Deep deterministic policy gradient-based combustion optimization method for coal-fired boiler

Boiler combustion optimization technology aims to improve boiler efficiency while reducing NOx emissions. However, the combustion optimization methods commonly used lack the self-learning capability and have low utilization of historical data. In this paper, a combustion optimization method based on the deep deterministic policy gradient (DDPG) algorithm is proposed. Firstly, dynamic modeling is realized based on online least squares support vector machine and multi-kernel learning, and the optimal kernel function weight coefficient and related kernel parameters are selected by an improved particle swarm optimization algorithm. Secondly, the DDPG algorithm is introduced to optimize the control variables. The simulation results demonstrate that the multi-kernel least squares support vector machine (MKLSSVM) model has better local fitting performance and global generalization performance. The DDPG optimizer and MKLSSVM predictor can effectively improve the boiler efficiency and reduce pollutant emissions while maintaining the stable reheat steam temperature.