We propose RHEA CL, which combines Curriculum Learning (CL) with rollinghorizon Evolutionary algorithms (RHEA) to automatically produce effective curricula during the training of a reinforcement learning agent. RHEA ...
详细信息
ISBN:
(纸本)9798350350685;9798350350678
We propose RHEA CL, which combines Curriculum Learning (CL) with rollinghorizon Evolutionary algorithms (RHEA) to automatically produce effective curricula during the training of a reinforcement learning agent. RHEA CL optimizes a population of curricula, using an evolutionary algorithm, and selects the best-performing curriculum as the starting point for the next training epoch. Performance evaluations are conducted after every curriculum step in all environments. We evaluate the algorithm on the DoorKey and DynamicObstacles environments within the Minigrid framework. It demonstrates adaptability and consistent improvement, particularly in the early stages, while reaching a stable performance later that is capable of outperforming other curriculum learners. In comparison to other curriculum schedules, RHEA CL has shown to yield performance improvements for the final Reinforcement learning (RL) agent at the cost of additional evaluation during training.
In a scenario characterized by a continuous growth of air transportation demand, the runways of large airports serve hundreds of aircraft every day. Aircraft sequencing is a challenging problem that aims to increase r...
详细信息
In a scenario characterized by a continuous growth of air transportation demand, the runways of large airports serve hundreds of aircraft every day. Aircraft sequencing is a challenging problem that aims to increase runway capacity in order to reduce delays as well as the workload of air traffic controllers. In many cases, the air traffic controllers solve the problem using the simple "first-come-first-serve" (FCFS) rule. In this paper, we present a rollinghorizon approach which partitions a sequence of aircraft into chunks and solves the aircraft sequencing problem (ASP) individually for each of these chunks. Some rules for deciding how to partition a given aircraft sequence are proposed and their effects on solution quality investigated. Moreover, two mixed integer linear programming models for the ASP are reviewed in order to formalize the problem, and a tabu search heuristic is proposed for finding solutions to the ASP in a short computation time. Finally, we develop an IRHA which, using different chunking rules, is able to find solutions significantly improving on the FCFS rule for real-world air traffic instances from Milano Linate Airport.
暂无评论