在压缩机组运行过程中,针对多个监测信号量发生微小故障作用下的隐含特征对机组运行的影响,提出了基于图卷积神经网络(Graph Convolutional Neural Network,GCNN)的压缩机组风险预警模型。将时序的压缩机组运行数据转化为基于时间点的信号网络,依据改进的密度峰值聚类(An Improved Density Peak Fast Search Algorithm,AIDP)算法对机组运行数据进行故障粗分类,获得标签样本;再通过GCNN训练标签数据的隐含特征,实现机组风险预警;在真实数据集中进行预警分析。实验结果表明,所提出的模型极大增强了压缩机组风险预警的识别能力,为保证整个压缩机组的安全平稳运行提供了理论与实践依据。

During the operation of the compressor unit, we have developed a risk warning model for compressor unit based on graph convolutional neural networks. This model takes into consideration the impact of the implied features of multiple small fault monitoring signal quantities on the unit operation. First, we convert the time-series unit operation data into a time-point-based signal network. Secondly, we use an improved density peak clustering algorithm to perform a coarse classification of the fault on the unit operation data, resulting in labeled samples. Finally, risk warning for compressor unit is achieved by deep learning on the implied features of labeled data using graph convolutional neural networks. Experimental results on real data sets show that our proposed risk warning model greatly enhances the identification capability of risk warning for compressor unit, and the proposed model provides a theoretical and practical basis for ensuring the safe and smooth operation of the whole compressor unit.