Scholarlink

<p>This study presents an innovative adaptive control strategy for a three-phase inverter connected to both ideal and non-ideal grids with photovoltaic system. This study examines the limitations of conventional PI-based phase-locked loop controllers which indicate insufficient frequency tracking and elevated total harmonic distortion in non-ideal grid conditions. To overcome these issues, a two-stage weight update method for an adaptive R-recurrent neural network-based phase-locked loop is recommended. In the initial step, RNN is trained offline using historical grid data to initialize the network with robust weight settings. During second stage, the recurrent neural network adjusts its weights in real time to deal with system variations and nonlinearities. This dual-phase method of learning enables accurate control of both DC-link voltage and output current. The controller is developed in MATLAB/Simulink and evaluated in comparison with a conventional PI-based phase-locked loop control system. Simulation results indicate that suggested method exhibits enhanced performance, decreasing total harmonic distortion to 0.53% under ideal grid conditions and 2.26% under non-ideal conditions, compared to 3.20% and 7.32% using PI-based phase-locked loop, respectively. The suggested recurrent neural network (RNN)-based phase-locked loop eliminates the necessity for additional filtering components, offering a resilient and efficient solution for grid-connected photovoltaic inverter systems operating under distortion and uncertainty.</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

Three-phase inverter connection to a non-ideal grid control based on two-stage weight updates tactics for recurrent neural network controller

  • 期刊论文

  • 发表日期：

    2026年1月13日

查看全文
  • Yechun Jin,
  • Jie Li

摘要

This study presents an innovative adaptive control strategy for a three-phase inverter connected to both ideal and non-ideal grids with photovoltaic system. This study examines the limitations of conventional PI-based phase-locked loop controllers which indicate insufficient frequency tracking and elevated total harmonic distortion in non-ideal grid conditions. To overcome these issues, a two-stage weight update method for an adaptive R-recurrent neural network-based phase-locked loop is recommended. In the initial step, RNN is trained offline using historical grid data to initialize the network with robust weight settings. During second stage, the recurrent neural network adjusts its weights in real time to deal with system variations and nonlinearities. This dual-phase method of learning enables accurate control of both DC-link voltage and output current. The controller is developed in MATLAB/Simulink and evaluated in comparison with a conventional PI-based phase-locked loop control system. Simulation results indicate that suggested method exhibits enhanced performance, decreasing total harmonic distortion to 0.53% under ideal grid conditions and 2.26% under non-ideal conditions, compared to 3.20% and 7.32% using PI-based phase-locked loop, respectively. The suggested recurrent neural network (RNN)-based phase-locked loop eliminates the necessity for additional filtering components, offering a resilient and efficient solution for grid-connected photovoltaic inverter systems operating under distortion and uncertainty.