Reconfigurable control of power plants using learning automata

A deaerating feedwater heater, equipped with a water level controller and a pressure controller, is used to investigate the feasibility of a reconfigurable control scheme for power plants by incorporating the concept of learning automata. The approach uses stochastic automata to learn the current operating status of the plant by dynamically monitoring the system performance and then switching to the appropriate controller on the basis of the observed performance. Simulation results based on a model of the experimental breeder reactor (EBR-II) at the Argonne National Laboratory site in Idaho are presented to demonstrate the efficacy of the scheme. The results show that it is capable of providing a sufficient margin for the net positive suction head at the feedwater pumps under loss of steam flow into the deaerator. Under similar circumstances, the existing controller in the deaerator would be incapable of maintaining the pressure and its decay rate within the safe margins, and so would oblige the plant operator to take additional measures to protect the feedwater pumps.<>