PHD Position Efficient Online Optimization for Time

Logo of EURAXESS Delft University of Technology (TU Delft) Engineering » Control engineering First Stage Researcher (R1) 23 Nov 2026 - 22:59 (UTC) Country Netherlands Type of Contract Permanent Job Status Not Applicable Hours Per Week 38.0 Is the job funded through the EU Research Framework Programme? Not funded by a EU programme Is the Job related to staff position within a Research Infrastructure? No Offer Description How to avoid learning too much - safe and efficient learning for control Job description Inefficient operation of industrial systems, electricity production, and heating systems in buildings is currently responsible for over 50% of overall greenhouse gases emissions. Improving energy efficiency of energy-intensive sectors by implementing advanced control systems has the potential to both improve efficiency and reduce the costs, all while continuing to meet safety requirements. The flexibility of safe and adaptive learning control methods makes them crucial for safety-critical applications. A key difficulty in these learning-based approaches lies in the fact that the algorithm must simultaneously learn the model and guarantee safety. This dual requirement significantly impacts the computational cost of the learning algorithm, particularly in situations where the optimal decision evolves over time. We now aim to develop computationally efficient safe learning methods for time-varying optimal control problems. Existing safe learning algorithms usually assume that they have unlimited access to the experiments and thus unlimited new data. Approaches using limited data sets have been restricted to Bayesian optimization without safety constraints. Moreover, the algorithms typically rely on the repetitive nature of the experiments, for instance in robotics or precision manufacturing. The unlimited data combined with repetitiveness limit the applicability of safe learning algorithms in problems evolvi...

EURAXESS 标志 代尔夫特理工大学（TU Delft） 工程 » 控制工程 第一研究阶段研究员（R1） 2026年11月23日 - 22:59（UTC） 国家 荷兰 合同类型 永久 工作状态 不适用 每周工时 38.0 该职位是否由欧盟研究框架计划资助？ 不由欧盟计划资助 该职位是否与研究基础设施内的员工职位相关？ 否 职位描述 如何避免学得过多——安全高效的控制学习 职位描述 工业系统、电力生产和建筑供暖系统的低效运行目前造成了超过50%的温室气体总排放。通过实施先进控制系统来提高能源密集型行业的能效，有可能在继续满足安全要求的同时，既提高效率又降低成本。安全自适应学习控制方法的灵活性使其对安全关键应用至关重要。这些基于学习的方法的一个关键困难在于，算法必须同时学习模型并保证安全性。这一双重需求显著影响学习算法的计算成本，尤其是在最优决策随时间演变的情况下。我们现在旨在为时变最优控制问题开发计算高效的安全学习方法。现有的安全学习算法通常假设它们可以无限制地进行实验，因而拥有无限的新数据。使用有限数据集的方法仅限于没有安全约束的贝叶斯优化。此外，这些算法通常依赖于实验的重复性，例如在机器人技术或精密制造中。无限数据与重复性相结合，限制了安全学习算法在演变中的问题中的适用性……

⚑ 岗位信息收录自公开招聘页（公开招聘平台），以雇主官网为准