A NEURAL NETWORK FOR THE CONTROL OF THE FEEDWATER SYSTEM AND STEAM WATER LEVEL IN A PRESURIZED WATER REACTOR

THIS RESEARCH APPLIES NEURAL NETWORK TECHNOLOGY TO THE MONITORING AND CONTROL OF THE FEEDWATER SYSTEM AND RESULTANTSTEAM GENERATOR WATER LEVEL DURING STARTUP AND LOW POWER OPERATIONS OF LARGE PRESSURIZED WATER REACTORS. DURING THESE OPERATIONAL MODES AND PLANT TRANSIENTS, CONTROL OF THEFEEDWATER SYSTEM IS DIFFICULT. THE NEURAL NETWORK WOULD BE USED TO MONITOR THE INPUT SIGNALS USED BY THE EXISTING AUTOMATIC CONTROL SYSTEM AND THEN "LEARN" THE RESPONSE CHARACTERISTICS OF THIS SYSTEM. IN THIS MANNER, THE NEURAL NETWORK WOULD BE TRAINED OVER A PERIOD OF TIME TO EXPECT THENORMAL RESPONSE SIGNALS AND DETECT ANY OFF-NORMAL SIGNALS FROM THE FEEDWATER SYSTEM OPERATION. PERSONNEL WOULD BE NOTIFIED IN ADVANCE OF ANY OFF-NORMAL ACTIVITY. THIS NEURAL NETWORK APPLICATION WOULD IMPROVE THE CONTROL OF THE FEEDWATER SYSTEM DURING STARTUP, LOW POWER OPERATIONS, AND PLANT TRANSIENTS. THIS WOULD REDUCE THE NUMBER OF FEEDWATERRELATED REACTOR TRIPS. DURING PHASE I, A THOROUGH STUDY OF THE SENSOR REQUIREMENTS WILL BE MADE TO ESTABLISH THE NEURALNETWORK ARCHITECTURE. A PROTOTYPE SUBSET OF THE SYSTEM WILLBE SIMULATED AND DEMONSTRATED USING AN IBM PC-AT. IMPELL CORPORATION AND COMMONWEALTH EDISON WILL PARTICIPATE IN DEVELOPING THE SYSTEM AND EVALUATING THE RESULTS. IN PHASE II, A FULL SCALE SYSTEM WILL BE DEVELOPED TO SIMULATE AND TEST THE APPLICATION OF THIS TECHNOLOGY TO THE FEEDWATER CONTROL SYSTEM.