【目的】针对柔性直流配电网故障现有方法准确度低且故障信号获取成本高的问题，本文提出一种基于Transformer算法和生成对抗网络(GAN)融合模型的故障精确检测方法。【方法】首先，将预处理后的直流配电网故障电压和电流信号经过自适应变分模态分解提取有效模态分量，并利用Transformer算法提取有效特征向量。然后，采用Transformer算法替代GAN模型生成器，据此建立TransGAN深度学习模型。最后，基于Matlab/Simulink建立仿真模型对所提方法的有效性和准确性进行验证。【结果】试验结果表明，该方法在提高故障检测准确度、降低误报率方面具有显著优势，相比于现有检测方法，具有更高故障的识别精度和更强的泛化能力。【结论】所提方法在实现更高检测准确度的同时，所需的数据集大小相比于其他方法更低，降低了数据获取成本和计算复杂度，提升了模型的实际应用价值和工程可行性。

[Objective] Aiming at the problem that the existing methods for detecting faults in flexible DC distribution networks have low accuracy and high cost of acquiring fault signals, this paper proposes a fault accurate detection method based on the Transformer algorithm and the generative adversarial network (GAN) fusion model. [Methods] Firstly, the effective modal components of the preprocessed DC distribution network fault voltage and current signals were extracted by adaptive variational mode decomposition method, and the effective feature vectors were extracted by Transformer algorithm. Then, the GAN model generator was replaced by Transformer algorithm, and the TransGAN deep learning model was established based on it. Finally, a simulation model was established based on Matlab/Simulink to verify the effectiveness and accuracy of the proposed method. [Results] Experimental results showed that this method has significant advantages in improving fault detection accuracy and reducing false alarm rate. Compared with the existing detection methods, it has higher fault recognition accuracy and stronger generalization ability. [Conclusion] The proposed method achieving higher detection accuracy, the required data set size is smaller than other methods, which reduces the data acquisition cost and computational complexity, and improves the practical application value and engineering feasibility of the model.

南方电网创新项目(070500KK5222005)