基于广义回归神经网络与遗传算法的煤灰熔点优化  

Combining general regression neural network and genetic algorithm to optimize ash fusion temperature

作　　者：石喜光[1] 郑立刚[2] 周昊[1] 陈习珍 邱坤赞[1] 岑可法[1] 

机构地区：[1]浙江大学热能工程研究所能源清洁利用国家重点实验室 [2]焦作工学院资源与材料工程学系,河南焦作454000 [3]大冶特殊钢股份有限公司动力公司,湖北黄石435001

出　　处：《浙江大学学报（工学版）》2005年第8期1189-1192,1242,共5页Journal of Zhejiang University：Engineering Science

基　　金：国家自然科学基金资助项目(50206018).

摘　　要：考虑固态和液态排渣锅炉对煤灰熔点的不同要求,采用广义回归神经网络建立了煤灰软化温度模型.神经网络的输入变量为7个,即煤灰中SiO2、Al2O3、Fe2O3、CaO、MgO、TiO2、Na2O&K2O的质量分数.以煤灰软化温度作为目标函数,采用遗传算法寻优计算获得当煤灰软化温度最高和最低时煤灰中氧化物的组成.广义回归神经网络仅需30个训练样本,最大和平均相对误差分别为21.8%和1.55%.优化结果表明,掺烧高钙煤或者向燃煤中添加石灰石等富含Ca的原料可以降低煤灰熔点;而增加Al2O3的质量分数可以提高煤灰熔点.Considering the different requirements of dry bottom furnace and wet bottom furnace for coal ash fusion temperature, general regession neural network （GRNN） was employed to model the relationship of ash softening temperature and the chemical composition of coal ash. The 7 input parameters of the neural network were the fractions of SiO2 ,Al2O3 ,Fe2O3 ,CaO,MgO,TiO2 ,Na2O ＆ K2O in coal ash. With ash softening temperature set as objective function, genetic algorithm （GA） was used to make a global optimization to find the suitable chemical compositions of coal ash corresponding to the maximum or minimum ash softening temperature. With 30 training sampies, the maximum and average relative prediction errors of GRNN were 2. 81% and 1.55%, respectively. The optimization results show that ash softening temperature can be decreased by adding coals with higher Ca content or limestone, while adding Al2O3 results in higher ash fusion temperature.

关 键 词：灰熔点 灰组分 广义回归神经网络 遗传算法 

分 类 号：TK222[动力工程及工程热物理—动力机械及工程]