, PID, and other conventional control strategies for heating, ventilation, and air conditioning (HVAC) systems in commercial buildings. Due to the size of large commercial HVAC systems, implementing MPC as a single monolithic optimization problem is not practical nor desirable given real-time operating requirements. In this paper, we present a hierarchical decomposition for economic MPC in large-scale commercial HVAC systems using a two-layer approach. We show a sample optimization for a campus of 25 buildings with 500 total zones and a central plant consisting of eight chillers. Then, we discuss an application of the ideas presented here in the recently completed $485-million replacement of the Stanford campus heating and cooling systems and conclude with some of the control theory challenges presented by this new class of applications. [ABSTRACT FROM AUTHOR]

Scheduling, optimization and control of power for three industrial cogeneration plants at one of Dow’s Louisiana site is presented in this paper. A first principle mathematical model that includes mass and energy balances for gas turbines, heat recovery units, steam turbines, pressure relief valves and steam headers is used to formulate multiple optimization problems to recommend the best strategy to trade power. The model has detailed operational information that includes equipment status and control curves for different operating scenarios. The scheduled power offer curve is obtained by solving multiple optimization problems using the validated process model along with operational and equipment limitations. Adjustment of power schedule offer is done in the real-time market thirty minutes prior to the hour and implementation of the dispatched power schedule is done using a model predictive controller. [ABSTRACT FROM AUTHOR]

Stochastic model predictive control (SMPC) provides a probabilistic framework for MPC of systems with stochastic uncertainty. A key feature of SMPC is the inclusion of chance constraints, which enables a systematic trade-off between attainable control performance and probability of state constraint violations in a stochastic setting. This paper presents an overview of core concepts in SMPC in relation to MPC and stochastic optimal control, with numerical illustrations on a typical chemical process. Estimation of stochastic disturbances as well as the impact of estimation quality of stochastic disturbances on the SMPC performance are discussed. Some avenues for future research in SMPC are suggested. [ABSTRACT FROM AUTHOR]

This paper presents a method for efficient optimization of a plunger lift process in shale gas wells. Plunger lift is a cyclic process consisting binary decision as well as continuous and discrete state variables. The time-series data comprising of surface measurements are converted into cycle-wise process-relevant performance outputs, while the binary manipulated variable is transformed into continuous threshold values. These transformed variables are used to develop a reduced order cycle-to-cycle model and corresponding receding horizon optimization problem that maximizes daily production while meeting operational constraints. The efficacy of the proposed algorithm is demonstrated on a simulated plunger lift process. [ABSTRACT FROM AUTHOR]

• The steady-state points' deviations of each working conditions is employed as the identification data. • The PWA model of 660-MW USC CFB boiler unit is developed using SIM. • The construction of the excitation signal for identification in practical applications is fully considered. • The MLD-MPC method is exploited to realize the control of the 660-MW USC CFB boiler unit. In this paper, fully considering the form of the nonlinear model after linearization at the steady-state point, an improved subspace identification method based on the steady-state points' deviations data was proposed to identify piecewise affine model. Besides, the construction of the excitation signal in practical applications is fully considered. The identification results demonstrate that the method has better adaptability to strong nonlinear systems. The identification normalized root mean square error of each working condition is almost all less than 10%. On this basis, a framework that is widely applicable to complex system control is established combining with the mixed logic dynamic model. The canonical form realization is performed to transfer the local models into the same state basis. Predictive control was carried out on the boiler-turbine system of the 660-MW ultra-supercritical circulating fluidized bed unit based on the above framework. The results indicate the effectiveness of the framework. [ABSTRACT FROM AUTHOR]

This paper discusses recent developments in the data-based modeling and control of nonlinear chemical process systems using sparse identification of nonlinear dynamics (SINDy). SINDy is a recent nonlinear system identification technique that uses only measurement data to identify model dynamical systems in the form of first-order nonlinear differential equations. In this work, the challenges of handling time-scale multiplicities and noisy sensor data when using SINDy are addressed. Specifically, a brief overview of novel methods devised to overcome these challenges are described, along with modeling guidelines for using the proposed techniques for process systems. When applied to two-time-scale systems, to overcome model stiffness, which leads to ill-conditioned controllers, a reduced-order modeling approach is proposed where SINDy is used to model the slow dynamics, and nonlinear principal component analysis is used to algebraically "slave" the fast states to the slow states. The resulting model can then be used in a Lyapunov-based model predictive controller with guaranteed closed-loop stability provided the separation of fast and slow dynamics is sufficiently large. To handle high levels of sensor noise, SINDy is combined with subsampling and co-teaching to improve modeling accuracy. The challenges of modeling and controlling large-scale systems using noisy industrial data are then addressed by using ensemble learning with SINDy. After summarizing the advances, a nonlinear chemical process is used to provide an end-to-end demonstration of process modeling using sparse identification with guidelines for chemical engineering practitioners. Finally, several future research directions for the incorporation of SINDy into process systems engineering are proposed. • Overview of sparse identification modeling approach. • Methods for handling time-scale multiplicity and noise in the data. • Use of sparse identification modeling in predictive control. • Examples of chemical process systems. [ABSTRACT FROM AUTHOR]

This paper describes a method for model predictive control (MPC) with model maintenance. A supervisor maintains the control model in real time by providing an estimate of disturbances and noise. A switch is triggered when the predictions of the control model deviate from disturbance estimate by more than a pre-determined amount. The predictive control algorithm described in the paper uses the innovation representation of a Markov–Laguerre model. A Monte Carlo study and an experiment show that good models and stable control are obtained. A simulation study based on models of a boiler–turbine unit shows that the algorithm can adapt to time-varying data. The performance is assessed using the area regulation (AR) test criterion currently adopted by PJM Interconnection LLC. The proposed adaptive MPC gives an AR test score of more than 90 with pressure fluctuations less than 3% even when the coal quality changes. [ABSTRACT FROM AUTHOR]

Commercially, the Czochralski process plays a key role in production of monocrystalline silicon for semiconductor and solar cell applications. However, it is a highly complex batch process which requires careful control throughout the whole crystal production. In the present paper, an iterative method based on model predictive control (MPC) for calculating a new and improved trajectory from a growth run to the next is explored. The method uses the results of the previous growth run in combination with an underlying model which incorporates the complex dynamic effect of the heater temperature on the pulling rate. The motivation behind this choice of strategy is to enhance the quality of the fully grown ingot from one run to the next by applying the most recent estimates of the unknown parameters. The results show that combining MPC, estimation and Run-To-Run control has enabled simulation of effective control of the Czochralski crystallization process. [ABSTRACT FROM AUTHOR]

Even if the performances of bioprocesses can be significantly improved by model-based control, there often remains a tradeoff between model complexity and control robustness. This paper proposes an original data-driven strategy for fast design of dynamic bioprocess models with minimal complexity (i.e., minimal number of bioreactions). Maximum likelihood principal component analysis (MLPCA) is applied to infer the minimal reaction scheme from a 25-state mammalian cell culture database. Then, a systematic algorithm is used to provide a continuous kinetic model formulation assuming all rates to occur simultaneously, which may be far from true cell metabolic conditions sometimes presenting discontinuous metabolic switches. A robust model predictive formulation is therefore adopted to reduce the impact of model structural uncertainty on the process performances. Additional numerical results show that the proposed strategy presents excellent performances in presence of unexpected metabolic switches. • A systematic procedure generating simple data-driven mechanistic models. • A fast kinetic structure generation based on Fisher information. • Application of the method to a 25-state variable model of mammalian cell cultures. • The kinetic structural uncertainty is tackled by a robust multi-stage NMPC. • The method pros and cons are highlighted, suggesting an optimal control design. [ABSTRACT FROM AUTHOR]

Abstract: This paper will discuss the evolution of reactor control over the last 25 years within the DuPont Company and its subsidiaries. It will focus on high level trends in control philosophy, systems and approaches. These changes have been necessary in order to achieve higher rates, better yields, improved uptime and a more sustainable footprint. This paper is an update to an article that was presented twenty-five years ago at Chemical Process Control (CPC) 3. The main focus will be on the use of Model Predictive Control (MPC) for reactor processes. [Copyright &y& Elsevier]

Abstract: The Modelica language, targeted at modeling of complex physical systems, has gained increased attention during the last decade. Modelica is about to establish itself as a de facto standard in the modeling community with strong support both within academia and industry. While there are several tools, both commercial and free, supporting simulation of Modelica models few efforts have been made in the area of dynamic optimization of Modelica models. In this paper, an extension to the Modelica language, entitled Optimica, is reported. Optimica enables compact and intuitive formulations of optimization problems, static and dynamic, based on Modelica models. The paper also reports a novel Modelica-based open source project, JModelica.org, specifically targeted at dynamic optimization. JModelica.org supports the Optimica extension and offers an open platform based on established technologies, including Python, C, Java and XML. Examples are provided to demonstrate the capabilities of Optimica and JModelica.org. [Copyright &y& Elsevier]

• Wood fibers are chemically separated and physically deformed in a pulp digester. • A multiscale model is developed to capture the deformation of fiber during pulping. • The extended purdue model is integrated with kMC model and column buckling theory. • A model predictive controller is designed to regulate fiber curl and kappa number. • The closed-loop operation successfully suppresses the fiber curvature during cooking. Lightweight packaging in the pulp and paper industry has received much attention as an effective way to reduce CO 2 emissions. One challenge is that the tensile strength of packaging papers, which determines the maximum load, has to be comparable to that of conventional packaging materials. In a pulp digester, wood fibers are chemically separated and physically deformed during pulping, resulting in the dynamic change of tensile strength of end-use papers. In this work, the deformation of fiber during Kraft pulping is described by integrating the multiscale modeling framework of Choi and Kwon (2019a) and the classical column buckling theory. Then, an approximate model is identified and employed to design a model predictive control system to regulate the fiber deformation during pulping. The proposed control system achieved a lower degree of fiber deformation than a conventional pulping strategy, thereby contributing to the enhanced tensile strength of end-use papers. [ABSTRACT FROM AUTHOR]

This paper is devoted to the multiple input multiple output (MIMO) model predictive control (MPC) of a catalytic reverse flow reactor (RFR). The RFR aims to reduce, by catalytic reaction, the amount of volatile organic compounds (VOCs) released in the atmosphere. The peculiarity of this process is that the gas flow inside the reactor is periodically reversed in order to trap the heat released during the reaction. The objective of this paper is to propose a solution to avoid the limitation seen in a previous study for the single input single output (SISO) control case. It was dealing with the impossibility to avoid degradation of the catalytic elements due to the excessive heating induced by the exothermic reaction. In order to overcome this issue, a ratio of the inlet (cold) gas flow is now bypassed into the central zone. This allows introducing a second manipulated variable: the dilution rate. The phenomenological model considered here for the MIMO MPC of the RFR is obtained from a rigorous first principles modeling. The resulting accurate nonlinear partial differential equation (PDE) model can be a drawback for the model based controller implementation. To overcome this issue, we use a MPC strategy previously developed: it combines a two phase approximation of the PDE model in an internal model control (IMC) structure. This strategy allows using a less accurate model and a less time-consuming control algorithm. Finally, efficiency of the control approach is shown in simulation. [Copyright &y& Elsevier]

This paper introduces an enhancement to the model predictive control (MPC) algorithm to address to address variable time delays that may occur in the control loop. These variable delays can arise from various sources such as measurement delays, human-in-the-loop, and communication delays. The specific focus of this paper is to investigate the effect of random communication delays on network-based process control systems.The process industry is moving towards sensor and control systems that are accessed over networks rather than hardwired. For fast responding processes, the communication delay introduced by the networks can lead to performance degradation and even instability. Our experimental characterization of network communication delays revealed that they were mostly white, had a baseline minimum and approached wide-tailed distributions. We propose the time-stamped model predictive control (TSMPC) algorithm, an extension to MPC that uses a communication delay model, along with time-stamping and buffering to improve reliability over networked control systems. Experimental validation of this new algorithm resulted in improved performance over traditional MPC. Where time-stamping is not possible, accounting just for the mean/median communication delay can still result in better performance, and this simplification to TSMPC is termed as the mean/median delay model predictive control (MDMPC). Future enhancements that are possible include addressing non-linearity, input constraints, disturbance rejection, and model mismatch within the algorithm as well as to characterize and tune for delays from other control loop elements. [Copyright &y& Elsevier]

• Volatile energy price based scalarization and weighting method for energy management systems. • Comparison of existing scalarization and weighting methods for a use case of thermal batch processes. • Investigation of the effect of model inaccuracy on the performance of energy management systems. Predictive energy management systems (EMSs) enable industrial plants to operate the energy supply systems at optimal efficiency, taking account of multiple objectives, including energy cost reduction. The performance of model-based EMSs depends on the appropriate design and correct scalarization of the resulting multiobjective function. This paper introduces the Volatile Energy Prices Scalarization (VEPS) method, which effectively designs, standardizes, and weighs the multiobjective function of model-based EMSs without the need for prior simulations or test runs. We present a case study, in which we compare the VEPS method to other state-of-the-art methods, utilizing a validated simulation model from an industrial food plant. The results show that the VEPS method outperforms other weighting methods with comparable tuning effort in this case-study. Moreover, the performance of the VEPS method is close to the Pareto-optimal performance. Economic weighting methods such as VEPS enable a fast and cost-effective implementation of EMS in the manufacturing industry. [ABSTRACT FROM AUTHOR]

Industrial model predictive control (MPC) typically includes a steady-state target calculation (SSTC) layer that provides targets for the dynamic MPC controller to track. This paper develops the first closed-loop robust SSTC method, which explicitly addresses the closed-loop uncertainty, for stable linear systems with time-invariant parametric uncertainty. The problem is initially formulated as a stochastic bi-level optimization problem, where the lower-level optimization problem describes the nominal MPC law. The problem is then transformed into a deterministic mixed-integer second-order cone program for ellipsoidal uncertainty. Case study results show that the proposed method is more robust than the classical nominal SSTC in that it helps reduce cycling of the steady-state targets. In addition, the proposed method is less conservative than a typical open-loop robust SSTC method, in that it drives the system to the steady state more quickly and can push the steady state closer to the operating boundary when desired. • The first closed-loop robust steady-state target calculation method for MPC. • A bilevel stochastic program where the lower-level models steady-state control. • Reformulation of the bilevel stochastic program to a deterministic MISOCP. • The proposed method outperforms the nominal method and the open-loop robust method. [ABSTRACT FROM AUTHOR]

In this paper, we propose a model predictive control (MPC) technique combined with iterative learning control (ILC), called the iterative learning model predictive control (ILMPC), for constrained multivariable control of batch processes. Although the general ILC makes the outputs converge to reference trajectories under model uncertainty, it uses open-loop control within a batch; thus, it cannot reject real-time disturbances. The MPC algorithm shows identical performance for all batches, and it highly depends on model quality because it does not use previous batch information. We integrate the advantages of the two algorithms. The proposed ILMPC formulation is based on general MPC and incorporates an iterative learning function into MPC. Thus, it is easy to handle various issues for which the general MPC is suitable, such as constraints, time-varying systems, disturbances, and stochastic characteristics. Simulation examples are provided to show the effectiveness of the proposed ILMPC. [ABSTRACT FROM AUTHOR]

Model predictive control (MPC) is an advanced control strategy that has a growing interest for research and applications because of its good performance in many kind of processes and its ability to handle constraints, perform optimization, and consider economic aspects and nonlinearities of the process. However, its design, evaluation and implementation require a high level of expertise which might restrict the developments in this area. This paper presents a software framework developed in Pyomo, a mathematical modelling language embedded in Python, for the assessment and implementation of ideal nonlinear MPC and advanced step nonlinear MPC. The framework automates many of the aspects of MPC defining new classes in Pyomo for ideal NMPC (iNMPC) and advanced step NMPC (asNMPC). The user only has to define the prediction model of the process using new classes for manipulated variables, disturbances and initial conditions, and the real plant function to access to the states of the process in a similar way of other algebraic modelling languages. The model discretizaion, controller set-up, receding horizon and solution are done automatically. Three examples are presented in detail for explaining the use and advantages of the framework to evaluate iNMPC and asNMPC controllers. The software is freely available upon request and in the future it is expected to be an official extension of Pyomo. [ABSTRACT FROM AUTHOR]

This paper proposes efficient step response model implementation strategies that lead to accurate control and high computational performance in an embedded Model Predictive Control (MPC) scheme. Different implementations of the step response prediction model are examined, and inherent properties that directly affect control performance in the presence of disturbances are discussed. Model errors that are inconsistent with bias updates (i.e. the model of unknown disturbances commonly used in step response MPC) are identified, and it is shown that the bias updates may worsen the effect of the errors in some cases. Particular attention is paid to the robustness of the prediction models to small truncation errors and errors in the input or measured disturbance history. Several implementation aspects that are crucial for embedded targets with limited resources are discussed. The findings are illustrated by simple simulation examples and an industrial case-study involving hardware-in-the-loop simulation of a subsea compact separation process. [ABSTRACT FROM AUTHOR]

Scheduling of power and real-time optimization for three industrial cogeneration plants at one of Dow's Louisiana site is presented in this paper. A first principle mathematical model that includes mass and energy balances for gas turbines, heat recovery units, steam turbines, pressure relief valves and steam headers is used to formulate an optimization problem to recommend the best strategy to trade power. The model has detailed operational information that includes equipment status and control curves for different operating scenarios. The model can also accurately predict the effect of ambient temperature, thereby resulting in an optimal day-ahead schedule. Adjustment of power schedule is done in the real-time market 30 min prior to the hour and implementation of the dispatched power schedule is done using a model predictive controller. [ABSTRACT FROM AUTHOR]

Many chemical processes exhibit disparate timescale dynamics with strong coupling between fast, moderate and slow variables. To effectively handle this issue, a model predictive control (MPC) scheme with a non-uniformly spaced optimization horizon is proposed in this paper. This approach implements the time intervals that are small in the near future but large in the distant future, allowing the fast, moderate and slow dynamics to be included in the optimization whilst reducing the number of decision variables. A sufficient condition for ensuring stability for the proposed MPC is developed. The proposed approach is demonstrated using a case study of an industrial paste thickener control problem. While the performance of the proposed approach remains similar to a conventional MPC, it reduces the computational complexity significantly. [ABSTRACT FROM AUTHOR]

An overview of the recent developments of time-series neural network modeling is presented along with its use in model predictive control (MPC). A tutorial on the construction of a neural network-based model is provided and key practical implementation issues are discussed. A nonlinear process example is introduced to demonstrate the application of different neural network-based modeling approaches and evaluate their performance in terms of closed-loop stability and prediction accuracy. Finally, the paper concludes with a brief discussion of future research directions on neural network modeling and its integration with MPC. • Review of neural network model approaches. • Training and parameter estimation of neural network models. • Neural network model performance evaluation and improvement. • Implementation in MPC and evaluation of closed-loop performance. [ABSTRACT FROM AUTHOR]

• Control approach suitable for implementation on embedded platforms. • Optimization-free evaluation of close-to-optimal control inputs. • Guarantee of asymptotic stability with respect to physical constraints. • The benefits of the approach demonstrated using a 32-bit microcontroller. • Reduction of computational time and energy consumption. Advanced, optimization-based control methods are implemented at each level of industrial production. Although the model predictive control (MPC) represents a state-of-the-art control strategy maximizing profit, and, simultaneously, minimizing the energy losses, its industrial implementation is limited by the requirements on the software and hardware resources. This paper proposes an approximated MPC for such cases when the implementation of implicit MPC has a prohibitive effort on industrial hardware and its explicit counterpart is limited by its memory footprint requirements. The recursive random shooting-based approach is introduced to eliminate the conventional optimization and minimize the memory requirements, meanwhile guarantying the asymptotic stability subject to the physical constraints on control inputs and system states. The benefits of the proposed method are significantly reduced computational complexity by 93%, and decreased energy consumption by 93%, compared to non-recursive random shooting, while the performance loss is approximately 8%. [ABSTRACT FROM AUTHOR]

This paper describes a procedure to find the best controlled variables in an economic sense for the activated sludge process in a wastewater treatment plant, despite the large load disturbances. A novel dynamic analysis of the closed loop control of these variables has been performed, considering a nonlinear model predictive controller (NMPC) and a particular distributed NMPC-PI control structure where the PI is devoted to control the process active constraints and the NMPC the self-optimizing variables. The well-known self-optimizing control methodology has been applied, considering the most important measurements of the process. This methodology provides the optimum combination of measurements to keep constant with minimum economic loss. In order to avoid nonfeasible dynamic operation, a preselection of the measurements has been performed, based on the nonlinear model of the process and evaluating the possibility of keeping their values constant in the presence of typical disturbances. [ABSTRACT FROM AUTHOR]

The implementation of supply chains based on biomass conversion requires the exploration of various aspects, including the selection of processing technologies, configuration of the supply chain, portfolio of products as well as the feedstock selection. One important feature of this system is that the composition of the available biomass changes drastically through the year because this depends significantly on the climatic conditions; this way, the dynamic behavior of this process is an important issue that must be considered. This study presents a dynamic optimization model for the optimal planning of a distributed biorefinery system taking into account the time dependence of the involved variables and parameters. In addition, this paper incorporates a model predictive control methodology to obtain the behavior of the storages and orders of the supply chain; where the objective function is the difference between the required and satisfied demands in the markets. Therefore, this study considers relevant issues, which include the multiple available biomass feedstocks at various harvesting sites, the availability and seasonality of biomass resources, potential geographical locations for processing plants that produce multiple products using diverse production technologies, economies of scale for the production technologies, demands and prices of multiple products in each consumer, locations of storage facilities and a number of transportation modes between the supply chain components. The model was applied to a case study for a distributed biorefinery system in Mexico. Results show that is possible to get the configuration and the behavior of the supply chain considering its dynamic behavior in a rigorous way; furthermore, the solutions obtained by the model illustrate that the supply chains based on biomass conversion are seriously affected by the availability of bioresources over the time. [ABSTRACT FROM AUTHOR]

In this paper, a novel enhanced-efficiency selection of operating variables based on self-optimizing control (SOC) method for the vapor compression refrigeration cycle (VCC) system is proposed. An objective function is proposed to maximize the energy efficiency of the VCC system while meeting with the demand of indoor thermal comfort. With the detailed analysis of operating variables, three unconstrained degrees of freedom are selected among all the candidate operating variables. Then two SOC methods are applied to determine the optimal individual controlled variables (CVs) and measurement combinations as CVs. The model predictive control (MPC) method based controllers and PID controllers are designed for different sets of CVs, and the experimental results indicate that the proposed selection of CVs can achieve a good trade-off between optimal (or near optimal) stable operation and enhanced-efficiency of the synthesized control structure. [ABSTRACT FROM AUTHOR]

This paper addresses inter- and intra-patient variability in the context of automated drug delivery during anesthesia. A combined strategy of model predictive control (MPC) and least squares online parameter estimation for the control of the hypnotic depth, measured by the Bispectral Index (BIS), under uncertainty is presented, where the uncertainty originates from patient variability. The parameter with the highest sensitivity, C 50 the effect site concentration at 50% drug effect, is estimated online. The performance of the closed loop control design is shown for induction and maintenance of volatile anesthesia. In the maintenance phase, the control strategy is evaluated for predefined disturbances that are commonly occurring during surgery. The presented approach shows an improved performance compared to the nominal MPC controller under uncertainty. [ABSTRACT FROM AUTHOR]

In this paper, several methodologies of integrated design are proposed and applied to the design of wastewater treatment plants and their control system, focusing on the activated sludge process, within a novel multiobjective framework. The scope of the problem considers both fixed plant layout and plant structure selection by defining a simple superstructure. The control strategy chosen is a linear Model Predictive Controller (MPC) with terminal penalty. The evaluation of the controllability has been performed using norm based indexes, and the robustness conditions for different uncertainty sources have been considered, in the frequency and time domains. The optimization strategies used are based on the integration of stochastic and deterministic methods, as well as genetic algorithms. The presented methodologies and their application to wastewater treatment plants can be considered as an illustrative example in the universe of integrated design techniques presented in the Part I article of this series. [ABSTRACT FROM AUTHOR]

Model predictive control (MPC) algorithms with state-space process modelling, both linear and nonlinear, and state estimation methods for this algorithms are the subject of this paper. The considerations are under realistic assumption that processes are under influence of external disturbances and their models are not precise. This leads in most cases to errors in state estimation and, further, may lead to errors in feedback control. Attention has been paid to this problem in recent years. Two approaches are now available, an earlier one with additional disturbance state modelling and process-and-disturbance state estimation and a more recently proposed approach with different way of disturbance modelling and estimation of the process state only. The main aim of this paper is to provide a comprehensive comparison of the two mentioned approaches, including also discussion of available state estimation algorithms. To the best knowledge of the authors, there is still a lack of clear understanding of the differences between the mentioned approaches, in particular from practical point of view. After short presentation of the two methods and analysis of their theoretical aspects, a comprehensive comparative analysis is provided on a nonlinear example of the polymerization reactor. [ABSTRACT FROM AUTHOR]