Abstract: Accurate NO_x emission prediction significantly improves operation safety and reduces pollutant emissions in coal-fired power plants. Machine learning-based NO_x emission prediction models have the advantages of fast computational speed and high fitting accuracy. However, these methods lack interpretability and overly rely on training samples, leading to poor generalization ability under insufficient data scenarios. Therefore, this paper introduces a novel forecasting method for NO_x emission based on a physics-informed neural network (PINN). In PINN, the monotonic relationships between coal feed rates, oxygen levels, openings of separated overfire air (SOFA), and NO_x emission concentration are embedded into the neural network model, prompting the model to obey the mechanisms and effectively inhibiting the overfitting or underfitting issues. This enhancement boosts the model's accuracy under broad boiler operating conditions. Taking a 660 MW utility boiler as the research object, the results indicate that the proposed method significantly outperforms three traditional machine learning methods (random forest, support vector machine, and artificial neural network), reflecting superior prediction performance and generalization capacity. Moreover, this PINN model maintains adherence to the monotonic relationship between parameters even in unknown operating modes.