With the increasing demand for flexible operation of power plant boilers

frequent variable-load operation leads to a wide range of fluctuations in pollutant concentrations and flue gas parameters. Modeling of key indicators such as single pollutant or flue gas parameter can no longer meet the actual demand

so it is necessary to consider the coupling of multiple key indicators for synergistic predictive modeling. Based on the historical operation data of coal-fired power plants

feature extraction is performed through kernel function mapping

and a long short-term memory neural network with a hard parameter sharing structure is constructed for multi task prediction modeling. The prediction model is optimized using uncertainty loss methods. The experimental results show that

the proposed prediction model exhibits high prediction accuracy under variable load conditions

and the prediction errors for the key metrics involved in this study are reduced by 25.5%

41.8%and 4.7%

respectively. The proposed method is capable of predicting several key indicators of utility boilers under variable load conditions

which can assist power plants to achieve pollution control and optimize the thermal efficiency of combustion

and provide technical support for intelligent operation of power plants.