末级过热器管壁频繁超温且机理复杂,快速准确地追溯机组超温原因,有助于从源头解决超温问题,对电厂的安全运行有重要意义。为此,本文提出基于改进对抗生成网络与深度迁移学习的末级过热器超温诊断方法。首先,通过格拉姆角差场GADF编码将SIS系统中的关键特征变量转换为二维图像,以适配迁移学习模型；其次,利用改进对抗生成网络CWGAN生成高质量图像样本,以显著提高Xception网络在稀疏不平衡源域中的泛化能力；最后,微调在源域机组上预训练后的Xception网络,并将其迁移到稀疏不平衡的目标域机组中。以某在役的两台600MW超临界机组末级过热器为对象进行验证,结果表明,本文所提方法的平均准确率达到了99.03%,相较于传统深度学习方法,该方法的收敛速度更快、准确率更高,实现了对目标域机组末级过热器超温的快速诊断。

The frequent overtemperature of the final-stage superheater tube wall is a complex issue, and accurately tracing the causes of unit overtemperature is crucial for addressing overtemperature problems at their source, which is of paramount importance for the safe operation of power plants. In this paper, a superheater overtemperature diagnosis method is proposed based on improved generative adversarial networks and deep transfer learning. First, the key feature variables collected by the SIS system are transformed into two-dimensional images using GADF encoding to adapt to the transfer learning model. Second, an improved conditional Wasserstein GAN (CWGAN) is utilized to generate high-quality image samples, significantly enhancing the generalization ability of the Xception network in the sparse and imbalanced source domain. Finally, the pre-trained Xception network on the source domain unit is fine-tuned and transferred to the sparse and imbalanced target domain unit. The proposed method is validated using two in-service 600MW ultra-supercritical units. The results indicate that the average accuracy of the proposed method reaches 99.03%. Compared with traditional deep learning methods, this method exhibits faster convergence and higher accuracy, achieving rapid diagnosis of overtemperature in the final-stage superheater of the target domain unit.

中国华能集团有限公司2022年度科技项目(HNKJ22-HF22)