Final Report Summary - MOBOCON (Model-based optimizing control - from a vision to industrial reality)
The chemical production in Europe is under strong pressures both from the economic side as wages are high and raw materials are more expensive than in other parts of the world and from the point of view of sustainability and protection of the environment. These pressures ask for continuous improvements in energy and resource efficiency of the production processes. Large savings of energy, better resource efficiency and better economic performance are possible at a relatively low cost compared to investments into new production technology by optimizing the operation of the processes and by controlling them better. In recent years, model-based control solutions have started to spread in the chemical industry beyond the petrochemical industry where they have first been introduced.
A recently proposed step forward towards even more efficient operation is to combine the calculation of control targets and model-predictive control into one model-based online optimizing control scheme where a rigorous nonlinear dynamic plant model is used to compute optimal trajectories for the control inputs that minimize the cost of production while meeting constraints on product quality, emissions, and equipment limitations. This approach has successfully been applied in polymerization reactors, but for a broad application, several challenges have to be overcome:
• Handling the inevitable mismatch between the model which is used in the optimizing controller and the behavior of the real process and adaptation to changes in the plant or other operating conditions (e.g. feed compositions)
• Efficiency and robustness of the numerical algorithms that are employed in optimizing control, providing fast feedback to avoid performance degradation due to delayed responses
• Improved interaction with and acceptance by the operators.
The MOBOCON project made significant contributions in the field of real-time optimization and optimizing control in the presence of plant-model mismatch. By using measured data to adapt the optimization problem by the so-called modifier adaptation, plants can be steered to their optimal operating point (e.g. minimal energy demand) even though the model employed is inaccurate. A new algorithm, called MAWQA – Modifier Adaptation with Quadratic Approximation was developed that is less vulnerable to measurement noise than other approaches.
In the area of optimizing control, the multi-stage formulation was further developed. Multi-stage MPC is based on a tree of scenarios for the further evolution of the process, e.g. due to different model parameters and computes optimal control inputs taking into account that the future control inputs can be adapted to new measurement information which will be obtained later. This reduces the conservatism of the method considerably. Several extensions of the approach, in particular to the situation where a state estimator is needed and to model adaptation and dual control were investigated.
For the efficient development and deployment of model-based robust optimizing control solutions, a flexible software platform called do-mpc has been realized where different techniques for optimization, estimation and performance monitoring can be combined with plant models of different level of detail in a modular, flexible manner and where efficient implementations of optimizing control can be connected easily to real plants. For applications with feedback rate demands in the range of milliseconds, the software platform MLI has been realized.
As a enabling technology for model-based optimizing control with high-frequency feedback rates, novel numerical methods for nonlinear model predictive control were developed, comprising decomposition algorithms and efficient sparse grid based scenario generation algorithms for Multi-stage MPC, adaptivity for Multi-Level Iterations, computational methods for mixed-integer optimal control with constraints involving integer variables, and new approaches for the problem of Dual Control.
The robust model-based optimizing control approach was demonstrated successfully for a very complex pilot plant in which a combined reaction and separation process takes place. This shows that by using state of the art software, model-based optimizing control can become a reality in complex industrial processes, leading to improved economic and ecologic performance.
A recently proposed step forward towards even more efficient operation is to combine the calculation of control targets and model-predictive control into one model-based online optimizing control scheme where a rigorous nonlinear dynamic plant model is used to compute optimal trajectories for the control inputs that minimize the cost of production while meeting constraints on product quality, emissions, and equipment limitations. This approach has successfully been applied in polymerization reactors, but for a broad application, several challenges have to be overcome:
• Handling the inevitable mismatch between the model which is used in the optimizing controller and the behavior of the real process and adaptation to changes in the plant or other operating conditions (e.g. feed compositions)
• Efficiency and robustness of the numerical algorithms that are employed in optimizing control, providing fast feedback to avoid performance degradation due to delayed responses
• Improved interaction with and acceptance by the operators.
The MOBOCON project made significant contributions in the field of real-time optimization and optimizing control in the presence of plant-model mismatch. By using measured data to adapt the optimization problem by the so-called modifier adaptation, plants can be steered to their optimal operating point (e.g. minimal energy demand) even though the model employed is inaccurate. A new algorithm, called MAWQA – Modifier Adaptation with Quadratic Approximation was developed that is less vulnerable to measurement noise than other approaches.
In the area of optimizing control, the multi-stage formulation was further developed. Multi-stage MPC is based on a tree of scenarios for the further evolution of the process, e.g. due to different model parameters and computes optimal control inputs taking into account that the future control inputs can be adapted to new measurement information which will be obtained later. This reduces the conservatism of the method considerably. Several extensions of the approach, in particular to the situation where a state estimator is needed and to model adaptation and dual control were investigated.
For the efficient development and deployment of model-based robust optimizing control solutions, a flexible software platform called do-mpc has been realized where different techniques for optimization, estimation and performance monitoring can be combined with plant models of different level of detail in a modular, flexible manner and where efficient implementations of optimizing control can be connected easily to real plants. For applications with feedback rate demands in the range of milliseconds, the software platform MLI has been realized.
As a enabling technology for model-based optimizing control with high-frequency feedback rates, novel numerical methods for nonlinear model predictive control were developed, comprising decomposition algorithms and efficient sparse grid based scenario generation algorithms for Multi-stage MPC, adaptivity for Multi-Level Iterations, computational methods for mixed-integer optimal control with constraints involving integer variables, and new approaches for the problem of Dual Control.
The robust model-based optimizing control approach was demonstrated successfully for a very complex pilot plant in which a combined reaction and separation process takes place. This shows that by using state of the art software, model-based optimizing control can become a reality in complex industrial processes, leading to improved economic and ecologic performance.