Abstract
A new Multi-Input Multi-Output Hybrid Intelligent (MIMO-HI) system is designed for Primary Pressure Control System (PPCS) of a highly non-linear Pressurized Heavy Water Reactor (PHWR)-type Nuclear Power Plant (NPP). The MIMO hybrid intelligent system is a nonlinear control system which is synthesized using Adaptive Feedforward Neural Network (AFNN) and Mamdani-type Fuzzy Inference System (MFIS). In the proposed Mamdani Adaptive Neuro-Fuzzy System (MANFS), a supervised learning technique has been adopted using AFNN for learning the design parameters of membership functions. A Clustering Algorithm (CA) is used to select the initial number and type of membership functions with each variable. A reduced order nonlinear primary pressure control system is developed based on neural optimization for rule reduction. The MIMO- HI system is designed based on five Multi-Input Single-Output (MISO) adaptive neuro-fuzzy systems in the framework of parallel learning and parallel data processing. These five MISO adaptive neuro-fuzzy systems are designed for intelligent predictions of two feed valves, two bleed valves and one spray valve positions of a PPCS of an operating PHWR-type NPP in Pakistan. A severe transient has been imposed in training, validation and testing phases with proposed MIMO-HI system. The proposed MIMO-HI system is designed in MATLAB and a Graphical User Interface (GUI) is developed for variables transfer and simulations in Visual Basic. The designed MIMO-HI system is observed to be extremely efficient, robust, with reduced oscillations and better transient and steady state characteristics. The performance of the designed MIMO-HI system is tested and the estimated results are in agreement with conventional PIDtype PPCS of PHWR.
