Research On NO_x Emission Prediction And Control Of Coal-fired Boiler

With the deep access to new energy generation system,the safe and stable operation of power grids is facing new challenges.It is imperative to improve the of capability of deeply regulating load of thermal power plant.Rapid and deep load changes can lead to drastic fluctuations in the combustion process,which in turn affects emission control.The establishment of a boiler NOx emission prediction model is a prerequisite and basis for subsequent denitrification control and combustion optimisation,and is of great significance for achieving deep load regulation and energy saving and emission reduction in thermal power plants.The paper focuses on the theme of NOx emission prediction and control research for coal-fired boilers from the following aspects.1.Various factors affecting NOx emissions are analysed.The infrared temperature measurement devices are installed on the main combustion zone of the boiler to obtain the main combustion zone temperature.Then the relationship between the temperature parameters and the NOx concentration in the furnace is explored.Various data analysis methods based on principal component analysis,variable importance projection,and mutual information are investigated.And the automatic selection of NOx prediction model variables based on mutual information is achieved.2.There are different time lags between the thermal process variables,which can be calculated based on the mutual information method.An ensemble strategy based on random subspace is proposed,including the data partition method and the model ensemble mode.Firstly,subspaces are constructed according to the component information extracted by partial least squares.Then,the deep belief network is used as a sub-model.Finally,a back propagation neural network is developed for model combination.The ensemble DBN model has been used to model the NOx emission prediction of a 660MW boiler.The simulation results show that the ensemble DBN model can fully exploit the nonlinear mapping relationship between input variables and NOx concentration by using various learning learners.3.Considering the time-dependence of NOx-concentration in coal-fired boilers,recurrent neural networks are applied to NOx emission prediction from a temporal perspective.The dynamic prediction of NOx emission concentration is achieved by making full use of the time-series information of the data.To address the problem of the degradation of model prediction performance due to increasing input length,a long short-term memory network prediction algorithm based on the attention mechanism is proposed.Simulation results show that the algorithm can achieve adaptive attention to data features and improve the feature extraction ability of the network.4.From a spatial perspective,a graph convolutional neural network prediction algorithm combining variable correlations is proposed to address the complex non-linear relationships between variables and their mutual coupling.The algorithm effectively fuses the information of features between different variables by constructing the adjacency relationship between variables to form graph data and obtaining the feature adjacency matrix.Then a NOx prediction model is constructed based on the graph convolutional network,according to the topology information and historical operation data of the coal-fired boiler.The results show that the graph convolutional network based on the variables correlation can effectively fuse the correlation information between auxiliary variables,thus improving the prediction performance of the model.5.For the characteristics of large inertia,large time delay and strong disturbance of SCR flue gas denitrification system in coal-fired power plants,a LADRC-based cascade control scheme for SCR denitrification is designed.The predicted NOx concentration at the SCR inlet is introduced into the control system as a feedforward signal to realize the composite control of SCR system.The simulation results show that the LADRC-based compound control scheme provides more timely adjustment of the ammonia injection and better control performance.