The direction of economic development and policy orientation has promoted the upgrading of coal-fired boilers in thermal power plants towards the direction of intelligence. The combustion efficiency of coal-fired boiler is an important indicator to measure the operating status of boiler. In order to meet the requirements of real-time calculation of boiler thermal efficiency，the following methods are used to calculate the boiler efficiency with the help of the daily measurement data of the power plant：Firstly，the corresponding combustion and operation characteristics of the boiler were analyzed; Secondly，according to the extracted features，the preprocessing methods of eliminating outliers，steady state discrimination，and similarity processing were carried out to generate better training samples. Finally，the neural network algorithm improved by genetic algorithm was used to establish the calculation model among the boiler exhaust temperature，fly ash carbon content and coal ash content. The calorific value of the coal into the furnace was calculated by using the proportional relationship between the calorific value of coal and the theoretical air volume，and the calculated value was used in the inverse balance calculation model of the boiler thermal efficiency. The calculation results show that the predicted value of the neural network model can meet the requirements of engineering calculation. The calculated exhaust gas temperature，fly ash carbon content and coal ash content can be used in the calculation of boiler efficiency to realize real-time dynamic boiler efficiency calculation. The change of the calculated boiler efficiency is approximately the same as that of the actual evaporation change. When the actual evaporation capacity of the boiler decreases，the efficiency of the boiler will decrease. When the actual evaporation capacity of the boiler is maintained above 60% of the rated evaporation capacity，the boiler efficiency is easily maintained at a high level.