基于人工神经网络的燃煤汞排放预测模型  

Study on prediction model for mercury speciation of coal combustion flue gases by using ANN

作　　者：王凡[1] 刘宇[1] 王相凤[1] 张辰[1] 曹晴[1] 张凡[1] 邓双[1] 陈育武[2] 王立刚[2] 

机构地区：[1]中国环境科学研究院,北京100012 [2]北京科技大学,北京100083

出　　处：《环境化学》2014年第1期93-99,共7页Environmental Chemistry

基　　金：国家高技术研究发展计划(863计划)(2012AA06A11303);环保公益性行业科研专项(200909025;201009048)资助

摘　　要：利用GA-BP的人工神经网络算法建立燃煤汞排放预测模型,确定煤中汞含量、煤的发热量、煤中硫含量、煤中氯含量、挥发份含量、排烟温度作为输入矢量,元素态汞、氧化态汞和颗粒态汞3个因素作为输出参数,通过对20个燃煤锅炉汞排放形态的测试数据进行模型训练,结合实际测试数据和预测数据对误差来源进行了分析.通过对3个样本进行验证,分析人工神经网络的实际预测效果.研究结果表明,训练与预测的精度都是符合汞排放预测实际要求的,预测精度达0.895,分析表明利用人工神经网络建立预测模型可对燃煤汞排放进行预测.A mercury emission prediction model of GA-BP was developed and improved based on traditional BP neural networks, in which mercury content of coal, calorific value, sulfur content of coal, chlorine content of coal, volatile content, and flue gas temperature had been evaluated and selected as the input characteristic variants, and 3 mercury speciations including elemental mercury, divalent mercury and particulate mercury were set as outputs. Analysis results of 20 coal-fired boilers had been used as training input samples, and source of training error had been discussed. 3 samples were used for testing the predicting model, and predicting accuracy of the prediction model also evaluated. The results showed that the results of training and predicting were greatly coordinated with the actual measurement results, and the training correlation efficiency was as high as O. 895. It is deduced that the GA-BP is achievable for prediction of mercury speeiation.

关 键 词：燃煤烟气 预测模型 汞形态 

分 类 号：X701[环境科学与工程—环境工程] TP183[自动化与计算机技术—控制理论与控制工程]