Knowledge-Guided Machine Learning for Inferring Unknown Measurement Information in Power Distribution Networks

Power system state estimation is a crucial component of the power system, which uses real-time measurement data to estimate the condition of the system. However, in current distribution systems, the limited number of measurement devices and imperfect communication channels result in a significant challenge for state estimation, as real-time measurements are much smaller than the minimum amount required for state estimation. To address this issue, we propose a knowledge-guided machine learning framework for inferring unknown measurement information in power distribution networks. The framework utilizes machine learning methods to complement the missing measurement information and enable full node state calculation. Specifically, we construct a loss function to represent the measurement equations and utilize physical constraints of power flow in the distribution network to design a reasonable physical inconsistency penalty term to optimize the loss function and reduce false measurement errors, achieving accurate full measurement calculation. We test our method on the IEEE-33 node system and perform distribution network state calculation based on completed measurement information using weighted least squares method. The results show that our approach can effectively improve the accuracy of distribution network state calculation.