Abstract Designing modern control systems is a critical factor in producing rigid control and effective performance of non-linear and complex systems—particularly employing intelligence-based controllers, derived from soft computing algorithms. It is known to be well capable of exploiting tolerances for uncertainty and nonlinearity. Fuzzy logic, which utilizes soft computing techniques, deals with vague information and approximate reasoning, yet it lacks effective learning capability, while ANN is known for exceptional learning and adaptation of training data. Adaptive neuro-fuzzy interference system (ANFIS), a hybrid of these two algorithms, is a fuzzy interference system optimized by neural networks. Most nuclear generating stations still use PID controllers, which does not show good controlling capabilities, resulting in large coolant level overshoot and longer settling time. This paper bridges this gap by testing an ANFIS approach that yields high predictive and control capability. In validating the configured control loop model, three loss-of-load transients from a Shippingport reactor were tested, trained and demonstrated using the simulations in the MATLAB/Simulink environment. The minimal training RMSE for the 51 MW(e), 74 MW(e), and 105 MW(e) loss-of-load transients are 18.85%, 19.37%, and 24.36%. The pressure response showed ideal output, with all three steady-state pressures close at setpoint (14 MPa), proving that the ANFIS controller is well capable of rejecting disturbances with lesser overshoot and faster settling times in maintaining level setpoint. Validating solely the pressurizer model was executed in a turbine leading power trip from a 100% to 75% reduction range in PCTRAN, a pressurized water reactor (PWR) simulator made available by IAEA. The comparative performance exhibited a good fit between the reported simulation data and model data. Both results against simulation and experimental transients captured the effective predictive and control capabilities of the pressurizer unit and control system.
Read full abstract