A Physics Parameter-Guided Metalearning Framework for Temperature Prediction of Double-Sided Double-Glass TopCon Photovoltaic Modules

IF 3.7 3区工程技术 Q3 ENERGY & FUELS

IEEE Journal of Photovoltaics Pub Date : 2026-07-01 Epub Date: 2026-04-22 DOI:10.1109/JPHOTOV.2026.3680213

Huajian Ding;Zhongyuan Su;Jun Wang

引用次数: 0

Abstract

Aiming at the need for accurate prediction operational temperature of N-type TOPCon bifacial double-glass photovoltaic modules, this article proposes an integrated forecasting method combining physical principles with metalearning. A metalearner model based on a Stacking ensemble framework is constructed, which integrates predictions from a thermodynamic model and a dual-branch long short-term memory (LSTM) network through a dynamic weighting mechanism. Cross-validation strategies are employed to train deep neural networks, enabling multiscale feature integration and weight self-adjustment. Experimental results achieve a root mean square error (RMSE) of only 0.95°C, representing a 66% reduction compared to the physical model (RMSE = 2.81°C) and a 14% reduction compared to the LSTM model (RMSE = 1.11°C), substantially improving the model’s adaptability in nonlinear and dynamic scenarios.

查看原文本刊更多论文

基于物理参数导向的双面双玻璃TopCon光伏组件温度预测元学习框架

针对n型TOPCon双面双玻璃光伏组件工作温度准确预测的需求，本文提出了一种物理原理与元学习相结合的综合预测方法。构建了一个基于堆叠集成框架的元学习器模型，该模型通过动态加权机制将热力学模型和双分支长短期记忆（LSTM）网络的预测相结合。采用交叉验证策略训练深度神经网络，实现多尺度特征集成和权值自调整。实验结果的均方根误差（RMSE）仅为0.95°C，与物理模型（RMSE = 2.81°C）相比降低了66%，与LSTM模型（RMSE = 1.11°C）相比降低了14%，大大提高了模型对非线性和动态场景的适应性。

本文章由计算机程序翻译，如有差异，请以英文原文为准。

求助全文

约1分钟内获得全文求助全文

来源期刊

IEEE Journal of Photovoltaics ENERGY & FUELS-MATERIALS SCIENCE, MULTIDISCIPLINARY

CiteScore

7.00

自引率

10.00%

发文量

206

期刊介绍： The IEEE Journal of Photovoltaics is a peer-reviewed, archival publication reporting original and significant research results that advance the field of photovoltaics (PV). The PV field is diverse in its science base ranging from semiconductor and PV device physics to optics and the materials sciences. The journal publishes articles that connect this science base to PV science and technology. The intent is to publish original research results that are of primary interest to the photovoltaic specialist. The scope of the IEEE J. Photovoltaics incorporates: fundamentals and new concepts of PV conversion, including those based on nanostructured materials, low-dimensional physics, multiple charge generation, up/down converters, thermophotovoltaics, hot-carrier effects, plasmonics, metamorphic materials, luminescent concentrators, and rectennas; Si-based PV, including new cell designs, crystalline and non-crystalline Si, passivation, characterization and Si crystal growth; polycrystalline, amorphous and crystalline thin-film solar cell materials, including PV structures and solar cells based on II-VI, chalcopyrite, Si and other thin film absorbers; III-V PV materials, heterostructures, multijunction devices and concentrator PV; optics for light trapping, reflection control and concentration; organic PV including polymer, hybrid and dye sensitized solar cells; space PV including cell materials and PV devices, defects and reliability, environmental effects and protective materials; PV modeling and characterization methods; and other aspects of PV, including modules, power conditioning, inverters, balance-of-systems components, monitoring, analyses and simulations, and supporting PV module standards and measurements. Tutorial and review papers on these subjects are also published and occasionally special issues are published to treat particular areas in more depth and breadth.