Supervised learning for more accurate state estimation fusion in IoT-based power systems

Concerned with deploying zero-emission energy sources, reducing energy wasted through transmission lines, and managing power supply and demand, monitoring and controlling microgrids have found utter importance. Accordingly, this paper aims to investigate the efficacy of state estimation fusion for a synchronous generator as well as an induction motor in order to ameliorate system monitoring. A third-order nonlinear state-space model, that operates based on actual input data taken from the Smart Microgrid Laboratory, is assumed for each of the electrical machines. The model parameters are set according to the parameters of the electrical machines. A fusion structure based on the internet of things communication network, which is modified to increase uncertainty, is presented for fusing the state estimates. The data fusion topology is distributed and relies on two data fusion models. The first model is a set of state estimators, referred to as data input-feature output model. The second one fuses the estimators' outputs based on supervised machine learning methods, referred to as feature input-feature output model. The simulation results in MATLAB and Python show the efficiency of linear regression methods compared with other leveraged methods for data fusion. By comparing the results obtained from both simple and complex estimation filters, it can be deduced that combining simple filters, extended Kalman filter in this case, with simple data fusion methods, linear regression in this case, can produce much more accurate results in a short period of time. Besides, this study shows that the averaging operators are unsuitable for estimation fusion by referring to their convexity condition.