Abstract: The selective non-catalytic reduction (SNCR) denitrification system of the municipal solid waste incineration plant mainly relies on the preset static mapping table of NOx concentration and ammonia injection amount at the outlet of the bag filter, which adjusts the ammonia injection amount by real-time detection of parameters such as flue gas temperature and flow rate and checking the table. Due to the lack of adaptability to the dynamic characteristics of the operating conditions, the control stability is poor. Based on it, an efficient composite control method for the SNCR denitrification system in domestic waste incineration plants is proposed. The K-means algorithm is used to iteratively update the clustering centers to cluster the typical operating conditions of the SNCR denitrification system. Based on the clustering results of typical operating conditions, the operating condition dataset is divided. The training process of the model is divided into initialization stage and online learning stage using an online sequential extreme learning machine (OS-ELM). The model parameters are adjusted to adapt to the dynamic characteristics of the denitrification system by combining real-time operating data of the denitrification system, and the NOx concentration at the outlet of the bag filter is predicted. Using NOx concentration and real-time flue gas parameters as state inputs, and ammonia injection amount adjustment as action space, a reinforcement learning algorithm based Actor network is used to generate ammonia injection strategy, and the intelligent agent dynamically optimizes ammonia injection control through online interactive learning. Further more, the control stability of the proposed method has been tested. The final test results show that when using the proposed method for composite control, the fluctuation rate of ammonia injection amount is 4.2%, which has a relatively ideal control effect.

Key words: domestic waste incineration plant, selective non-catalytic reduction, denitrification system, composite control, control stability