Bayesian model selection and calibration applied to composite load identification

Power system operation, control and planning rely heavily on numerical simulations based on power system models. While generators and transmission lines have been relatively well understood, load is least known since it changes all the time and utilities have little control on. Inaccurate load models may result in unsafe operating conditions, leading to power outages, under-utilization of system capacity, or inappropriate capital investment. The Western Electricity Coordinating Council´s (WECC) Load Modeling Task Force has recommended the use of a composite load model that has among other elements a feeder connecting different types of induction motors, a discharge lighting model, and a constant power load. At the transmission or sub-transmission level, the composite load model can capture the aggregate behavior of loads. Nevertheless, the validation of these models remains a challenging and critical step to ensure model accuracy. This paper presents the use of state-of-the art Bayesian methods for composite load model selection and calibration. The approach aims at identifying configuration and reducing parameters uncertainty of the WECC composite load model in the presence of measured field data. The success of the approach is illustrated with synthetic field data and a simplified model.