Noise-Trained Deep Learning-Based Distribution System State Estimation Considering the Penetration of Distributed Energy Resources

Abstract

Lack of measurements has always been one big challenge in Distribution System State Estimation (DSSE). The increased local measurements from Distributed Energy Resources (DERs), such as the measurements from smart meters, Photovoltaic (PV) inverters and battery inverters, pose a potential solution, but have not been explored sufficiently so far. Meanwhile, these local measurements in the real world are usually carried with anomaly data points, such as extremely large and extremely small values, missing data etc. Therefore the accuracy of conventional weighted least square (WLS)-based DSSE is challenged by the penetration of DERs and the abnormal measurements. Furthermore, WLS-based DSSE can only be applied to over-determined systems. This paper proposes a deep learning-based DSSE algorithm that is trained with noise. The algorithm aims to learn the relationship between the measurements and system states, especially when the measurements are taken with extreme values or missing data. This is particularly important when there is an under-determined system and high penetration of DERs. Case studies demonstrate that the proposed model can effectively estimate the system states under different levels of abnormal and missing data interference, and can also provide reliable system states when considering reverse power flow caused by DER penetrations.