基于残差网络的光伏组件故障诊断

Fault Diagnosis of Photovoltaic Modules Based on Residual Network

光伏的故障诊断对于光伏的运维具有重要的意义。传统的故障诊断方法准确率较低,识别速度较慢,受外界因素的影响较大。文章通过使用残差网络对电致发光的公开数据集进行故障诊断。基于Pytorch框架搭建,先对数据集进行预处理,避免外界因素对试验产生不必要的影响,通过比较不同深度网络对性能的影响,从而选择性能最好的ResNet34;为更好地提取EL图像的特征,在ResNet34的基础上引入注意力机制;因数据集数量较少无法对网络进行训练,所以使用ImageNet上预训练好的ResNet34模型来进行分类,并分析了不同的Dropout概率和不同梯度下降算法对测试集性能的影响。试验结果表明,最终准确率达到84.4%,能够有效地对光伏故障进行诊断。

The fault diagnosis of photovoltaic systems is of great significance for the operation and maintenance of photovoltaic systems.Traditional fault diagnosis methods have low accuracy,slow recognition speed,and are greatly influenced by external factors.The article conducts fault diagnosis on the publicly available dataset of electroluminescence using residual networks.Based on the Python framework,the dataset is preprocessed to avoid unnecessary external factors affecting the experiment.By comparing the performance of different deep networks,the ResNet34 with the best performance is selected;To better extract the features of EL images,attention mechanism is introduced on the basis of ResNet34;Due to the limited number of datasets and the inability to train the network,the pre trained ResNet34 model on ImageNet was used for classification,and the impact of different Dropout probabilities and gradient descent algorithms on the performance of the test set was analyzed.The experimental results show that the final accuracy reaches 84.4%,which can effectively diagnose photovoltaic faults.