Model Predictive Control

The Model Predictive Control, mostly Model Predictive Control (MPC ) or Receding Horizon Control ( RHC ) called, is a modern method for predictive control of complex, usually multi - variable processes.

Operation

MPC in a time- discrete dynamic model of the controlled process is employed to calculate the future behavior of the process as a function of the input signals. This allows the calculation of the - in the sense of a quality function - optimum input signal, leading to optimum output signals. This can be considered the same input, output and state constraints. While the model behavior is predicted to a certain time frame N, only the input signal u typically used for the next time step, and then repeats the optimization. The optimization in the next time step is then performed with the actual ( measured ) condition which can be regarded as a feedback MPC and does, in contrast to optimal control of a settlement. This permits the consideration of interference, but also requires a considerable amount of computing power.

The process models may be of various shape, for example, Transfer function or Zustandsraumdartsellung. Besides the mostly linear process models, artificial neural networks are sometimes used to create a process model. These regulators are then used to grade the NMPC (Non- linear Model Predictive Control ), as well as forms of adaptive controllers.

Areas of application

Unlike many other modern control procedures MPC was due to its ability to take into account constraints explicitly, already widely applied in industry. MPC controllers are preferred in industrial processes (including combustion processes in power plants, waste incineration plants, paper machines, rolling mills and cement plants ) used gain in which conventional controls (P, D, PID controller) and fuzzy controller insufficient control performance, and the relevant system dynamics are slow enough to perform an optimization in each sampling can. Often MPC also serve as a parent, arrangements for basic automation, eg in the form of a cascade as the manipulated variable of a PID controller.

Procedural processes are often automated process control systems. The optimization algorithm of model predictive control is thereby usually not executed within the process-related components / controller, but implemented in an external process computer, the example communicates with the control system via OPC. This is attributable to the computing power needed for the calculation of the algorithm and the rather low computing capacity of the process-oriented controller. The computer power required is also dependent on the number of inputs and outputs of the process. A goal is to integrate MPC into the process-related components and thus avoid the costs of integrating specialized hardware. This is promising and particularly useful for processes with low number of inputs and outputs. In addition to the 'Online calculation of the algorithm in the controller, another approach is the computation of all solutions of an optimization problemes in advance a possibility. These precalculated results are then stored in the controller and searched during the operation.

Variants

• Move blocking
• Explicit MPC
• Minimum -time MPC
• Infinite horizon MPC