【摘要】 针对火电机组锅炉燃烧过程中预测NO_x排放过程存在的非线性和时序性特点，提出一种基于核主成分分析(KPCA)和注意力机制(AM)的门控循环神经网络(GRU)氮氧化物预测模型。首先选用KPCA对模型的输入变量进行降维，消除冗余变量；其次，将筛选的变量数据作为GRU的输入，并采用网格搜索优化GRU的超参数；最后，引入AM计算权值，实现区分输入特征功能，提高NO_x预测模型精度。通过某330 MW电站锅炉实际数据对AGRU预测模型仿真验证，并将AGRU模型、GRU模型和BP神经网络模型的预测结果进行对比。结果表明：基于AGRU的NO_x预测模型的均方根误差和平均绝对误差较BP神经网络和GRU模型均有减少，可精准预测非线性时序燃烧过程的NO_x排放。更多还原

【Abstract】 Aiming at the nonlinear and sequential characteristics of NO_x emission prediction in the boiler combustion process of thermal power units, a NO_x prediction model of gated recurrent unit(GRU) neural network based on kernel principal component analysis(KPCA) and attention mechanism(AM) was proposed. Firstly, KPCA was selected to reduce the dimension of the input variables of the model and eliminate redundant variables. Secondly, the filtered variable data was used as the input of GRU, and the grid search was used to optimize the superparameters of GRU. Finally, AM calculation weight was introduced to realize the function of distinguishing input characteristics and improve the accuracy of the NO_x prediction model. The AGRU prediction model was simulated and verified by the actual data of a 330 MW power plant boiler, and the prediction results of the AGRU model, the GRU model, and the BP neural network model were compared. The results show that the root mean square error and average absolute error of the NO_x prediction model based on AGRU are less than those of the BP neural network and GRU model, which can accurately predict the NOx emission in the nonlinear sequential combustion process.更多还原