4.8 Article

Satellite based fault diagnosis of photovoltaic systems using recurrent neural networks

期刊

APPLIED ENERGY
卷 305, 期 -, 页码 -

出版社

ELSEVIER SCI LTD
DOI: 10.1016/j.apenergy.2021.117874

关键词

Photovoltaics; Recurrent neural network; Time series classification; Fault diagnosis; Fault detection

资金

  1. DAPPER project - Flux50 and Flanders Innovation & Entrepreneurship, Belgium [HBC.2020.2144]

向作者/读者索取更多资源

Current techniques for PV fault diagnosis, while accurate, are costly and limit widespread adoption. This study proposes a machine learning-based approach that utilizes satellite weather data and low-frequency inverter measurements for precise fault diagnosis in PV systems.
Due to manufacturing defects and wear, faults in photovoltaic (PV) systems are often unavoidable. The effects range from energy losses to risk of fire and electrical shock, making early fault detection and identification crucial. Literature focuses on PV fault diagnosis using dedicated on-site sensors or high-frequency current and voltage measurements. Although these existing techniques are accurate, they are not economical for widespread adoption, leaving many PV systems unmonitored. In contrast, we introduce a machine learning based technique that relies on satellite weather data and low-frequency inverter measurements for accurate fault diagnosis of PV systems. This allows one to adopt machine learning based fault diagnosis even for PV systems without on-site sensors. The proposed approach relies on a recurrent neural network to identify six relevant types of faults, based on the past 24 h of measurements, as opposed to only taking into account the most recent measurement. Additionally, whereas state-of-the-art methods are limited to identifying the fault type, our model also estimates the output power reduction stemming from the fault, i.e., the fault severity. Comprehensive experiments on a simulated PV system demonstrate that this approach is sensitive to faults with a severity as small as 5%, reaching an accuracy of 96.9% +/- 1.3% using exact weather data and 86.4% +/- 2.1% using satellite weather data. Finally, we show that the model generalizes well to climates other than the climate of its training data and that the model is also able to detect unknown faults, i.e., faults that were not represented in the training data.

作者

我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。

评论

主要评分

4.8
评分不足

次要评分

新颖性
-
重要性
-
科学严谨性
-
评价这篇论文

推荐

暂无数据
暂无数据