![]() | Only 14 pages are availabe for public view |
Abstract Nuclear Power Plants (NPPs) are indispensable for maintaining a steady supply of electricity and reducing the need for frequent refuelling interruptions, constituting a substantial portion of global electricity generation. Among the various types of NPPs, Pressurized Water Reactors (PWRs) are widely utilized, relying on robust safety mechanisms for their operation. Within PWRs, the Pressurizer (PZR) unit holds particular significance, serving crucial functions such as regulating primary coolant loop pressure, managing coolant temperature, and stabilizing the reactor during transients or accidents to ensure the integrity of the containment system. Given the pivotal role of the PZR in PWR systems, there is a critical need to develop accurate modeling and control techniques. In this context, this thesis aims to investigate and develop modeling and control techniques specifically tailored for the PZR, with a special emphasis on leveraging Artificial Intelligence (AI)-based methods. Initially, the thesis focuses on developing a mathematical model capable of accurately forecasting pressure and water level dynamics within the PZR under normal operation and during load power changes. This model is constructed as a non-equilibrium three-region model grounded in the principles of thermodynamics, encompassing mass and energy interactions within the water and steam phases of the PZR. This model integrates thermodynamic principles and accounts for multiple factors affecting pressure and temperature. Validation using simulator data demonstrates its effectiveness. To validate the accuracy of the developed model, data generated from the Water-Water Energetic Reactor (VVER-1200) simulator, a widely used simulation platform for nuclear reactors, is utilized for parameter estimation, verification, and validation, particularly during load power changes. Comparative analysis with existing models documented in the literature demonstrates the superior sensitivity and performance of the developed model. To explore the application of the developed model in control systems, an analytical Fuzzy Proportional-Integral-Derivative (FPID) controller with various configurations is designed to regulate and control the pressure and water level of the PZR system. Stability analysis of the FPID controllers with variable gains is conducted, and conditions for Bounded-Input Bounded-Output (BIBO) stability conditions are derived using the small gain theory. Various scenarios are employed to assess the dynamic response of the applied controllers under different operating conditions. Additionally, performance indices are compared between the developed intelligent controller and conventional Proportional-Integral-Derivative (PID) controllers. Building upon the developed mathematical model, the thesis explores the application of AI-based techniques, specifically a TakagiSugano-Kang Fuzzy Neural Network (TSKFNN), for predicting pressure and water level dynamics within the PZR. The identification of the TSKFNN model involves multiple steps, including structure and parameter identification. The Fuzzy C-Means (FCM) clustering method is utilized to separate input data into clusters and obtain rule antecedent parameters, while kernel ridge regression defines initial rule consequent parameters. Subsequently, the sliding-window Kernel Recursive Least Squares (KRLS) algorithm, in conjunction with the gradient method, adapts TSKFNN model parameters. Utilizing input/output data from the VVER-1200 simulator, the TSKFNN model is trained, tested, and evaluated, demonstrating its effectiveness in predicting PZR behaviour. The simulation results further validate the efficiency of the TSKFNN model, particularly when compared with existing mathematical models. Furthermore, the thesis delves into the self-organizing aspect of the TSKFNN model, which enables automatic adjustment of fuzzy rules based on the complexity of input data. Simulation results confirm the viability of the self-organized TSKFNN model for predicting pressure and water level dynamics. Additionally, the thesis explores the utilization of the self-organized model to identify and adapt a Controlled Auto Regressive Integrated Moving Average (CARIMA) model, commonly employed in control approaches such as Generalized Predictive Control (GPC). This adapted CARIMA model serves as the basis for designing a GPC controller for the PZR system. Finally, various scenarios are employed to assess the dynamic response of the designed GPC controller. Comparative analysis is conducted with other control methods, including FPID controllers and conventional PID controllers, to evaluate the effectiveness and superiority of the GPC controller. The simulation results provide insights into the performance of different control strategies and underscore the potential of AI-based methods in enhancing the safety and efficiency of nuclear power plant operations. |