Neural Networks Based Dynamic Load Modeling for Power System Reliability Assessment

Abstract

The reliability of a power system is considered as a critical requirement in planning and operating the system due to the increasing demand for more reliable service with a lower frequency and duration of interruption. Hence, reliability is also considered as a major challenge in the development of future power systems as they become more advanced and complex, making the accuracy of the reliability assessment dependent on several factors such as supply and load modeling. Recent studies on power systems’ reliability and stability have focused on load modeling, where loads are either assumed to be static or dynamic, by introducing significant constraints. However, the emergence of new types of loads necessitates the development of models that can incorporate them with accuracy, as this would facilitate their effective use in flow and stability simulation studies, as well as reliability analyses. In this study, dynamic loads are modeled using a feed-forward neural network where a simulation test bed is developed in MATLAB/Simulink to generate operating data used during training and validating of the neural network model. Subsequently, Electrical Transient Analyzer Program (ETAP) software is used to verify the effect of load modeling on power system reliability assessment platform. Bus 2 of Roy Billinton Test System (RBTS) is employed as a case study to investigate the sensitivity of the reliability indices, such as System Average Interruption Duration Index (SAIDI) and System Average Interruption Frequency Index (SAIFI), on the load modeling technique with mixed loads (dynamics and statics).