基于拓扑分解训练的配电网协同超分辨率量测生成算法

Collaborative generation of super-resolution measurements algorithm for distribution networks through topological decomposition training

配电网超分辨率量测生成算法可以利用有限量测产生时空稠密量测,提升配电网状态感知能力。大规模配电网超分辨率量测生成模型结构复杂、训练耗时长,影响其实际应用。提出了一种基于拓扑分解训练的协同超分辨率量测生成方法,该方法先将大规模配电网拓扑分解为多个分区系统和协调侧系统,分别训练超分辨率模型;再汇总分区系统超分辨率模型并固定其参数,采用分解协调算法微调协调侧系统超分辨率模型参数。应用阶段,各分区系统模型先独立生成超分辨率量测,再由协调侧系统超分辨率模型更新边界节点量测,完成全配电网超分辨率量测生成。采用IEEE-33和IEEE-123节点系统开展仿真测试,验证了所提方法具有模型训练快、计算代价小和量测生成精度高等特点,适用于大规模配电网全局状态感知。

Super-resolution(SR)measurement generation algorithms for distribution networks can generate spatiotemporally dense measurements with limited actual measurement data,thereby improving the capability of state awareness.SR models of large-scale distribution networks is complex in structure and requires a long training time,which affects their practical applications.A collaborative SR measurement generation method based on topological decomposition training is proposed.Firstly,the large-scale distribution network is decomposed into multiple sub-systems and coordination systems.The SR model of each sub-system and each coordination system is trained separately.Subsequently,the SR models of all the sub-systems are gathered and their parameters are fixed.The parameters of the SR models of coordination systems are fine-tuned with a decomposition and coordination algorithm.During the application phase,each sub-system’s SR model first independently generates SR measurements,then the coordination system's SR model updates boundary node measurements to complete the super-resolution measurement generation for the entire distribution network.Finally,in test cases of IEEE 33-bus and IEEE 123-bus systems,it is verified that the method can reduce the computational resource occupation and model training time while improving the accuracy of super-resolution results for large-scale distribution networks.