Core Concepts
Developing PEARL algorithm for multi-objective optimization in engineering design.
Abstract
A novel method, Pareto Envelope Augmented with Reinforcement Learning (PEARL), addresses challenges of multi-objective problems in engineering. PEARL learns a single policy, outperforming traditional methods. It is evaluated on classical benchmarks and practical PWR core Loading Pattern optimization problems. PEARL efficiently uncovers a Pareto front without additional efforts. Future works include sensitivity analysis and extension to complex problems.
Stats
Several versions inspired from deep learning and evolutionary techniques have been crafted.
PEARL specifically the PEAL-NdS variant efficiently uncovers a Pareto front.
Outperforms classical approaches across multiple performance metrics including Hyper-volume.