A Fitness Landscape View on the Tuning of an Asynchronous Master-Worker EA for Nuclear Reactor Design
This work addresses the specific challenge of parameter tuning in expensive black-box optimization for nuclear power plant control, representing an incremental improvement in domain-specific applications.
The authors tackled the problem of tuning algorithm parameters for an asynchronous master-worker Evolutionary Algorithm used in nuclear reactor design optimization, showing that fitness landscape analysis can guide mutation parameter tuning based on low-cost feature estimation.
In the context of the introduction of intermittent renewable energies, we propose to optimize the main variables of the control rods of a nuclear power plant to improve its capability to load-follow. The design problem is a black-box combinatorial optimization problem with expensive evaluation based on a multi-physics simulator. Therefore, we use a parallel asynchronous master-worker Evolutionary Algorithm scaling up to thousand computing units. One main issue is the tuning of the algorithm parameters. A fitness landscape analysis is conducted on this expensive real-world problem to show that it would be possible to tune the mutation parameters according to the low-cost estimation of the fitness landscape features.