基于遗传神经网络敏感度分析的飞灰含碳量测量方法  

Prediction method of unburned carbon content in fly ash based on genetic neural network with sensitivity analysis

作　　者：冯旭刚[1] 钱家俊[1] 章家岩[1] 

机构地区：[1]安徽工业大学电气与信息工程学院,马鞍山243000

出　　处：《电子测量与仪器学报》2016年第7期1083-1089,共7页Journal of Electronic Measurement and Instrumentation

基　　金：安徽省教育厅自然科学重点项目(KJ2015A058)资助

摘　　要：飞灰含碳量是反映电站锅炉燃烧效率的重要指标,准确测量飞灰含碳量有利于监测和调整锅炉燃烧,降低煤耗,提高锅炉运行的经济性和安全性。分析了工程上普遍采用的人工取样、实验室化验方法以及在线监测仪器所存在的缺陷。研究了以软测量和信息融合技术为代表的人工智能方法的不足,采用遗传神经网络算法,对的连接权值、阈值和隐层节点个数进行了优化计算,以增强网络的泛化能力,并利用Grason敏感性分析算法,对影响飞灰含碳量的输入参数进行了筛选,实现了以较少的工况输入来有效地测量输出。在此基础上,通过实际样本进行仿真分析和对比,验证了所用方法具有更好的测量精度,最后通过实测数据的对比进一步验证了该结论。The unburned carbon content of fly ash is an important index reflecting the combustion efficiency of utility boiler. Measuring the carbon content in the fly ash accurately is beneficial to the detection and adjustment of boiler combustion, the reduction of the coal consumption and the improvement of the economy and security. The defects of manual sampling and on-line detecting instruments were analyzed in this paper. And the deficiencies of artificial intelligence such as soft sensor and information fusion technology were also studied. This paper used a genetic neural network optimized the initializing weights, thresholds and numbers of node in hidden layer of BP neural network. And the Garson method was also used to screen the input parameters that could influence the unburned carbon content of fly ash. The Garson method adopted in this paper can predict the outputs effectively by using less input parameters. The simulation result illustrates that the method is more accurate than the genetic neural networks and standard BP neural network.

关 键 词：飞灰含碳量 发电锅炉 BP神经网络 隐层节点 Garson方法 

分 类 号：TN081[电子电信—物理电子学] TM621[电气工程—电力系统及自动化]