Linear Load Model For Robust Power System Analysis

Extension of constant power (PQ) load models to more accurately represent the electric load behavior in the grid has produced models (e.g. ZIP) that have been shown to improve the accuracy of load characterization, but like PQ load models, they introduce nonlinearities in the power flow formulation that make it more susceptible to divergence or convergence to a non-physical solution. In this paper a first-order load model (BIG) based on equivalent circuit principles is proposed that offers accuracy comparable to a ZIP model, but unlike ZIP, is linear and captures angle information when used in a current-voltage based power flow formulation. Advanced machine learning techniques are applied to fit parameters for the BIG model and traditional models using time series measurement data from our university campus and from mu PMUs installed at Lawrence Berkeley National Laboratory. The results show that the linear BIG model characterizes the load behavior far better than PQ load models while having a similar fit to that of more complex non-linear ZIP load model while offering complexity and modeling benefits.