基于支持向量回归的煤粉工业锅炉烟气氧含量预测

Oxygen content prediction of pulverized coal-fired industrial boiler flue gas based on SVR

锅炉氧含量优化对锅炉的效率和排放均有重要的影响。为了克服直接测量成本高、积灰后稳定性差的问题,提出使用间接测量法并基于支持向量回归模型以建立煤粉工业锅炉烟气的软测量模型。根据煤粉锅炉的燃烧工艺对现有的数据进行采集,通过F检验选择合适的变量,并对数据进行Z-score标准化;再基于控制变量、状态变量和两者合并,建立煤粉工业锅炉氧含量支持向量回归预测模型,并使用采集历史数据的70%对模型进行训练;最后使用采集历史数据的30%对模型的预测效果进行验证。结果表明,3种模型预测结果均符合真实值变化趋势,虽控制变量的调节及时显现在采集的数据中,但锅炉工况及氧含量需要一定的反应时间才可较为稳定地运行在新的状态,基于控制变量所预测得到的烟气氧含量与真实值差距较大,而状态变量与氧含量相对同步变化,因此,3种模型中基于状态变量的支持向量回归模型对烟气氧含量的预测最为准确,其可为工业煤粉锅炉燃烧系统优化提供指导。

The oxygen content optimization of boiler flue gas has important influence on boiler efficiency and emissions.Direct measuring has high costly price and unsteady trait after gathering dust.To overcome these shortcomings,indirect measuring was proposed.A soft measuring model was established for pulverized coal-fired industrial boiler flue gas based on SVR(Support Vector Regression).Data were collected according to pulverized coal-fired industrial boiler combustion process and appropriate auxiliary variables were chosen through F test.Data was normalized by the Z-score.Then,oxygen concentration support vector regression prediction model of pulverized coal-fired industrial boiler was established based on control variables,state variables and both of them,and the model was trained by 70%of history data.Finally,the prediction effect of the model was verified by 30%of history data.The simulation results on real data show that three kinds of SVR model are successful to predict the trend of real value.Although the regulation of the control variables appears in the collected data in time,the boiler working conditions and oxygen content still need a certain reaction time to run in new station.The model based on the control variables predicts the biggest gap between the flue gas oxygen content and the real value,and the state variables change synchronously with the oxygen content.Therefore,the support vector regression prediction model based on state variables predict the flue gas oxygen content most accurately in three models.It can provide guidance for the optimization of combustion system of industrial pulverized coal boiler.