Scholarlink

<p>Accurate and real-time estimation of photovoltaic (PV) parameters is critical for maximum power point tracking, fault diagnosis, and long-term system performance monitoring under dynamic environmental conditions. Existing metaheuristic-based approaches, such as PSO, DE and other hybrid optimization methods, suffer from limitations including static initialization, weak causality enforcement, lack of real-time adaptability, and high computational cost, which hinder their performance under rapidly varying irradiance and temperature. To address these challenges, this study proposes a Multi-Stage Metaheuristic–Sequential Filtering Framework that integrates a Causality-Aware Snow Leopard Optimization (CASLO) metaheuristic with an Unscented Kalman Filter (UKF) for online PV parameter tracking, supplemented by an adaptive feedback loop that selectively re-optimizes parameters when deviations exceed predefined thresholds. CASLO enforces parameter causality and physical consistency, while UKF ensures smooth, noise-resilient tracking of all five PV parameters (Iph, I0,n, Rs, Rsh​) in real time. The proposed method enforces physical feasibility and causal parameter dependencies during global optimization, followed by noise-resilient sequential tracking of PV parameters in real time. An event-driven feedback loop selectively re-optimizes only affected parameters when estimation deviations exceed predefined thresholds, ensuring computational efficiency and stability. The framework is implemented in Python. Experimental results demonstrate a significant enhancement in estimation performance, achieving approximately 34% reduction in RMSE and nearly 35% reduction in MAE compared to state-of-the-art GA, PSO, DE, and GWO-based methods, while consistently improving the goodness-of-fit (R² ≈ 0.96) under dynamic irradiance and temperature conditions. The proposed framework operates without reliance on large historical datasets or deep learning models, making it suitable for real-time deployment. Experimental results show that our method achieves 34% lower RMSE and 35% lower MAE results compared to traditional methods which produce better goodness-of-fit results that reach R² values of 0.96. The proposed framework successfully delivers precise and dependable photovoltaic parameter assessments which maintain operational accuracy throughout changing environmental conditions.</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

Causality-constrained snow leopard optimization with adaptive filtering for intelligent real-time parameter identification

  • 期刊论文

  • 发表日期：

    2026年4月29日

查看全文
  • Natrayan Lakshmaiya

摘要

Accurate and real-time estimation of photovoltaic (PV) parameters is critical for maximum power point tracking, fault diagnosis, and long-term system performance monitoring under dynamic environmental conditions. Existing metaheuristic-based approaches, such as PSO, DE and other hybrid optimization methods, suffer from limitations including static initialization, weak causality enforcement, lack of real-time adaptability, and high computational cost, which hinder their performance under rapidly varying irradiance and temperature. To address these challenges, this study proposes a Multi-Stage Metaheuristic–Sequential Filtering Framework that integrates a Causality-Aware Snow Leopard Optimization (CASLO) metaheuristic with an Unscented Kalman Filter (UKF) for online PV parameter tracking, supplemented by an adaptive feedback loop that selectively re-optimizes parameters when deviations exceed predefined thresholds. CASLO enforces parameter causality and physical consistency, while UKF ensures smooth, noise-resilient tracking of all five PV parameters (Iph, I0,n, Rs, Rsh​) in real time. The proposed method enforces physical feasibility and causal parameter dependencies during global optimization, followed by noise-resilient sequential tracking of PV parameters in real time. An event-driven feedback loop selectively re-optimizes only affected parameters when estimation deviations exceed predefined thresholds, ensuring computational efficiency and stability. The framework is implemented in Python. Experimental results demonstrate a significant enhancement in estimation performance, achieving approximately 34% reduction in RMSE and nearly 35% reduction in MAE compared to state-of-the-art GA, PSO, DE, and GWO-based methods, while consistently improving the goodness-of-fit (R² ≈ 0.96) under dynamic irradiance and temperature conditions. The proposed framework operates without reliance on large historical datasets or deep learning models, making it suitable for real-time deployment. Experimental results show that our method achieves 34% lower RMSE and 35% lower MAE results compared to traditional methods which produce better goodness-of-fit results that reach R² values of 0.96. The proposed framework successfully delivers precise and dependable photovoltaic parameter assessments which maintain operational accuracy throughout changing environmental conditions.