Accommodating power plant anomalies via an artificial neural network-based reconfigurable control

A reconfigurable control strategy has been developed using a feed-forward Artificial Neural Network (ANN) implemented on a Bailey Network 90 digital control system requiring relatively minor modifications to the plant controller. An experimentally verified simulation of a secondary system's deaerating feedwater heater of a nuclear power plant was used to demonstrate the feasibility of the concept. The controller uses the ANN to decide when to switch from normal level control to alternate pressure control (via the level control valves) in the event of a rapid pressure decrease, possibly due to loss of feedwater heating, i.e., a phenomenon known as deaerator quenching. A prolonged rapid pressure decrease would result in feedpump cavitation (i.e., Net Positive Suction Head less than zero) due to the change in pressure reaching the pump suction before the corresponding change in saturation temperature. The ANN uses the deaerator level, pressure, and rate of change of pressure to develop a system status signal (i.e., reconfiguration switch signal) between 0.0 and 1.0. The existing level controller, with some minor modifications, is reconfigured to a pressure controller upon receiving a switch signal of 0.485. When tested with a simulated pressure regulator malfunction the reconfigurable control scheme displayed superior performance (compared with the original system) by arresting the pressure decrease and maintaining NPSH above zero for a prolonged period of time.