Reconfigurable Control of Power Using Learning Automata Plants

A deaerating feedwater heater, equipped with a water level controller and a pressure controller, has been chosen to investigate the feasibility of a reconfigurable control scheme for power plants by incorporating the concept of leaming automata. Simulation results based on a model of the Experimental Breeder Reac- tor (EBR-11) at the Argonne National Laboratory site in Idaho are presented to demonstrate the efficacy of the reconfigurable control scheme. basis. The situation becomes more complex with unstructured disturbances. An alternative approach is to apply the concept of leaming automata (2),(3) to continually monitor the system performance. In this approach, the control structure can be reconfigured, and the individual learning modules are made to be parts of the integrated control and decision- making system which dynamically responds to changes in the plant conditions. This time- dependent response results from the enforce- ment of an appropriate control law selected from a bank of pre-designed controllers (4),(5). The possible control laws could be formulated using the time-domain and fre- quency-domain analytical techniques (6),(7), fuzzy logic (8) or simple heuristics. The key idea is that no single controller can satisfy the performance requirements for all conditions of plant operations. The learning agent evaluates the system performance and recon- figures the system by switching to the control- ler that it determines to be most likely to achieve the desired performance at the given instant. However. poor performance may result due to incorrect identification of the plant operating c onditions, imprecise