陈鸿伟 梁占伟# 赵 斌 高明非 赵争辉 张志远

(1.华北电力大学能源动力与机械工程学院，河北 保定 071003；2.华北理工大学冶金与能源学院，河北 唐山 063009)

调峰燃煤机组选择性催化还原烟气脱硝的建模预测优化*

陈鸿伟1梁占伟1#赵 斌2高明非2赵争辉1张志远1

(1.华北电力大学能源动力与机械工程学院，河北 保定 071003；2.华北理工大学冶金与能源学院，河北 唐山 063009)

调峰燃煤机组的运行对选择性催化还原烟气脱硝(SCR-DeNOx)系统的性能提出了新的挑战。通过比较BP神经网络(BP)算法和遗传算法优化BP(GABP)算法的直接模型和间接模型，得到出口NOx浓度和氨逃逸浓度预测的最佳模型。结果表明，对出口NOx浓度和氨逃逸浓度预测的最佳模型均为GABP算法的间接模型。该模型不仅精度高，可以适应不同的负荷，而且可以同时预测出口NOx浓度和氨逃逸浓度。对出口NOx浓度预测的相对误差均在±10%以内，均方根误差和平均绝对百分比误差分别仅为2.76mg/m3和4.49%；对氨逃逸浓度预测的相对误差基本都在±50%以内，均方根误差和平均绝对百分比误差分别为0.13mg/m3和24.29%。

调峰燃煤机组 选择性催化还原 脱硝NOx氨逃逸 神经网络算法

Abstract: The selective catalytic reduction dinitration (SCR-DeNOx) systems are facing new challenges since coal-fired peak regulation units ran. The direct and indirect models by BP neural network algorithm (BP algorithm) and genetic algorithm modified BP algorithm (GABP algorithm) were applied to predict outlet NOxconcentration and ammonia escaping concentration in SCR-DeNOxsystem. Results showed that the best model for both outlet NOxconcentration and ammonia escaping concentration prediction was the indirect model by GABP algorithm. This model was accurate and able to fit different loads. Moreover,the model could predict outlet NOxconcentration and ammonia escaping concentration simultaneously. The relative error,root-mean-square error and mean-absolute-percentage error were within ±10%,2.76 mg/m3and 4.49% for outlet NOxconcentration,respectively;and those almost within±50%,0.13 mg/m3and 24.29% for ammonia escaping concentration,respectively.

Keywords: coal-fired peak regulation units； selective catalytic reduction； denitration； NOx； ammonia escaping； neural network algorithm

近年来，受国内经济形势下行以及新能源发电并网等因素的影响，很多地区供电负荷峰谷差增大。据统计，京津冀地区峰谷差已达到30.6%[1]。单靠抽水蓄能发电及30万kW的燃煤机组调峰已不能满足电网调峰需求，60万kW及以上的燃煤机组参与深度调峰已成常态化。调峰燃煤机组的运行对选择性催化还原烟气脱硝(SCR-DeNOx)系统的性能提出了新的挑战。……