Composite Load Model Parameter Identification with Distributed Machine learning for the Stability Study of Microgrids

Precise load modeling is essential for the study of the dynamic behavior of large power grids and microgrids. This is challenging as many distributed generation units, energy storage elements, and non-linear loads such as power electronic interfaced loads, electric vehicles, and plug-and-play loads have become integrated into modern power systems. A standard load modeling approach is to model the behavior of a cluster of small loads with a spot load so called composite load model (CLM). The CLM has a fixed structure and its parameters are selected such that the behavior of CLM imitate the behavior of cluster of small loads in study of power system dynamic behavior. In order to estimate the CLM parameters different loads parameters are transferred to a central unit to be processed. This, however, increases system vulnerability against cyberattacks and raises privacy concerns for owner of the loads that do not want to disclose all the local load information. In this paper, a new method of estimating CLM parameters in power distribution systems by using distributed machine learning is proposed to correct these issues. The performance of the method is verified with an example using the recognized standard IEEE 37 bus test feeder.