How to improve the optimization of recharging schedules with Neutronet: experience and multi-objective function

The search and optimisation process for PWR loading pattern designs is inherently complex and resourceintensive, requiring substantial time and computational capabilities. Traditionally, this process has relied on an intuitive and iterative approach largely driven by the expertise of a nuclear designer. Multiple loading pattern configurations would be evaluated using a neutronic code until a core design was identified that met the plant’s energy production goals and operational constraints, while ensuring all safety requirements were satisfied. This approach, although effective, involved considerable effort and demanded a high level of specialized experience. In this context, and with the objective of improving efficiency without compromising design standards, ENUSA developed Neutronet.

From a manual process to intelligent automation

Neutronet is a tool that integrates an artificial neural network (ANN) with an optimization algorithm to accelerate the search for suitable loading patterns. Since its development some years ago, it has been employed as a support tool for PWR reload design studies in several Spanish nuclear power plants, achieving highly satisfactory results.

This work builds upon ENUSA’s accumulated experience with Neutronet in PWR loading pattern design. Based on this operational insight, enhancements to the tool are proposed to better address the inherent complexity of core design. In particular, to increase Neutronet’s flexibility, the implementation of multiobjective functions is explored, enabling the simultaneous optimisation of multiple key variables in the PWR core design process.

Toward more efficient multi-objective optimization

The scope of this study includes an analysis of Neutronet’s performance, illustrated through its application in the search for a specific reload project. The report also details the development and implementation of different multiobjective functions designed to optimise two parameters concurrently. Finally, the impact of these new objective functions on both the efficiency and the effectiveness of the loading pattern search process is evaluated. The proposed improvements aim to enhance Neutronet’s precision and adaptability, thereby contributing to a more streamlined and robust coredesign workflow.