MIT Nuclear Reactor Lab researcher Dr. Akshay Dave, along with co-authors Dr. Kaichao Sun and Jarod Wilson, received the Mathematics and Computation Division Best Summary and Presentation Award at the American Nuclear Society Winter 2020 meeting. Their paper, “Deep Surrogate Models for Multi-dimensional Regression of Reactor Power,” was presented virtually to peers in the nuclear industry.

Their paper focuses on developing surrogate models for multi-dimensional power regression of the MIT Research Reactor; using artificial neural networks as surrogates in particular. The work establishes a baseline for the performance of neural networks, their capability to generalize in this context, and highlights findings on hyperparameter optimization. Their contributions include development of an open source package to abstract surrogate model development.

The researchers are currently utilizing their findings towards demonstrating an autonomous framework to control the subcritical MIT Graphite Exponential Pile. Their project addresses the current lack of experimental application of machine learning control in nuclear systems. Working towards autonomous control for next-generation small modular reactors and micro-reactors is essential in addressing the large O&M costs of traditional nuclear power plants.