In the United States, Nuclear energy can provide more carbon-free electricity, in comparison with solar and wind energy combined, which makes it a significant player in the fight against climate change. However, the nuclear fleet present in the U.S. has aged, and the operators are under constant pressure to compete with coal and gas-fired plants.
To win against their counterparts, nuclear energy plants must reduce their costs, and the best place to do it is the reactor core, wherein the energy is produced. The fuel rods present in the plants are liable to drive reactions. If they are idealistically placed, they are likely to burn less fuel and would also require less maintenance. Through constant research, engineers have tried to extend these pricey fuel rods' lives but have been unsuccessful in getting a satisfactory result till now.
A team of researchers have taken a big step towards solving this problem by turning the design problem into a game. The team has stated that Artificial Intelligence can be taught to generate dozens of optimal configurations. These results can help extend each rod's life by about 5%, thus saving a regular power plant about $3 million a year. This new system is a significant development in the Nuclear Power Plant and Equipment Market as it can improve the economics of nuclear energy in the United States. It can help in reducing global carbon emissions and may be able to attract the youth towards working in the clean energy sector.
Deep reinforcement learning is an AI technique that has achieved great success at games like Go and chess. The team investigated if they could use this technique to make the screening go faster. Deep reinforcement learning combines deep neural networks, wherein deep neural networks excel at catching patterns within reams of data. In contrast, reinforcement learning uses ties learning to reward signals like winning a game.
In this study, the researchers guided their agents to position the fuel rods under a set of constraints. This would earn them more points as they make a favorable move, one after the other. All the limitations chosen by the team are based on expert knowledge of the physics laws. It was noted that once the rules were wired, neural networks make good decisions.
Researchers are now testing a beta version of the AI system in a virtual scenario that would imitate an assembly within a boiling water reactor. Also, 200 assemblies within a pressurized water reactor are known to be the most common type of reactor globally.