Data-driven Stochastic Model Predictive Control Method for Voltage and Power Regulation in Active Distribution Networks

Siyun Li,Fan Zhang,Wenchuan Wu,Wuhua Hu

DOI: https://doi.org/10.1109/pesgm51994.2024.10688952

2024-01-01

Abstract:High penetration of distributed renewable energy has introduced significant uncertainties to active distribution networks (ADNs). What's more, the lack of accurate system model parameters has posed challenges in implementing conventional model-based control methods. To address these problems, this paper presents a data-driven voltage and power regulation method based on stochastic model predictive control (SMPC) method for ADNs, which doesn't rely on accurate system model parameters. To incorporate stochastic control strategies, the prediction errors of distributed renewable energy generation and system loads are characterized using the Gaussian mixture model (GMM). To approximate the system model from data with a good balance between simplicity and expressiveness, we leverage a piecewise linear (PWL) regression method combing the support vector linear regression (SVR) model, which can overcome the issues of measurement outliers and data collinearity in traditional PWL regression methods. With the PWL model trained offline, a novel data-driven SMPC model is proposed, where the coordination of fast and slow regulating resources and the uncertainties of prediction errors are well considered. The effectiveness of the proposed method is demonstrated via numerical simulations.