基于时间序列模型的化工设备状态的预测研究
|
摘要
故障预报和预警是20世纪后期才发展起来的一个新兴的研究方向,也是当前预测控制理论研究的热点之一。随着对系统可靠性和安全性要求的进一步提高,人们希望能够在故障对系统的危害显现之前就能够更加准确的预测故障发生的有关信息,由此提出了对故障预报更加严格的要求。目前故障预报研究进展较为缓慢,基于时间序列预测的方法是现有的主要预报方法。本文分析了在时间序列预测领域中应用比较广泛的几种基础的方法,比较了它们在预测中的应用范围和各自的优劣,并在这基础的方法上提出了改进。结合当今预测领域新兴的数据挖掘思想,介绍了国外一种新的预测方法,并将上述预测方法和化工监测诊断系统相结合。本文着眼于预测模型的实际应用,主要工作如下:
1、总结并分析了时间序列预测领域中应用最广泛的自回归—滑动平均模型(ARMA模型),并针对其在定阶中的存在的模糊性和人为判定的情况,本文在实际应用中采用了F检验逐步自动定阶的方法,避免了ARMA模型定阶中所遇到的问题,实现了数据处理、模型定阶和状态预测的实际应用。
2、针对基础的Elman神经网络,本文基于已有的神经网络思想,改进了其网络的内部结构,即在基础的Elman神经网络的承接层增加一个自反馈,增强了Elman神经网络的动态跟踪性能。
3、本文把时间序列序贯学习的思想引入Elman网络的在线建模过程中,实现了网络在线学习,通过时间窗的推移,不断地利用新采集数据对网络进行训练,提高了网络的建模效率和对真实数据的跟踪和拟合能力。
4、针对故障样本数据丰富的情况,借鉴美国辛辛那提大学的IMS(Intelligent Maintenance Systems)中心提出的匹配矩阵的思想,本文介绍了一种基于数据挖掘技术的匹配矩阵预测模型,并将其应用于设备状态的真实数据,取得了良好的预测效果。而且在该匹配矩阵模型的建模过程中融合了ARMA模型,拥有较高的建模效率和反应速度,体现了多种预测方法相互融合的现代预测技术的发展方向。
5、面对目前matlab软件不支持建模函数编译成动态链接库被vc调用的问题,通过重新编写算法和利用其他组织编写的时间序列工具箱等方法,应用vc和matlab混合编程的技术,将论文中提出的预测方法加以实现并应用到化工领域的维修与安全保障完整性管理平台中,并进行仿真实验。仿真结果表明了文章中提出的的预测方法,预测精度和效率能够达到实际应用的要求。
Fault forecasting and advance warning is a new subject started from the late 20th century, and is one of the hot research areas of predictive control theory. With further improvement of the system requirements for reliability and security, it is hoped that we can more accurate predict the fault information before the fault harm to the system appears.And because of that strict demands on fault forecasting and early warning are made. Now progress in the prediction field is slow,based on time series prediction method is the main existing forecasting methods. This paper analyzes several more extensive application of basic methods in the time series prediction, and compares of their scope of application and their respective advantages and disadvantages, and makes suggestions to improve. The paper introduces a new method of prediction based on data mining.
The main contributions of this paper are as follows:
1、Summarized and analyzed the Auto Regression-Moving Average model, and in the light of the concrete problems existing in the model order confirmation,this paper adapts auto-order method step by step and F significant to avoid the ARMA model order confirmation.
2、Aimd at the basic Elman neural network, this paper improved the internal structure of the network, enhanced dynamic tracking performance of Elman neural network.
3、Introduced a sequential learning algorithm of neural network and its application, and based on that realised network online study. Goes through the time window, continue to take advantage of new data collection for training.
4、Based on the condition of rich fault sample data,the paper use the idea of match matrix put forward by the IMS ( Intelligent Maintenance Systems) of University of Cincinnati for reference to describe a new prediction method depends on data mining.
5、In view of the current problem that matlab dosen't support mixed programming of model building function, through rewriting algorithm and making use of time series tool box that other organizations write etc, using mixed programming to take the forecasting method that puts forward in the thesis to carry out and apply to areas of production.
1、总结并分析了时间序列预测领域中应用最广泛的自回归—滑动平均模型(ARMA模型),并针对其在定阶中的存在的模糊性和人为判定的情况,本文在实际应用中采用了F检验逐步自动定阶的方法,避免了ARMA模型定阶中所遇到的问题,实现了数据处理、模型定阶和状态预测的实际应用。
2、针对基础的Elman神经网络,本文基于已有的神经网络思想,改进了其网络的内部结构,即在基础的Elman神经网络的承接层增加一个自反馈,增强了Elman神经网络的动态跟踪性能。
3、本文把时间序列序贯学习的思想引入Elman网络的在线建模过程中,实现了网络在线学习,通过时间窗的推移,不断地利用新采集数据对网络进行训练,提高了网络的建模效率和对真实数据的跟踪和拟合能力。
4、针对故障样本数据丰富的情况,借鉴美国辛辛那提大学的IMS(Intelligent Maintenance Systems)中心提出的匹配矩阵的思想,本文介绍了一种基于数据挖掘技术的匹配矩阵预测模型,并将其应用于设备状态的真实数据,取得了良好的预测效果。而且在该匹配矩阵模型的建模过程中融合了ARMA模型,拥有较高的建模效率和反应速度,体现了多种预测方法相互融合的现代预测技术的发展方向。
5、面对目前matlab软件不支持建模函数编译成动态链接库被vc调用的问题,通过重新编写算法和利用其他组织编写的时间序列工具箱等方法,应用vc和matlab混合编程的技术,将论文中提出的预测方法加以实现并应用到化工领域的维修与安全保障完整性管理平台中,并进行仿真实验。仿真结果表明了文章中提出的的预测方法,预测精度和效率能够达到实际应用的要求。
Fault forecasting and advance warning is a new subject started from the late 20th century, and is one of the hot research areas of predictive control theory. With further improvement of the system requirements for reliability and security, it is hoped that we can more accurate predict the fault information before the fault harm to the system appears.And because of that strict demands on fault forecasting and early warning are made. Now progress in the prediction field is slow,based on time series prediction method is the main existing forecasting methods. This paper analyzes several more extensive application of basic methods in the time series prediction, and compares of their scope of application and their respective advantages and disadvantages, and makes suggestions to improve. The paper introduces a new method of prediction based on data mining.
The main contributions of this paper are as follows:
1、Summarized and analyzed the Auto Regression-Moving Average model, and in the light of the concrete problems existing in the model order confirmation,this paper adapts auto-order method step by step and F significant to avoid the ARMA model order confirmation.
2、Aimd at the basic Elman neural network, this paper improved the internal structure of the network, enhanced dynamic tracking performance of Elman neural network.
3、Introduced a sequential learning algorithm of neural network and its application, and based on that realised network online study. Goes through the time window, continue to take advantage of new data collection for training.
4、Based on the condition of rich fault sample data,the paper use the idea of match matrix put forward by the IMS ( Intelligent Maintenance Systems) of University of Cincinnati for reference to describe a new prediction method depends on data mining.
5、In view of the current problem that matlab dosen't support mixed programming of model building function, through rewriting algorithm and making use of time series tool box that other organizations write etc, using mixed programming to take the forecasting method that puts forward in the thesis to carry out and apply to areas of production.
引文
[1]张正道,胡寿松.复杂非线性系统故障检测与故障预报[D].南京:南京航空航天大学自动化学院,2006.2
[2]周成容,杨大地.故障预测诊断维修系统开发框架[D].重庆:重庆大学计算数学专业,2006.4
[4]何书元.应用时间序列分析[M].北京:北京大学出版社,2003
[5]Gang Yu,Hai Qiu,Dragan Djurdjanovie,Jay Lee.Feature signature prediction of a boring process using neuralnetwork modeling with confidence bounds[J].Int J Adv Manuf Tectmol:2006,30:614-621
[6]采峰,曾凤章.基于非平稳时序的城市用水量ANN-ARMA预测模型[J].环境科学,2005,26(1):42-46
[7]Gordo J M,Guesdes S C.Approximate load shortening curves for stiffened-plates under uniaxial compression[A].In:Faulkner D,Cowling M J,Incecik A,et al.Integrity of Offshore Structures-5[C].Arly:EMAS,1993,189-211
[8]吉乔伟,毛根海.基于改进ARMA模型的时用水量预测[D].江南大学学报(自然科学版),2008,4
[9]唐玉娜,李启会.ARMA模型在预测问题中的应用[D].嘉兴学院数学与信息科学学院,2006,6
[10]杨素春,王朝晖,于晓红.基于ARMA模型的往复式压缩机状态预测[J].石油化工设备,2007
[11]李晓静,邓国和.基于遗传算法和改进Elman神经网络的股价预测[J].广西师范大学学报:自然科学版,2008
[12]陈敏.基于BP神经网络的混沌时间序列预测模型研究[D].长沙:中南大学,2007
[13]王波.基于神经网络的金融时间序列分析[D].天津:天津大学,2005
[14]神经网络模型及其MATLAB仿真程序设计[m],清华大学出版社,2005
[15]杨戈,吕剑虹.一种新型RBF网络序贯学习算法[J].中国科学E辑:工程科学材料科34(7):763-775
[16]任丽娜.基于Elman神经网络的中期电力负荷预测模型研究[D].兰州:兰州理工大学,2007
[17]苏刚,王玲玲,徐永生.基于改进Elman网络的燃气负荷预测[J].东南大学学报,2006.Vol.36
[18]韩旭明.Elman神经网络的应用研究[D].天津大学计算机科学与技术学院,2006
[19]徐留兴,梁川,秦远清.改进的Elman模型在紫坪铺月径流预测中的应用[J].四川大学学报(工程科学版),2006,5,Vol.38
[20]宋崇智,吴玉国,王璐.改进Elman网络在发动机齿轮箱故障诊断中的研究[J].仪器仪表学报,2008,Vol.29
[21]邓超,熊范伦,谭营等.一种序贯学习神经网络及其应用[J].电子科学学刊,1999,Vol.21
[22]Gang Yu,Hai Qiu.Performance Prediction Using Recurrent Neural Network Modeling with Confidence Bounds[J/OL].Center for Intelligent Maintenance Systems,www.imscenter.net,2002
[23]Lee,J.Measurement of Machine Performance Degradation using a Neural Network Model,Computers in Industry,Vol.30,pp.193-209,1996.
[24]边海燕.基于数据挖掘技术的动态检测与故障预测分析[D].北京:北京化工大学,2004,5
[25]裴双喜.基于数据挖掘的金融时间序列预测分析与研究[D].大连:大连海事大学,2008,3
[26]康丽锋.基于数据挖掘的短期负荷预测方法研究[D].北京:华北电力大学,2006,12
[27]Watchdog Agent Tools Documentation[R].Cincinnati:Center for Intelligent Maintenance System,2007
[28]熊腾科.基于数据挖掘的金融预测模型[D].厦门:厦门大学,2006
[29]Jianbo Lin,Dragan Djurdjanovic.Similarity based method for manufacturing process perfbrmance prediction and diagnosis[J].Computers in Industry 58(2007) 558-566
[30]梁平,龙新峰,吴庚申.基于ARMA及神经网络的汽轮机振动故障诊断研究[J].热能动力工程,2007,Vol.22
[31]屈亚玲,周建中,刘芳等.基于改进的Elman神经网络的中长期径流预报[J].JOURNAL OF CHINA HYDROLOGY,2005,Vol,26
[32]神经网络模型及其MATLAB仿真程序设计[M].清华大学出版社,2005
[33]易文周,刘海燕,张超英,王强.VC++与MATLAB混合编程实现遗传算法在图像分割上的应用[J].计算机应用与软件,2008,Vol,23
[34]黄金明,武玉强,邢西深.VC++与MATLAB的混合编程研究[J].电脑开发与应用,2008,Vol,21
[2]周成容,杨大地.故障预测诊断维修系统开发框架[D].重庆:重庆大学计算数学专业,2006.4
[4]何书元.应用时间序列分析[M].北京:北京大学出版社,2003
[5]Gang Yu,Hai Qiu,Dragan Djurdjanovie,Jay Lee.Feature signature prediction of a boring process using neuralnetwork modeling with confidence bounds[J].Int J Adv Manuf Tectmol:2006,30:614-621
[6]采峰,曾凤章.基于非平稳时序的城市用水量ANN-ARMA预测模型[J].环境科学,2005,26(1):42-46
[7]Gordo J M,Guesdes S C.Approximate load shortening curves for stiffened-plates under uniaxial compression[A].In:Faulkner D,Cowling M J,Incecik A,et al.Integrity of Offshore Structures-5[C].Arly:EMAS,1993,189-211
[8]吉乔伟,毛根海.基于改进ARMA模型的时用水量预测[D].江南大学学报(自然科学版),2008,4
[9]唐玉娜,李启会.ARMA模型在预测问题中的应用[D].嘉兴学院数学与信息科学学院,2006,6
[10]杨素春,王朝晖,于晓红.基于ARMA模型的往复式压缩机状态预测[J].石油化工设备,2007
[11]李晓静,邓国和.基于遗传算法和改进Elman神经网络的股价预测[J].广西师范大学学报:自然科学版,2008
[12]陈敏.基于BP神经网络的混沌时间序列预测模型研究[D].长沙:中南大学,2007
[13]王波.基于神经网络的金融时间序列分析[D].天津:天津大学,2005
[14]神经网络模型及其MATLAB仿真程序设计[m],清华大学出版社,2005
[15]杨戈,吕剑虹.一种新型RBF网络序贯学习算法[J].中国科学E辑:工程科学材料科34(7):763-775
[16]任丽娜.基于Elman神经网络的中期电力负荷预测模型研究[D].兰州:兰州理工大学,2007
[17]苏刚,王玲玲,徐永生.基于改进Elman网络的燃气负荷预测[J].东南大学学报,2006.Vol.36
[18]韩旭明.Elman神经网络的应用研究[D].天津大学计算机科学与技术学院,2006
[19]徐留兴,梁川,秦远清.改进的Elman模型在紫坪铺月径流预测中的应用[J].四川大学学报(工程科学版),2006,5,Vol.38
[20]宋崇智,吴玉国,王璐.改进Elman网络在发动机齿轮箱故障诊断中的研究[J].仪器仪表学报,2008,Vol.29
[21]邓超,熊范伦,谭营等.一种序贯学习神经网络及其应用[J].电子科学学刊,1999,Vol.21
[22]Gang Yu,Hai Qiu.Performance Prediction Using Recurrent Neural Network Modeling with Confidence Bounds[J/OL].Center for Intelligent Maintenance Systems,www.imscenter.net,2002
[23]Lee,J.Measurement of Machine Performance Degradation using a Neural Network Model,Computers in Industry,Vol.30,pp.193-209,1996.
[24]边海燕.基于数据挖掘技术的动态检测与故障预测分析[D].北京:北京化工大学,2004,5
[25]裴双喜.基于数据挖掘的金融时间序列预测分析与研究[D].大连:大连海事大学,2008,3
[26]康丽锋.基于数据挖掘的短期负荷预测方法研究[D].北京:华北电力大学,2006,12
[27]Watchdog Agent Tools Documentation[R].Cincinnati:Center for Intelligent Maintenance System,2007
[28]熊腾科.基于数据挖掘的金融预测模型[D].厦门:厦门大学,2006
[29]Jianbo Lin,Dragan Djurdjanovic.Similarity based method for manufacturing process perfbrmance prediction and diagnosis[J].Computers in Industry 58(2007) 558-566
[30]梁平,龙新峰,吴庚申.基于ARMA及神经网络的汽轮机振动故障诊断研究[J].热能动力工程,2007,Vol.22
[31]屈亚玲,周建中,刘芳等.基于改进的Elman神经网络的中长期径流预报[J].JOURNAL OF CHINA HYDROLOGY,2005,Vol,26
[32]神经网络模型及其MATLAB仿真程序设计[M].清华大学出版社,2005
[33]易文周,刘海燕,张超英,王强.VC++与MATLAB混合编程实现遗传算法在图像分割上的应用[J].计算机应用与软件,2008,Vol,23
[34]黄金明,武玉强,邢西深.VC++与MATLAB的混合编程研究[J].电脑开发与应用,2008,Vol,21