利用改进ResNet和暂稳态时间序列的光伏阵列在线故障诊断方法

光伏阵列在线故障诊断方法主要采用实时电压电流时序信号作为输入故障特征。然而，这些时序信号因受最大功率点跟踪和时变环境因素的影响，往往包含暂态和稳态交替过程以及时变噪声，显著制约了故障诊断精度及可靠性。针对这些问题，本文首先利用相对位置矩阵方法将三种一维暂稳态时序数据，包括加权总电流以及光伏阵列时序电压和电流，转换为二维数据，以此生成红、绿、蓝三通道图像。而后，将图像输入到所提的基于与坐标注意力结合的残差网络（Residual Network, Resnet）模型中，该模型能提取其丰富的故障信息，有效地提升故障诊断精度。最后，通过仿真和实际的故障模拟实验获取故障样本数据，以训练和测试所提的网络模型，并与多种其它网络模型进行对比，还对仿真数据集进行了可靠性验证。经实验分析证明，本文提出的故障检测与诊断方法在准确性和稳定性方面都有更佳的表现，根据仿真平台获得的数据集也有较高的可靠性。

On-line fault diagnosis method of photovoltaic array using improved ResNet and transient steady-state time series

The on-line fault diagnosis method of photovoltaic array mainly uses real-time voltage and current time series signals as input fault characteristics. However, due to the influence of maximum power point tracking (MPPT) and time-varying environmental factors, these time-series signals often contain transient and steady-state alternating processes, which significantly restrict the accuracy and reliability of fault diagnosis. Aiming at these problems, Firstly, three one-dimensional transient steady-state timing data, including weighted total current and photovoltaic array timing voltage and current, are converted into two-dimensional data by using the relative position matrix method to generate red, green, and blue (RGB) three-channel images. Then, the images are input into the proposed residual network (Resnet) model based on the combination of coordinate attention, which can extract its rich fault information and effectively improve the fault diagnosis accuracy. Finally, fault sample data are obtained through simulated and actual fault simulation experiments to train and test the proposed network model, compare it with a variety of other network models, and verify the reliability of the simulation data set. The experimental analysis proves that the fault detection and diagnosis method proposed in this paper has the better performance in terms of accuracy and stability, and the data set obtained according to the simulation platform also has high reliability.