Artificial Intelligence Applications in Nuclear Power Systems

Artificial intelligence is increasingly integral to the design, operation and safety of nuclear power systems. Machine learning methodologies are employed across the plant life cycle, from reactor core optimisation and in-core fuel management to predictive maintenance and real-time monitoring. Digital twins—virtual replicas of physical systems driven by sensor data—enable accurate prognosis of equipment degradation and support condition-based maintenance, thereby reducing unplanned outages. Convolutional neural networks and recurrent architectures such as long short-term memory networks facilitate early detection of anomalies in thermal-hydraulic parameters and enable rapid prognosis of accident sequences. Reinforcement learning frameworks are under investigation for autonomous control of start-up, load-following and cold shutdown procedures, aiming to reduce operator workload and enhance operational resilience. Human-in-the-loop decision-support platforms, integrating hidden Markov models with adaptive intervention strategies, improve situational awareness under complex scenarios. Crucially, interpretable AI methods incorporating tools like Shapley Additive Explanations ensure transparency in risk assessment and regulatory compliance. Collectively, these innovations promise to bolster safety margins, optimise fuel utilisation and advance the global transition to low-carbon energy generation.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Research from all publishers

Recent studies have advanced interpretable frameworks for severe-accident forecasting, combining gated recurrent units with Shapley Additive Explanations to predict core parameter trends during loss-of-coolant and main steam-line break scenarios, thereby strengthening emergency response strategies. Investigations into deep learning for reactor safety have evaluated model reliability and explainability, developing neural-network surrogate models to predict turbulent eddy viscosity in computational fluid dynamics simulations and applying local interpretable model-agnostic explanations to elucidate predictions. A novel zigmoid-based long short-term memory architecture has been introduced for multivariate time-series analysis of loss-of-coolant accidents, enhancing long-term memory retention and delivering improved accuracy in post-accident parameter prediction across multiple system variables.

Artificial Intelligence Applications in Nuclear Power Systems publication trend

The graph below shows the total number of articles in artificial intelligence applications in nuclear power systems across all publications each year (not limited to Nature Index journals).

Technical terms

Digital twin: A virtual representation of a physical system that synchronises with real-time sensor data to simulate and predict system behaviour.

Long short-term memory (LSTM): A recurrent neural network architecture that captures long-range dependencies in sequential data through gated memory cells.

Gated recurrent unit (GRU): A simplified recurrent neural network variant that uses gating mechanisms to control information flow and maintain sequence context.

Shapley Additive Explanations (SHAP): A model-agnostic interpretability technique that assigns each feature an importance value for individual predictions based on cooperative game theory.

Reinforcement learning: A machine learning paradigm in which an agent learns to make decisions by receiving rewards or penalties through trial-and-error interactions with an environment.

References

1. Analyzing Operator States and the Impact of AI-Enhanced Decision Support in Control Rooms: A Human-in-the-Loop Specialized Reinforcement Learning Framework for Intervention Strategies. International Journal of Human-Computer Interaction (2024).
2. Deep learning for safety assessment of nuclear power reactors: Reliability, explainability, and research opportunities. Progress in Nuclear Energy (2022).
3. An Overview of AI Methods for in-Core Fuel Management: Tools for the Automatic Design of Nuclear Reactor Core Configurations for Fuel Reload, (Re)arranging New and Partly Spent Fuel. Designs (2019).
4. Comparison of Deep Reinforcement Learning and PID Controllers for Automatic Cold Shutdown Operation. Energies (2022).
5. Multivariate Time Series Prediction for Loss of Coolant Accidents With a Zigmoid-Based LSTM. Frontiers in Energy Research (2022).
6. An Interpretable Time Series Data Prediction Framework for Severe Accidents in Nuclear Power Plants. Entropy (2023).