基于改进SqueezeNet的光伏阵列故障诊断

Photovoltaic Array Fault Diagnosis Based on Improved SqueezeNet

光伏阵列常受极端条件影响,易产生各种故障,影响电力系统安全运行。为了提高光伏阵列故障诊断模型的综合性能,提出了一种基于改进SqueezeNet的轻量化光伏阵列故障诊断模型,用于对短路、断路、阴影、老化故障的精确诊断。采用深度卷积代替传统卷积降低模型的参数计算量,使用通道洗牌操作提高模型的特征交互能力,并加入残差连接允许信息在层与层之间流通。基于光伏阵列的I-U和P-U曲线,构建了I,U,P三维通道特征图作为模型的输入。实验结果表明,所提模型能够准确地进行故障分类,平均准确率为99.65%,优于其他对比模型,且相较于其他模型,所提模型在保持高准确率的同时,大幅降低了计算复杂度,可实现光伏阵列故障的实时在线诊断。

Photovoltaic arrays are often affected by extreme conditions and are prone to various failures,affecting the safe operation of the power system.In order to improve the comprehensive performance of the photovoltaic array fault diagnosis model,a lightweight photovoltaic array fault diagnosis model based on improved SqueezeNet is proposed for accurate diagnosis of short circuit,open circuit,shadow and aging faults.Deep convolution is used instead of traditional convolution to reduce the parameter calculation amount of the model,channel shuffling operation is used to improve the feature interaction ability of the model and residual connections are added to allow information to flow between layers.Based on the I-U and P-U curves of the photovoltaic array,a three-dimensional channel feature map of I,U and P is constructed as the input of the model.Experimental results show that the proposed model can accurately classify faults with an average accuracy of 99.65%which is better than other comparison models.Compared with other models,the proposed model greatly reduces the computational complexity while maintaining high accuracy and can realize real-time online diagnosis of photovoltaic array faults.