针对燃煤电厂普遍缺少煤炭元素分析数据的现状，以我国商品煤煤质数据库中的3 000余条煤质数据为基础，分别采用线性回归、BP神经网络、SSABP神经网络模型对煤炭工业分析数据进行建模，预测煤炭元素分析含碳量，进而从原料侧计算燃煤碳排放，3种模型对于煤炭元素分析含碳量预测的相对误差分别为8.40%,2.51%，1.30%。选取某百万机组燃煤电厂平稳负荷、波动负荷、升负荷、降负荷4种典型工况，从原料侧通过上述3种模型开展电厂燃煤连续碳排放计算，并与电厂烟气侧检测碳排放值进行比较。结果表明：线性回归、BP神经网络、SSABP神经网络模型可以较好地推测元素分析含碳量。3种模型在平稳负荷的低负荷、中负荷、高负荷3种工况下，与锅炉烟气侧测量所得燃煤碳排放的均方根误差RMSE分别为0.35，0.08，0.07；0.87，0.37，0.09；0.23，0.19，0.17。在升负荷、降负荷、波动负荷工况下，3种模型计算值的均方根误差RMSE分别为1.00，0.84，0.71；1.43，1.24， 0.73；1.33，1.15，0.93。以某电厂典型工作日为例，3种模型对日总碳排放计算值与烟气检测法获得的碳排放相对偏差分别为12.28%，5.52%，0.22%。SSABP神经网络模型煤质预测和碳排放计算结果与烟气侧测量值偏差最小。

In view of the general lack of coal ultimate analysis data in coalfired power plants, based on more than 3 000 pieces of quality data in China′s commercial coal quality database, a linear regression model, a BP neural network model and a sparrow search algorithm (SSA) optimized BP neural network model were established. The coal proximate analysis data were fitted in three models to predict the carbon content of the coal ultimate analysis, which was further applied to calculate the carbon emission of coal combustion from stock side, and the relative errors of the carbon content of the coal ultimate analysis predicted by three models were 8.40%, 2.51% and 1.30%, respectively. A 1 000 MW power plant unit under four typical load conditions of stationary load, fluctuating load, load up and load down was selected to calculate the continuous coalfired carbon emissions through the proposed three models from stock side, and the carbon emission value was compared with that detected from flue gas side of power plant. The results show that the proposed linear regression, BP neural network and SSABP neural network models can predict the carbon content of coal ultimate analysis well. The root mean square error (RMSE) of carbon emissions of coal combustion obtained from flue gas side under three working conditions of low, medium and high stationary loads are 0.35, 0.08, 0.07 and 0.87, 0.37, 0.09 as well as 0.23, 0.19, 0.17. The RMSEs of computational values of three models under three working conditions of load up, load down and load fluctuation are 1.00, 0.84, 0.71 and 1.43, 1.24, 0.73 as well as 1.33, 1.15, 0.93. Taking a typical working day of a power plant as an example, the relative deviations between the total daily carbon emissions calculated by three models and the carbon emissions obtained by flue gas detection method are 12.28%, 5.52% and 0.22%, respectively. SSABP neural network model has the smallest deviation of the coal quality prediction and carbon emission calculation result from the measured values on the flue gas side.

TQ533

国电南瑞南京控制系统有限公司科技项目（524609220030）