基于交叉验证深度置信网络的少样本柔直计量装置故障诊断方法研究

为识别柔性直流输电系统计量装置的故障，提出了一种基于深度置信网络的故障诊断方法。该方法首先从合并单元端获取、解析数据并分成训练样本和测试样本；然后将这些数据用于训练深度置信网络。 最后将模型的故障诊断结果和实际样本的标签组合为一个交叉验证集合，从而测试深度置信网络性能。仿真结果表明，相比于支持向量机和BP神经网络，该文提出的基于深度置信网络的方法可以更加稳定、可靠地识别故障样本少的柔性直流计量装置的故障。

Research on Fault Diagnosis of Small-Sample Measuring Devices of Flexible DC Transmission System Based on Cross-Validation Deep Confidence Network

In order to identify faults of the measuring devices in the flexible DC power system, a method based on Deep Belief Network is proposed in this paper. First, the data are acquired and parsed from the merging unit and divided into training samples and test samples. Second, these data are used to train the deep confidence network. Finally, the fault diagnosis results of the model and the labels of the actual samples are combined into a cross-validation set to test the performance of the deep confidence network. The simulation results suggest that compared with the support vector machine and BP neural network, the proposed method based on deep confidence network can identify the faults of flexible DC metering devices having fewer fault samples more stably and reliably.