Scholarlink

<p>This paper proposes a new robust nonlinear state estimator, termed Trust Region Policy Optimization-based q-Rényi Maximum Correntropy Unscented Kalman Filter (TRPO-MCUKF), for joint estimation of power system harmonic amplitudes, phases, frequency, and noise parameters. Conventional Gaussian filters often underperform in non-Gaussian and non-stationary environments, necessitating alternative approaches. By integrating reinforcement learning with the maximum correntropy criterion and the Unscented Kalman Filter, the proposed approach enhances robustness. The q-Rényi kernel effectively models heavy-tailed noise, while TRPO adaptively tunes process and measurement noise covariances during online operation. Additionally, an MLE-based variant (MLE-MCUKF) is developed to address the unknown noise covariance problem through classical adaptation. The algorithm is evaluated on synthetic power system harmonic signals under complex noise conditions and is shown to outperform conventional Gaussian filters. Real-world validation is conducted using data from a grid-connected inverter setup, further demonstrating the superior estimation accuracy and adaptability of the proposed TRPO-MCUKF approach.</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

Adaptive maximum correntropy UKF with Trust Region Policy Optimization for power system harmonic estimation under non-Gaussian noise with unknown noise covariances

  • 期刊论文

  • 发表日期：

    2026年1月13日

查看全文
  • Pravir Yadav,
  • Jayanta Piri,
  • Aparajita Sengupta,
  • Mainak Sengupta

摘要

This paper proposes a new robust nonlinear state estimator, termed Trust Region Policy Optimization-based q-Rényi Maximum Correntropy Unscented Kalman Filter (TRPO-MCUKF), for joint estimation of power system harmonic amplitudes, phases, frequency, and noise parameters. Conventional Gaussian filters often underperform in non-Gaussian and non-stationary environments, necessitating alternative approaches. By integrating reinforcement learning with the maximum correntropy criterion and the Unscented Kalman Filter, the proposed approach enhances robustness. The q-Rényi kernel effectively models heavy-tailed noise, while TRPO adaptively tunes process and measurement noise covariances during online operation. Additionally, an MLE-based variant (MLE-MCUKF) is developed to address the unknown noise covariance problem through classical adaptation. The algorithm is evaluated on synthetic power system harmonic signals under complex noise conditions and is shown to outperform conventional Gaussian filters. Real-world validation is conducted using data from a grid-connected inverter setup, further demonstrating the superior estimation accuracy and adaptability of the proposed TRPO-MCUKF approach.