Predictive control
When driving a car we not only observe the position of the vehicle in a given moment but we also monitor the road, traffic, weather etc. conditions ahead, controlling and steering the car accordingly. This is a typical example of “predictive” control which is partly based on future information.
The basic idea behind predictive control is the following. Based on the process model, we predict, pre-calculate the expected change in the output signal in a given future domain. In this domain the future basic signal is considered as known. We define a quadratic cost function which regards the quadratic difference of the future reference signal and output signal in the given time domain (“horizon”), and also takes input signal increments into consideration. Minimising the cost function we get the series of manipulating signals but we only apply the first value of the series to the input of the process, and repeat the procedure at the next sampling point (receding horizon technique).
Predictive control algorithms are more efficient in improving control quality than PID algorithms, especially when the reference signal is known and the process has long dead time. If more information is available and the expected behaviour can be foreseen, we can make better decisions than by checking the difference between the reference signal and the output signal at the actual point of time and determine the manipulating signal accordingly. As the control foresees the behaviour expected beyond the dead time, the manipulating signal takes effect before the changes to keep track of or prevent the effects of such changes.
In industrial computer process control, predictive controls are the second most common control algorithms after PID controls. The general share of PID controls accounts for about 90%, while the remaining 10% is split between more advanced (e.g. optimal, feedback, predictive etc.) control techniques. Old customs are hard to change. Larger companies usually have a separate department responsible for the implementation and dissemination of advanced control techniques. They have already developed software which facilitates measurement, identification and modelling, as well as the simulation and real-time application of controls.