Multi-Model Predictive Control for SCR Denitrification Systems in Coal-Fired Power Plants Based on Hybrid Data-Driven and Model Ensemble

In order to address the difficulty of controlling the ammonia spraying amount quickly and precisely in the selective catalytic reduction (SCR) denitrification system of a coal-fired power plant, a multi-model predictive control approach based on hybrid data drive and model ensemble is proposed. Firstly, the maximum information coefficient (MIC) and Elastic Net algorithm are adopted to reconstruct the modeling data for time delay and select variables to improve the model's prediction accuracy. Secondly, the kernel fuzzy C-mean (KFCM) clustering algorithm that introduces the introduction of Xie-Beni metrics is employed to work on the modeling samples. Thirdly, an ensemble model based on extreme gradient boosting (XGBoost), least squares support vector machine (LSSVM), and genetic algorithm optimized BP neural network (GA-BP) model is utilized to establish a prediction model for the SCR system. Finally, a multi-model prediction controller based on integrated sub-models and model switching strategy is designed and developed. The experimental results indicate that the multi-model predictive control method proposed in this paper can obtain higher control precision and reliability, which can provide guidance for the actual operation of the SCR system.