Scholarlink

<p>This paper presents a data-driven Intelligent Inner Control Loop (IICL) for voltage-controlled inverters in islanded AC microgrids, employing a Deep Neural Network (DNN) as a high-performance alternative to traditional Proportional-Resonant (PR) controllers. Integrated within the primary layer of the microgrid's hierarchical control architecture, the proposed DNN-based controller is trained offline on clustered and preprocessed voltage and current data, capturing the system's dynamic trajectory toward stability under diverse resistive-inductive (RL) load conditions. The network—structured as a five-layer fully connected feedforward architecture with skip connections to facilitate gradient flow—learns complex system dynamics without requiring any explicit plant model or manual gain tuning. Once deployed, it generates real-time PWM reference voltages using only local measurements, enabling fully decentralized, communication-free operation. Simulation results in MATLAB/Simulink demonstrate that the DNN controller significantly improves dynamic and steady-state performance: it reduces voltage tracking root mean square error (RMSE) to 2.065&#xa0;mV, lowers THD from 3.75% to 0.3%, and shortens settling time from 0.1&#xa0;s to 0.03&#xa0;s. Additionally, it ensures stable voltage regulation and accurate active/reactive power sharing within 5% deviation—even in the absence of secondary control. By generalizing across operating conditions and adapting to nonlinearities and disturbances, the proposed approach not only replaces the PR controller but also outperforms it across various operating conditions, offering a more robust and generalizable solution for microgrid inverter regulation.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Deep Neural Network-Based Inner Loop Control for Improved Power Sharing and Voltage Regulation in AC Microgrids

  • 期刊论文

  • 发表日期：

    2025年11月10日

查看全文
  • Afshin Hasani,
  • Hossein Heydari

摘要

This paper presents a data-driven Intelligent Inner Control Loop (IICL) for voltage-controlled inverters in islanded AC microgrids, employing a Deep Neural Network (DNN) as a high-performance alternative to traditional Proportional-Resonant (PR) controllers. Integrated within the primary layer of the microgrid's hierarchical control architecture, the proposed DNN-based controller is trained offline on clustered and preprocessed voltage and current data, capturing the system's dynamic trajectory toward stability under diverse resistive-inductive (RL) load conditions. The network—structured as a five-layer fully connected feedforward architecture with skip connections to facilitate gradient flow—learns complex system dynamics without requiring any explicit plant model or manual gain tuning. Once deployed, it generates real-time PWM reference voltages using only local measurements, enabling fully decentralized, communication-free operation. Simulation results in MATLAB/Simulink demonstrate that the DNN controller significantly improves dynamic and steady-state performance: it reduces voltage tracking root mean square error (RMSE) to 2.065 mV, lowers THD from 3.75% to 0.3%, and shortens settling time from 0.1 s to 0.03 s. Additionally, it ensures stable voltage regulation and accurate active/reactive power sharing within 5% deviation—even in the absence of secondary control. By generalizing across operating conditions and adapting to nonlinearities and disturbances, the proposed approach not only replaces the PR controller but also outperforms it across various operating conditions, offering a more robust and generalizable solution for microgrid inverter regulation.