LDM: A Generic Data-Driven Large Distribution Network Operation Model

With the growing intelligence of power grids, the application of data-driven AI technologies has been widely studied in distribution network (DN) control and operation. However, most existing studies can only address a specific task. The recent surge of powerful, versatile AI models has inspired us to explore whether the grid controller can also evolve toward greater intelligence, enabling it to perform multiple DN operation tasks. To this end, this letter proposes a novel generic data-driven Large Distribution network operation Model (LDM) based on multitask reinforcement learning (MTRL). It can concurrently learn multiple DN operation skills and perform distinct tasks separately. Specifically, to effectively handle the unaligned heterogeneous action spaces across different tasks, action-masking is incorporated. Case studies on a modified 33-bus system prove the generalization capabilities of LDM.