In modern tokamak machines, exploration and successful development of improved plasma regimes is impossible without adequate control systems. In ASDEX Upgrade, the control tasks are performed by two systems, the continuously operating machine control and the plasma control active as long as a plasma discharge lasts. Machine control based on programmable logic controllers operates on a relatively slow timescale of = 100 ms to configure and monitor the machine's technical systems. Real-time plasma controllers run on faster cycle times of a few milliseconds to feedback (FB) control plasma shape and performance quantities. During the burn of a discharge, a real-time supervisor monitors the full technical and physical system state ( = 10 ms) and applies alternate discharge program segments to optimize discharge performance or react to failures. The supervisor is fully integrated with a layered machine protection system.

Plasma position and shape control in ASDEX Upgrade is particularly difficult: Since the poloidal magnetic field (PF) coils are located reactor relevant outside the toroidal magnetic field coil system and distant from the plasma, each PF coil has a global effect on all shape quantities. This makes simultaneous control of shape parameters a multivariable problem. The feedback control algorithm is based on a matrix proportional-integral-derivative method, adapted to handle saturation of coil currents, excess of coil forces, or to balance loads among coils. Control cycle time is ~3 ms.

In parallel, the plasma performance control (sometimes called kinetic control ) acts on particle fueling and auxiliary heating systems. It consists mainly of FB loops each controlling a single variable. These circuits can be freely combined to simultaneously control a number of different plasma quantities. A clear hierarchy in the control processes allows special real-time processes to override the programmed plasma discharge feedback action: The set of controlled quantities may be changed dynamically, depending on the plasma regime detected; stabilizing actions may be triggered when plasma instabilities grow; and discharge termination by means of impurity addition is initiated when a neural network indicates an imminent disruption. The computation of the needed plasma parameters and instability indicators requires signal inputs from many diagnostic systems during each controller cycle.

Currently, a new plasma control system is being implemented as a distributed system of real-time controllers and diagnostic systems, which are connected via a deterministic communication network.