International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 11 Issue: 10 | Oct 2024
p-ISSN: 2395-0072
www.irjet.net
Development of AI model for Reactivity estimation of NPPs Rahul Raj1, P. Parimalam2 1Scientific officer C , Indira Gandhi Centre for Atomic Research, Kalpakkam, Tamil Nadu, India
2Scientific officer G , Head- DDCSS, Indira Gandhi Centre for Atomic Research, Kalpakkam, Tamil Nadu, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - The reactivity of nuclear power plants (NPPs)
Among ML techniques, Deep Neural Networks (DNNs), particularly Long Short-Term Memory (LSTM) networks, have demonstrated superior performance with lower prediction errors compared to traditional methods. To train these models, simulated datasets of neutron concentration versus reactivity are generated using input reactivity values ranging from +20 pcm to -20 pcm. The datasets incorporate three types of reactivity inputs: step, ramp, and sinusoidal reactivity.
is a critical parameter, essential for ensuring the safe and secure operation of these systems. In Pressurized Heavy Water Reactors (PHWRs), reactivity measurement typically relies on the Inverse Point Kinetics Equation (IPKE), a deterministic approach based on predefined mathematical formulations, which can be rigid in practice. This paper presents an alternative methodology using Machine Learning (ML), a subset of Artificial Intelligence (AI). ML models offer a data-driven approach for reactivity estimation, providing greater flexibility by learning from historical data. Several ML models are explored, including Linear Regression, Regression Trees, Random Forest, Support Vector Regression (SVR), and Recurrent Neural Networks (RNNs), particularly Long Short-Term Memory (LSTM) networks. Simulated datasets of reactivity versus neutron concentration, generated using Point Kinetics Equations (PKEs), are employed to train the ML models. After training, the models are evaluated on testing datasets, and their performance is assessed using Mean Squared Error (MSE), which quantifies the squared deviation between predicted and actual values. A comparative performance analysis based on MSE values identifies the most optimal ML model for accurate reactivity estimation.
The generated datasets are used to train multiple ML models, and their performance is evaluated through comparative analysis using random test datasets. This study aims to assess the feasibility of ML-based reactivity estimation as a complementary or alternative approach to the conventional IPKE method, providing insights into the potential of ML models for enhancing the safety and control of FBR operations. AI/ML as an Alternative to Conventional Methods. Conventional models like IPKE offer a deterministic, physics-based approach to reactivity estimation but are limited by their assumptions, rigidity, and sensitivity to input parameter inaccuracies. They are reliable within well-defined conditions but struggle with dynamic and complex scenarios. AI/ML models such as LSTMs provide a flexible, data-driven alternative that excels in handling time-dependent, non-linear, and noisy data. They learn directly from historical patterns, are more adaptable to changing conditions, and offer robust performance in scenarios where conventional models may fall short.
Key Words: Artificial intelligence (AI), Reactivity estimation, Machine Learning (ML), LSTM, Point Kinetics Equation (PKE), Nuclear Power Plants (NPPs).
1. INTRODUCTION Fast Breeder Reactors (FBRs) are a class of nuclear reactors that produce more fissile material than they consume, making them a critical component of sustainable nuclear energy production. However, the inherently dynamic behavior of FBRs poses significant challenges for reactor control and safety management, necessitating precise and timely reactivity estimation to ensure safe and efficient operation.
2. RELATED WORKS Ma and Bao [1] document a project by Idaho National Laboratory for the Nuclear Regulatory Commission (NRC), investigating the use of advanced computational tools, including artificial intelligence (AI) and machine learning (ML), in nuclear plant operations. The report establishes connections between statistics and AI/ML, highlights commonly used supervised and unsupervised learning algorithms, and explores their applications in enhancing the safety and efficiency of nuclear plants and advanced reactors. Saurabh [2] presents research on applying LSTMRNN models for stock price prediction, focusing on identifying the most effective optimizers for these models. Suman [3] discusses the complexities of nuclear reactors and the challenges of developing real-time AI models, primarily due to the lack of reliable datasets and the
Traditionally, reactivity in FBRs is estimated by monitoring power changes through the solution of Inverse Point Kinetics Equations (IPKE), which account for six groups of delayed neutrons. This deterministic approach relies on predefined mathematical models, limiting adaptability in dynamic conditions. In contrast, Machine Learning (ML) models provide a data-driven alternative, offering greater flexibility by learning from historical data trends.
© 2024, IRJET
|
Impact Factor value: 8.315
|
ISO 9001:2008 Certified Journal
|
Page 870