Prediction of NOx emission from SCR zonal ammonia injection system of boiler based on ensemble incremental learning

In order to avoid the over-injection of ammonia and reduce ammonia slip, the zonal ammonia injection transformation of selective catalytic reduction (SCR) system has been carried out in many units. Due to the limitations of the measurement principle and economic cost, there is a large lag in the measurement of NOx emission from the SCR zonal ammonia injection system. Abnormal values are also found in the measurement during the purge period. Therefore, an ensemble incremental learning method is proposed for estimating NOx emission. Firstly, a multi-feature ensembled broad learning system (BLS) is designed for constructing individual learners under various working conditions. Then, a residual analysis method based on the Yeo-Johnson transformation and 3σ criterion is developed for conditional judgment of incremental learning, so as to balance the model's speeds of learning new knowledge and forgetting old knowledge. Finally, an ensemble incremental learning based on the working condition discriminator and residual analysis is designed for estimating outlet NOx and areal NOx of the SCR zonal ammonia injection system. Based on the operational data of a 660 MW unit, simulation and experiment are carried out. It is also proved that the proposed method has good fitting accuracy and certain anti-data drift ability.