•  
  •  
 

Keywords

multi-time scale control; new energy grid integration; robust reinforcement learning; reinforcement learning; voltage control; Markov decision process

Abstract

To address the problems of frequent voltage fluctuations and significantly increased voltage control complexity caused by high-proportion new energy connected to the power grid, a voltage control strategy considering fast response and fine regulation capabilities is constructed for large-scale new energy collection areas, combined with the requirements of multi-time scale voltage regulation, to realize the joint control between devices with discrete response characteristics (CB) and devices with continuous regulation capabilities (SVC). According to the day-ahead expected results, coarse adjustment of discrete reactive power devices is realized by this strategy in the day-ahead stage; robust reinforcement learning is utilized in the real-time stage to achieve fast adjustment of continuous reactive power devices. First, a non-convex nonlinear stochastic optimization model considering both voltage deviation and economic cost as the objective function is established, and calculation efficiency is improved through model simplification and convex optimization processing. Secondly, in view of the uncertainty of new energy output, an objective function based on the economic cost of continuous device regulation and the improved voltage distribution index is established; a Markov decision process is constructed, and solving is performed by a real-time voltage control model based on robust reinforcement learning to achieve fast response in the real-time stage. Finally, it is verified through an improved case that the proposed strategy effectively improves voltage control accuracy and response speed of the power grid under new energy fluctuation scenarios and possesses strong engineering practicability and promotion value.

DOI

10.19781/j.issn.1673-9140.2026.02.021

First Page

236

Last Page

245

Download

Share

COinS