基于ST-TCN的太阳能光伏组件故障诊断方法

Fault diagnosis method of solar panel module based on ST-TCN

实地调研并收集电站光伏组件常见的故障类型,并对光伏组件在不同工作状况下的电流特征曲线进行分析,发现光伏组件的电流数据叠加了复杂的表现特征和高噪声。为能精准诊断光伏组件的故障类型,提出一种软阈值化的时序卷积神经网络(Soft Thresholding Temporal Convolutional Network,ST-TCN)光伏组件故障诊断模型。ST-TCN网络使用多个残差模块的膨胀卷积层、ReLU层、Dropout层提取电流数值特征和时序特征,再使用残差模块的软阈值化对所提取的特征降噪,最终使用全连接层对残差模块提取的特征进行故障诊断分类。实验结果表明,ST-TCN网络不仅结构简单,收敛速度快,而且故障诊断准确率高,达到92.99%。

This paper analyzes the current characteristic curves of photovoltaic modules under different working conditions and finds that the current data of photovoltaic modules superpose complex performance characteristics and high noise.In order to accurately diagnose the fault types of photovoltaic modules,a soft thresholding temporal convolutional network(ST-TCN)photovoltaic module fault diagnosis model is proposed.The ST-TCN network uses the dilated convolution layer,ReLU layer,and Dropout layer of multiple residual modules to extract current numerical and time series features,uses the soft thresholding of residual modules to de-noise the extracted features,and finally uses the full connection layer to diagnose and classify the extracted features of residual modules.The experimental results show that the ST-TCN network has a simple structure,fast convergence,and high accuracy in fault diagnosis,reaching 92.99%.