摘要
针对水轮发电机组的故障诊断及预测问题,提出一种基于卷积神经网络(CNN)与注意力机制融合的智能诊断模型(CNN-Attention)。通过模拟低泥沙磨损、中泥沙磨损、高泥沙磨损和撞击故障4种故障类型,结合振动信号特征分析,验证模型的有效性。实验结果表明,CNN-Attention模型的综合诊断准确率达92.5%,较传统决策树和随机森林模型平均提升5%~15%。该模型在小样本条件下仍能保持80%的准确率,展现出较强的特征提取能力和噪声鲁棒性。
This study proposes an intelligent diagnostic model(CNN-Attention)based on the fusion of convolutional neural network(CNN)and attention mechanism for the fault diagnosis and prediction of hydroelectric generating units.By simulating four types of faults including low,medium,and high sediment abrasion and impact faults,and combining with the analysis of vibration signal features,the effectiveness of the model is verified.The experimental results show that the comprehensive diagnostic accuracy of the CNN-Attention model reaches 92.5%,which is on average 5%to 15%higher than that of traditional decision tree and random forest models.The model can still maintain an accuracy rate of 80%under small sample conditions,demonstrating strong feature extraction ability and noise robustness.
作者
蒋晨
曹琦
Jiang Chen;Cao Qi(Urban Flood Control Project Management Office of Changzhou City,Jiangsu Province,Changzhou 213000,China)
出处
《吉林水利》
2025年第10期45-49,共5页
Jilin Water Resources
关键词
水轮发电机组
故障诊断
机器学习
Hydroelectric generating units
Fault diagnosis
Machine learning
