Technology Model Predictive Control

Cybernetica delivers Nonlinear Model Predictive Control (NMPC) based on mechanistic models. Our software product, Cybernetica CENIT, offers a flexible architecture that can meet any industrial challenge with optimal solutions.

Advantages of applying MPC

  • Multivariable optimal control
  • Predictive control – intelligent feed forward
  • Optimal constraint handling
  • Adaptive control through state and parameter estimation
  • Feedback from indirect measurements through process model

Advantages of NMPC over linear MPC

  • Nonlinear models are valid over larger operating ranges
  • Improved control of nonlinear processes
  • Less need for step response experiments
  • Improved state and parameter estimates

Typical applications

  • Control of batch and semi-batch processes
  • Control of nonlinear processes operated under varying conditions
  • Optimal grade transition in continuous processes
  • Safe control of exothermal processes
  • Control of unmeasured variables, such as conversion rates and product quality

Typical benefits

  • Increased productivity
  • Consistent product quality
  • Reduced energy consumption and carbon footprint
  • Improved safety

What is Model Predictive Control?

Model Predictive Control (MPC) is an Advanced Process Control (APC) algorithm where a mathematical model of the process is used to predict and optimise future behaviour.

The MPC algorithm combines model prediction with an optimisation algorithm to calculate the optimal future input sequence. The optimisation is repeated on regular intervals in order to implement feedback from available measurements, through state and parameter estimation.

Figure2

Model predictive control is well suited for multivariable control problems as couplings between variables are taken into account through the prediction model. Another advantage is the built-in handling of constraints, which allows constraints to be formulated on both measured and calculated variables. The controller may for example be easily configured to adhere to operational constraints, such as actuator limits, or specification limits on derived variables, such as quality parameters.

Nonlinear Model Predictive Control

Nonlinear Model Predictive Control (NMPC) is a subset of Model Predictive Control where the optimisation algorithm is designed to handle nonlinear models. Nonlinear models often possess better predictive capabilities than linear models, which for MPC application implies better controller performance for processes with nonlinear characteristics.

Nonlinear Model Predictive Control with Cybernetica CENIT

Cybernetica CENIT is a powerful and versatile software suite for nonlinear model predictive control. Our applications use nonlinear mechanistic models, which is better and more robust than alternative solutions that often rely on linear models.

The models used in CENIT are developed specifically for nonlinear model predictive control. Model uncertainties and unknown disturbances are handled through advanced estimation algorithms, such as Kalman Filtering or Moving Horizon Estimation (MHE). The mechanistic structure of the models ensures robust and efficient online state and parameter estimation. The combination of mechanistic models with on-line model adaption are crucial elements for successful application of NMPC technology.

Distributed CENIT applications

CENIT applications can be organised in a distributed architecture where the individual applications may exchange full information. This allows for a wide range of architecture, such as application in series, for efficient control of downstream units in a production line, or a multilevel control hierarchy.

The latter is particularly useful for doing optimisation on different timescales, with a supervisory RTO (Real Time Optimisation) level optimising a reference trajectory for an NMPC controller.

Model and application development

The process model and specific code for estimation and control is implemented in a Cybernetica Model and Application Component, which is linked into the CENIT system. This separation allows for very specific tailoring of the application to meet the customers need and efficient deployment of tailored online solutions.

 

Safety and DCS interaction

CENIT applications ensure safe operation through in spite of process disturbances and changing process characteristics. The estimator algorithms will continuously adjust the model to compensate for any deviation between model calculations and process measurements.

In addition to the inherent safety associated with model-based techniques, CENIT has built in features for application specific fault detection:

  • Internal fail check in CENIT algorithms
  • Custom validation of input signals
  • Custom validation of state estimates
  • Diagnostics signals provided to DCS and operators over OPC

Offline engineering tools

The Model and Application components developed for CENIT can be used with the entire Cybernetica toolchain. Our product portfolio contains several tools to facilitate efficient workflow for application development and maintenance.

The developed model may also be used in other Cybernetica tools for process simulation, soft sensing or batch optimisation.

Cybernetica ModelFit

Cybernetica ModelFit is a desktop application used by control engineers for:

  • Design and tuning of online estimators for use in CENIT
  • Estimation of constant or time-varying parameters
  • Estimation of initial states
  • Parameter identifiability analysis
  • Simultaneous estimation based on multiple datasets

ModelFit

Cybernetica RealSim

Cybernetica RealSim is a plant replacement simulator used for developing and testing CENIT applications before commissioning. RealSim communicates over the OPC protocol in order to replicate the plant architecture.

The plant replacement model may be the same as the CENIT model, but may also differ in order to evaluate controller performance in the presence of model uncertainties or unknown disturbances.

RealSim