摘要
针对基于支持向量机(Support Vector Machine,SVM)的间歇过程故障诊断准确率低的问题,结合间歇过程的时段特性,提出了一种基于子时段MPCA-SVM的间歇过程在线故障诊断方法。首先,利用多向主成分分析(Multi-way principal component analysis,MPCA)提取出间歇过程正常运行状态下的每个采样点的主成分,将相邻的且具有相同主成分个数的采样点归到同一粗划分时段内,再在每一个粗时段内利用相邻采样点的负载矩阵的角度信息作为相似性判据来细化分时段;其次,对每个时段建立MPCA在线过程监测模型,同时,利用MPCA提取每个时段内各个类型故障的特征,并用特征数据建立SVM故障诊断模型;最后,MPCA监测模型实施监测功能,当检测到故障时,相应时段的SVM故障诊断模型进行诊断。将该方法应用于青霉素发酵过程仿真平台进行验证,该方法相比于不分时段的SVM的故障诊断方法,平均可提高故障诊断准确率11%,实验结果表明了该方法的有效性和可行性。
Since the accuracy of fault diagnosis of batch process based on Support Vector Machine （SVM） is low generally, an on-line fault diagnosis for batch process based on sub-phase MPCA-SVM is proposed combing with the multiplicity of phases in batch process. Firstly, Multi-way Principal Component Analysis （MPCA） was utilized to extract the principal components （PCs） of each continuous time slices of normal batch process, which had the same number of PCs would be put together, so the coarser partition was achieved, then use the load matrix angle information as the similarity standard to complete the subdivision partition. Secondly, the detection models of each phase which were used to detect the faults were built, and PCs from all kinds of fault data were extracted by MPCA in each phase, the models of fault diagnosis were built by using the PCs as the input. Finally, the detection models devoted to monitor the batch process, when the model detects faults, the corresponding SVM fault diagnosis model was used to conduct to diagnose. The method was applied on the fed-batch penicillin fermentation simulation platform, the accuracy of fault diagnosis can increase 11% averagely, which shows this method has high efficient and can be used in production process extensively.
出处
《计算机与应用化学》
CAS
2016年第4期465-471,共7页
Computers and Applied Chemistry
基金
国家自然科学基金资助项目(61174109
61364009)
关键词
间歇过程
多向主成分分析
支持向量机
过程监测
故障诊断
batch processes
multi-way principal component analysis
support vector machine
process monitor
fault diagnosis
