基于联邦学习的分布式电采暖互动模式设计与展望

Design and prospect of distributed electric heating interactive mode based on federated learning

随着“双碳”目标的提出,以及“以电代煤”政策的贯彻落实,大量电采暖设备取代传统燃煤取暖投入运行并接入电网将成为必然趋势。大量电采暖设备可以作为需求侧可调资源进行新能源消纳,但是分布式电采暖所处地理区域较为分散,传统集中式管理的方式又存在隐私泄露、数据孤岛等问题。联邦学习作为一种分布式技术可在保护隐私的前提下支撑电采暖负荷互动,在分布式电采暖互动领域具有较强的适用性。分析了基于联邦学习的分布式电采暖互动需求,以及边缘缓存、隐私防护、通信传输优化和异构资源融合等技术在基于联邦学习的电采暖互动场景中的应用方式,并展望了未来基于联邦学习的分布式电采暖互动前景。

With the introduction of the dual-carbon target and the implementation of the"replacing coal with electricity"policy,substantial electric heaters are bound to be connected to the power grid and replace the traditional coal-fired heaters.The electric heater can be used as demand-side adjustable resources for new energy consumption.With regard to their management methods,distributed electric heaters are geographically scattered,while the traditional centralized heaters are vulnerable to privacy leakage and data islands.As a distributed technology,federated learning can support the interaction of electric heating loads under the premise of protecting privacy,and has strong applicability in the field of distributed electric heating interaction.In the analysis on the requirements of distributed electric heating interaction based on federated learning,the applications of edge caching,privacy protection,communication transmission optimization and heterogeneous resource fusion in the interactive modes are expounded.The prospect of distributed electric heating interactive modes based on federated learning is made.