DrivAerNet等五个工程数据集
收藏资源简介:
本研究构建了九个新的工程连续学习基准,这些基准基于五个代表性的3D工程数据集,包括DrivAerNet数据集,用于评估不同连续学习策略在工程代理模型任务中的表现。这些数据集模拟了新设计和新约束下工程数据的发展,要求模型能够随着时间的推移整合新知识。数据集的创建是为了解决传统代理模型方法在动态环境下的局限性,通过持续学习来实现更高效的模型更新,避免从零开始的重训练。这些数据集的应用领域主要是工程设计,旨在解决随着新数据、新约束或性能目标的出现,模型必须适应数据分布变化的问题。
This study constructs nine novel engineering continual learning benchmarks, which are developed based on five representative 3D engineering datasets including the DrivAerNet dataset, to evaluate the performance of different continual learning strategies on engineering surrogate model tasks. These datasets simulate the evolution of engineering data under novel designs and constraints, requiring models to integrate new knowledge over time. The datasets are created to address the limitations of traditional surrogate model methods in dynamic environments, enabling more efficient model updates via continual learning and avoiding retraining from scratch. These datasets are mainly applied in engineering design, aiming to solve the problem where models must adapt to changes in data distribution as new data, constraints or performance objectives emerge.




