PMU measurements enabled real-time adaptive load modeling framework

Abstract

Accurate load modeling plays a critical role in analyzing dynamic voltage stability events, such as Fault-Induced Dynamic Voltage Recovery (FIDVR), which may escalate into system-wide instability. Two important considerations in this context are the structure of the load model and the method used for parameter estimation. While significant progress has been made in developing advanced load models that account for renewable integration, demand response, etc., parameter estimation techniques remain comparatively underexplored. To address this gap, a Statistical Similarity Index (SSI)-based adaptive windowing scheme is introduced to identify faulty measurement intervals and, in turn, guide the selection of an appropriate load model structure. Specifically, a composite load model is employed during transient conditions, whereas a simpler ZIP model is used under steady-state conditions. Parameter estimation is carried out using Adam’s Stochastic Gradient Descent (ASGD) to obtain initial values, which are subsequently refined through Sequential Quadratic Programming (SQP). Simulation studies on a two-bus test system and the IEEE 39-bus network demonstrate the superior performance of the proposed method compared to conventional nonlinear algorithms, such as the Levenberg–Marquardt algorithm. Furthermore, real-time validation on RTDS and dSPACE1104 platforms confirms its suitability for practical applications.