基于数据融合的目标模式识别技术及应用

发布时间：2018-03-31 06:39

本文选题：目标模式识别　切入点：非平稳过程　出处：《华北电力大学》2017年硕士论文

【摘要】：目标模式识别在工业设备及系统状态监测、故障诊断、优化控制、协同调度等方面获得广泛应用。但常用如主元分析、聚类分析、证据理论、支持向量机等方法本质上都立足于时间域,只能对离散域内的信息进行处理,仅可实现平稳过程的目标模式识别。现场存在大量非平稳过程的目标模式识别问题,直接应用以上方法分析时域内信号时,则会因信号动态时间上的差异导致分析结果出现巨大误差,容易造成误判和漏判。受电网负荷指令频繁变化、煤质波动加剧等因素影响,火电机组动态变工况运行所占时间的比例大大增加。课题研究思路为,将时域内信号处理方法与离散域内信息融合方法相结合,实现对动态过程的目标模式识别。并分别进行了以下尝试:(1)将能够反映信号变化特征的信息熵理论应用于机组AGC(自动发电控制)指令模式识别方面;(2)采用多尺度滤波器对变化速度不一致的信号进行分解,选择相同的时间尺度下的信号进行D-S融合处理,并将此方法应用于中速磨煤机“临界堵塞”故障判断方面;(3)利用软测量技术构造出静态精度好和动态响应速度快的中间信号并进行动态补偿,保证原始信号时间尺度一致后再进行判断,并将此方法应用于供热机组供热季内安全运行模式识别方面。经过机理分析、仿真分析、现场运行数据分析和实际调试后,应用结果表明此类方法能够有针对性的解决一些现场问题,获得了较好的实际效果。课题研究在非平稳过程的目标模式识别方面作出了有益的探索。
[Abstract]:Target pattern recognition has been widely used in industrial equipment and system condition monitoring, fault diagnosis, optimal control, cooperative scheduling and so on. However, it is commonly used such as principal component analysis, cluster analysis, evidence theory, etc. Support vector machine (SVM) is essentially based on time domain. It can only process the information in discrete domain and can only realize the target pattern recognition of stationary process. There are a lot of non-stationary process target pattern recognition problems in the field. When the above method is used directly to analyze the signals in the time domain, there will be huge errors in the analysis results due to the difference in the dynamic time of the signals, which will easily lead to misjudgment and missing judgment, which will be affected by the frequent changes of power network load orders and the aggravation of coal quality fluctuations. The proportion of time occupied by the dynamic off-condition operation of thermal power units is greatly increased. The research idea is to combine the signal processing method in time domain with the information fusion method in discrete domain. The target pattern recognition of dynamic process is realized. The following attempts are made: 1) the information entropy theory, which can reflect the characteristics of signal variation, is applied to the pattern recognition of AGC (automatic Generation Control) instruction of generating set. (2) Multi-scale is adopted. The filter decomposes the signal with inconsistent speed. D-S fusion processing is performed by selecting signals of the same time scale. The method is applied to the judgment of "critical blockage" fault of medium speed coal mill. The intermediate signal with good static precision and fast dynamic response is constructed by soft sensor technology and dynamically compensated. The method is applied to the safe operation pattern recognition of heating units in the heating season. After mechanism analysis, simulation analysis, field operation data analysis and actual debugging, The application results show that this method can solve some field problems and obtain good practical results.
【学位授予单位】：华北电力大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.4;TM621

【参考文献】