【摘要】 新能源在全社会发电中占比不断增高，为了能够辅助电网进行削峰填谷，火电机组需要增加在低负荷范围内的运行时间，这会对选择性催化还原（selective catalytic reduction,SCR）的脱硝效果产生负面影响。因此，对SCR入口烟气温度（SCR gas temperature,SCRIGT）进行准确预测非常重要。首先，采用极限梯度提升（extreme gradient boosting,XGBOOST）模型，以电厂的运行参数作为输入，功率与SCRIGT的比值作为输出进行预测，然后计算得到SCRIGT。结果显示，对于两个不同锅炉类型的火电机组，平均绝对百分比误差（mean absolute percentage error,MAPE）分别为3.07%和2.49%。其次，采用不可知模型的局部可解释（local interpretable model-agnostic explanations,LIME）算法分析XGBOOST模型的预测结果，显示功率和功率与SCRIGT的比值之间存在线性关系，且R~2(R-squared)为0.994，基于此构建了一种可解释关系模型进行SCRIGT预测。最后，比较分析显示，对于试验中的两个机组，可解释关系模型预测结果的MAPE分别改进至0.68%和0.97%。更多还原

【Abstract】 The proportion of new energy in the power generation in the whole society is constantly increasing.To shave peaks and fill valleys,the thermal power plants must increase their operating time in the low-load range,which has a negative effect on the denitrification of selective catalytic reduction(SCR).Therefore,accurate prediction of SCR gas temperature(SCRIGT)at the inlet is crucial.Firstly,using the operating parameters of the power plant as input and the ratio of power to SCRIGT as output,the XGBOOST model was used for prediction to obtain SCRIGT.The results show that for two different types of thermal power units,the average absolute percentage error(MAPE)is 3.07% and 2.49%,respectively.Then,the local interpretable model agnostic explanations(LIME)was applied to explain the prediction result of XGBOOST model,which finds a linear relationship between load and load·SCRIGT-1 with R-squared of 0.994,based on which an interpretable relationship model was constructed.Finally,the comparative analysis shows that the mean absolute percentage errors of the interpretable relationship model have been improved to 0.68% and 0.97%,respectively.更多还原