摘要
在管道输送过程中,燃气轮机的健康状态与管道的安全平稳运行密切相关。为提高燃驱压缩机组燃气轮机气路故障诊断的准确率,以双轴燃气轮机为对象,通过建模获得不同气路故障类型的数据集,引入相关向量机(RVM)模型对典型气路故障进行分类诊断,并采用改进鸡群算法(ECSO)对RVM模型的超参数进行寻优。结果表明,ECSO-RVM模型的分类精度为95.55%,明显优于支持向量机(SVM)、径向基核函数(RBF)神经网络、反向传播(BP)神经网络、改进蜂群算法-自适应模糊推理系统(IABC-ANFIS)和深度置信网络(DBN)等模型;在添加不同程度高斯白噪声的条件下,ECSO-RVM模型的分类精度始终保持在80%以上,说明该模型的分类精度、抗噪声能力和泛化能力较好,可用于燃气轮机气路故障的快速识别和诊断。
The relevance vector machine(RVM) model has been employed to classify and diagnose typical gas path faults in a biaxial gas turbine in order to investigate the degradation faults in the gas path components of the combustion drive compressor unite. An enhanced chicken swarm optimization(ECSO) was used to optimize the hyperparameters of the RVM model. The classification accuracy of the ECSO-RVM model was 95.55%, which is significantly better than the results obtained with SVM, RBF, BP neural network, IABC-ANFIS or DBN models. The classification accuracy of the ECSO-RVM model remains above 80% when different degrees of Gaussian white noise are added, indicating that the model has good classification accuracy, anti-noise resistance and generality. The model can be used for rapid identification and diagnosis of gas turbine gas path faults.
作者
张平
罗易洲
王子聪
李亚民
ZHANG Ping;LUO YiZhou;WANG ZiCong;LI YaMin(State Pipe Network Group United Pipe Co.,Ltd.,Western Branch,Urumqi 830011;Liaoning World Walker Technology Co.,Ltd.,Shenyang 110121,China)
出处
《北京化工大学学报（自然科学版）》
CAS
CSCD
北大核心
2022年第6期73-82,共10页
Journal of Beijing University of Chemical Technology(Natural Science Edition)
