A Reconstruction Method for Missing Data in Power System Measurement Based on LSGAN

The integrity of data is an essential basis for analyzing power system operating status based on data. Improper handling measurement sampling, information transmission, and storage can lead to loss, thus destroying the hindering mining. Traditional imputation methods are suitable low-latitude, low-missing-rate scenarios. In high-latitude, high-missing-rate scenarios, applicability traditional in doubt. This paper proposes a reconstruction method missing LSGAN (Least Squares Generative Adversarial Networks). designed train unsupervized learning mode, enabling neural network automatically learn data, distribution patterns, other complex correlations that difficult model explicitly. It then optimizes generator parameters using constraint relations implied by true sample trained Generator generate highly accurate reconstruct proposed approach entirely data-driven does not involve mechanistic modeling. still case high latitude loss rate. We test effectiveness comparing three GAN derivation our experiments. experimental results show feasible effective, accuracy reconstructed higher while taking into account computational efficiency.