基于卷积神经网络的时序变换证据融合方法  

Evidence Fusion Method of Time Sequence Transformation Based on Convolutional Neural Network

作　　者：李小艳 田亮[1] LI Xiaoyan;TIAN Liang(School of Control and Computer Engineering,North China Electric Power University,Baoding 071003,China)

机构地区：[1]华北电力大学控制与计算机工程学院,河北保定071003

出　　处：《电力科学与工程》2025年第2期42-50,共9页Electric Power Science and Engineering

基　　金：国家重点研发计划资助项目(2022YFB4100400)。

摘　　要：证据理论在不确定性推理、数据融合、故障诊断领域应用较广泛,但对证据的时序同步性要求限制了其在工业中的应用。当电站锅炉磨煤机及制粉系统运行状态受机组负荷和煤质变化影响时,系统内不同对象的动态特性会导致信号间存在时序差异,及时准确获知其运行工况对优化设备运行、预防事故发生具有重要意义。研究了一种基于卷积神经网络的消除证据间时序差异的方法,首先通过卷积神经网络对磨煤机运行数据进行时序变换,得到消除时序差异的数据,再利用典型样本法构造信度函数分配,最后用证据理论进行融合得到诊断结果。实例分析表明该方法能够提前6min诊断出系统存在堵磨故障,取得较好的应用效果。Evidence theory is widely used in the areas of uncertainty reasoning,data fusion and fault diagnosis,but its application in industry is limited by the requirement of time synchronization of evidence.When the operation state of coal mill and pulverizing system of power plant boiler is affected by unit load and the change of coal quality,the dynamic characteristics of different objects in the system will lead to time sequence differences between signals,so it is of great significance to know its operation conditions in time and accurately for optimizing equipment operation and preventing accidents.A method based on convolutional neural network to eliminate the time sequence difference between evidences is studied.Firstly,the time sequence transformation of coal mill operation data is carried out by convolutional neural network to get the data that eliminates the time sequence difference,then the typical sample method is used to construct the reliability function assignment,and finally the diagnosis result is obtained by fusion with evidence theory.The case analysis shows that this method can diagnose blockage of coal mill 6 minutes in advance,and has achieved good application results.

关 键 词：证据理论 卷积神经网络 时序变换 磨煤机 故障诊断 

分 类 号：TK39[动力工程及工程热物理—热能工程] TK28