Work

Dual-horizon Proactive Physics-informed Model-based Reinforcement Learning Control for 10 Mw Offshore Pmsg Wind Energy Systems Under Lvrt Faults and Turbulent Wind

Year: 2026
Type: article
Abstract: Multi-megawatt offshore wind energy conversion systems based on surfacemounted Permanent Magnet Synchronous Generators (PMSGs) face severe operational challenges in maintaining DC-link voltage stability and power quality under simultaneous sub-cycle grid faults and high-turbulence wind regimes.Conventional model-free Reinforcement Learning (RL) controllers rely on reactive error-driven updates, often resulting in delayed transient responses, high-frequency control chatter, and frequent hardware brake chopper activation during Low-Voltage Ride-Through (LVRT) events.Conversely, purely data-driven Model-Based RL (MBRL) approaches require extensive fault-state training datasets and suffer from multi-step prediction drift.To overcome these limitations, this paper proposes a Dual-Horizon Proactive Physics-Informed Model-Based Reinforcement Learning (PI-MBRL) control architecture for a 10 MW offshore PMSG wind energy system.By embedding analytical DC-link and converter state-space equations directly into the predictive rollout loop, the framework evaluates two decoupled look-ahead horizons: a fast 15 ms electrical horizon for subcycle grid voltage sag detection and pre-emptive active power de-loading, and a slow 300 ms aerodynamic horizon for feedforward turbulent wind power gradient smoothing.
Language: English
Open Access status: bronze
OpenAlex ID: W7235264942
FWCI:
Cites: 3
Cited by:
Related to: 10