轻量化光伏电弧故障检测模型及边缘部署方法
Lightweight PV Arc Fault Detection Model and Edge Deployment Method
摘要
随着光伏发电系统的广泛应用,光伏串联电弧故障以其隐蔽性与危害性威胁着光伏系统的稳定运行.光伏系统电弧故障信号较弱且不同天气情况差异较大,存在难以快速准确识别的问题.通过搭建光伏系统电弧故障实验平台,完成故障数据采集与故障特征分析;基于故障特征分析结果选取ShuffleNetV2卷积神经网络并进行改进,融合双注意力机制等模块强化模型信息捕捉能力;利用构建的电弧故障数据集开展模型训练与测试,结果表明改进模型对光伏电弧故障的检测准确率达99.5%.在此基础上设计了8通道的光伏系统电弧故障检测边缘终端,将模型部署到终端设备,该设备不仅实现对光伏电弧故障的可靠检测,在云层遮挡、逆变器干扰等复杂工况下也保持良好的鲁棒性,为光伏系统安全运维提供了可靠的技术方案.
With the widespread application of photovoltaic(PV)systems,series arc faults have become a potential hazard to system safety due to their concealment and destructive characteristics.The weak amplitude and strong weather-dependent variability of arc fault signals bring challenges to rapid and accurate detection.Therefore,an experimental platform for PV series arc faults is established for data collection and fault characteristic analysis.According to the characteristic analysis,an improved ShuffleNetV2 network integrated with a dual-attention mechanism module is proposed to enhance model information capture ability.The model is trained and tested on the established arc fault dataset and the result indicate the detection accuracy of the improved model reaches 99.5%.An 8-channel edge detection terminal is designed and deployed with the proposed model.The proposed device enables reliable detection of PV arc faults,while maintaining strong robustness under complex working conditions including cloud shading and inverter interference,which provides a feasible technical scheme for the safe operation and maintenance of PV systems.
年,卷(期):
2026, 39(10)
分类号:
TM501.2
TM615
TP301.6
机标分类号:
TP393.08
TN925.93
R575.4
在线出版日期:
2026年10月14日
基金项目:
52789925000W:国家电网公司综合能源服务集团有限公司科技项目2023H0007:福建省科技计划工业引导性(重点)项目,