Abstract: In order to ensure the prediction precisionand reduce theprediction time, a soft sensor model based on gradient boosting decision tree（GBDT） and particle swarm optimization（PSO） for nitrogen oxides（NOx） at the inlet of the system in coal-fired power（SCR） plantsis proposed.Firstly, the objective function is designed based on the model’s prediction accuracy and running time.Secondly, the five most important features are selected through the Pearson correlation coefficient between the input features and target, the reduced dimensional data is standardized and divided intotraining dataset and testing dataset for the prediction model.Then, a GBDT model is constructed, and the model parameters are optimized through the objective function.Finally, in order to reduce the optimization timea PSO algorithm is used to optimize the hyperparameters of prediction model, so the soft sensoris achieved.The simulation results show that the average generalization error of the PSO-GBDT model for SCR inlet NOx concentration is less than 5%.At the same time, the model structure is simply, and the running time is less than the GBDT model.