WECC Composite Load Model Parameter Identification Using Evolutionary Deep Reinforcement Learning

doi:10.1109/TSG.2020.3008730

Return

WECC Composite Load Model Parameter Identification Using Evolutionary Deep Reinforcement Learning

2020-11-01

DOI：10.1109/TSG.2020.3008730

Original

Save

View PDF

Abstract

En 中文

Due to the increasing penetration of distributed energy resources (DERs), the load composition in distribution grids has significantly changed. This inverter-based device has notably different behavior from traditional static and induction motor loads. To accurately represent the combination of static load, induction motor and the emerging inverter-based devices, the composite load model with distributed generation (CMPLDWG) has been developed by Western Electricity Coordinating Council (WECC). Due to the large number of parameters and model complexity, the CMPLDWG model brings new challenges to parameter identification, which is critical to power system studies. To address these challenges, in this paper, a cutting-edge approach inspired by the evolutionary deep reinforcement learning (EDRL) with an intelligent exploration mechanism is innovatively proposed to perform parameter identification for CMPLDWG. First, to extract parameters contributions to dynamic power, parameter sensitivity analysis is conducted using a data-driven feature-wise kernelized Lasso (FWKL). Then, the EDRL with intelligent exploration, which can handle the natural high nonlinearity and nonconvexity of CMPLDWG, is employed to perform parameter identification. In the parameter identification process, the extracted parameter sensitivity weights are innovatively integrated into the EDRL with intelligent exploration to improve discovery efficiency. Finally, the proposed approach is validated using numerical experiments.

Keywords:

WECC composite load model

parameter identification

evolutionary strategy

intelligent exploration

AI Summary

Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.