With the completion of ultra-low emission transformation of thermal power plants, problems such as increased costs and excessive ammonia injection have arisen. Modeling and optimization of power plant operation data through machine learning has become an important means to solve the above problems. This article reviews the commonly used machine learning algorithms and their application scenarios in reducing nitrogen oxides. In terms of algorithm, the main algorithms of data preprocessing, modeling prediction and parameter optimization and their applicability to nitrogen oxides removal are summarized. The research directions of multi-operating condition data preprocessing method and the construction method of the objective function in multi-objective optimization are proposed. For the application level of the machine learning methods, such as low nitrogen combustion in the furnace, optimization of SCR denitration system, and comprehensive energy saving and consumption reduction of the whole system, the implementation methods and corresponding effects are summarized. The future research directions of long-period dynamic modeling control and multi-power plant joint modeling have prospected.

Micro-mixing combustion is a kind of potential technology to reduce pollutant emissions, and the microtube structure as a basic unit has significant influence on combustion characteristics. Taking the single microtube model burner as the object, the computational fluid dynamics (CFD) method is employed to analyze methane micro-mixing combustion characteristics with different microtube diameters, rear-section lengths, and chamber-to-tube area ratios, under the air-preheated and pressurized condition and the atmospheric condition. The numerical simulation results show that, the flame length increases obviously under the atmospheric condition, and the increase rate is greatly affected by the microtube diameter and the area ratio. With the increase of microtube diameter, the flow and temperature fields in the flame chamber are similar, and the dimensionless flame length does not change significantly. However, when the microtube diameter is smaller, the change of flame length is smaller between two working conditions, and the flexibility is better. With the increase of microtube rear-section length, the turbulence intensity at the microtube outlet reduces greatly, and the flame length increases obviously, but the change trend of flame length is consistent under two working conditions. With the increase of chamber-to-tube area ratio, the flame length decreases at first and then increases. When the area ratio is 4.0~9.0, the flame length is relatively short, and the combustion performance is relatively good. The results have reference value for the gas turbine combustor, and also have certain reference significance for free jet flame.

The variable operating condition of thermal power units makes the data show multi-modal characteristics, which leads to the decrease of prediction accuracy of the regression soft sensor model based on shallow network structure. An improved BP neural network (back propagation neural network, BPNN) soft sensor method is studied. Firstly, the original data features are extracted by using the strong deep learning ability of stacked sparse autoencoder (SSAE), and then the extracted features are analyzed by BPNN. The experimental results show that, the mean square error of the SSAE+BPNN soft sensor method is 0.135 8×10^–3 and the square correlation coefficient is 0.983 2. It is proved that its prediction accuracy and generalization ability are significantly better than those of BPNN. It is applied to the soft sensor of carbon content in fly ash of a flexible peak-shaving 660 MW ultra-supercritical generator set, and the average relative error of the prediction results is 0.91%, the overall relative error is less than±5%, indicating the method has good engineering application value.

With high power generation efficiency and low carbon emission, (ultra) supercritical unit is the main type in China's development of thermal power units. Optimization on feed water operating condition of the (ultra) supercritical unit is of great guiding significance. The optimal control indexes of different feed water operating conditions of (ultra) supercritical units are investigated by in-situ sampling test, and the influence of centralized sampling on feed water quality is researched. Moreover, the intelligent control system for feed water operating condition is developed. The results show that, the optimal target value of feed water conductivity is 6.7 μS/cm under AVT(O) condition, the optimal target value of feed water dissolved oxygen is 15 μg/L and the optimal target value of feed water conductivity is 3.0 μS/cm under OT condition. Long sampling pipeline of feed water will cause the deposition of iron corrosion products, and interfere with the judgment of unit corrosion. The chromates produced by adding oxygen to feed water is only related to the oxidation of the sampling tube and has nothing to do with the water vapor system.

Welding of dissimilar steels is a widely used connection method for super-heater tubes of boilers in power plants. Taking the TP304H/T22 dissimilar steel welded joint with backing plate at the superheater outlet of a thermal power unit as the research object, the formation mechanism and influence of cracks in the welded joint are analyzed through macro measurement, microstructure analysis, EDS energy spectrum detection and tensile test. The results show that, the differences in thermal expansion coefficients of the three substrates and stress concentration are the cause of the cracks of the dissimilar steel weld with backing plate and crack propagation in nickel base weld filler. The weld is the weak link in room temperature tension, while the fusion line and base metal are the fracture parts in high temperature tension.

Voltage transformer (TV) interturn short circuit at generator outlet will cause unbalanced voltage in system, resulting in zero sequence stator ground fault protection of generator fundamental. The external characteristics of TV fault are of great significance to stable operation of the unit. In order to understand the external characteristics of TV fault, based on Maxwell and Simplorer software, a field circuit direct coupling finite element model for the electromagnetic conversion process of single-phase voltage transformer is established, which can realize data exchange between field circuits at each calculation step. In view of the large number of turns of TV primary winding, which is difficult to model directly, winding grouping technology is proposed. An experimental platform is designed and built to verify the accuracy of the simulation model. Based on the calculation results of finite element eigenvalues, the electrical and magnetic eigenvalues of TV in different operating states are analyzed, and the analytical model of TV inter turn fault based on winding grouping technology is further given. The simulation results show that, the TV turn-to-turn fault model is accurate and effective in analyzing the primary current, circulating current and other fault characteristic quantities.