TY - JOUR
T1 - A synergetic control framework for pulverizing systems of coal-fired power plants
T2 - Performance optimization based on DRL and DMC
AU - Jing, Zheng
AU - Hu, Xiaoyang
AU - Wang, Shuai
AU - Qiu, Binbin
AU - Wang, Zhu
AU - Chen, Weixiong
AU - Wang, Jinshi
N1 - Publisher Copyright:
© 2026 Elsevier Ltd.
PY - 2026/8
Y1 - 2026/8
N2 - Enhancing operational flexibility has emerged as a pivotal technical requirement in the transformation and modernization of coal-fired power plants. Among the various subsystems, the pulverizing system constitutes a primary factor constraining overall unit flexibility. Under transient load fluctuations, conventional PID control strategies frequently encounter limitations. Such systems often lack the responsiveness required to adapt effectively to rapid variations in outlet parameters. To address these challenges, a data-mechanism hybrid model of the pulverizing system was established. Based on the model, this study attempted to employ the advanced control algorithms to improve pulverizing system performance. Specifically, the effectiveness of a deep reinforcement learning (DRL)-compensated fuel control framework was evaluated, and the utilization of the deep deterministic policy gradient (DDPG) algorithm with additive compensation was proposed. The results showed significant improvements in control accuracy due to the capability of data representation and intelligent decision-making. Compared to the PID control system, the mean absolute error (MAE) of pulverized coal output using DDPG algorithm was reduced by 18%, while the MAE of unit power output decreased by 22%. The performance of the dynamic matrix control (DMC) algorithm was evaluated for the coupled regulation of mill hot and cold air valves. Relative to the PID control system, the multi-parameter DMC strategy significantly reduced the overshoot of the mill outlet temperature by 24% and the response time by 38%. Furthermore, the response time of the primary air flow rate was reduced by 18%, and the response time for suspended coal storage within the suspension zone decreased by 22%. In conclusion, the control strategy significantly strengthens the control performance of the pulverizing system and attains a notable enhancement in the operational flexibility of the coal-fired power plant.
AB - Enhancing operational flexibility has emerged as a pivotal technical requirement in the transformation and modernization of coal-fired power plants. Among the various subsystems, the pulverizing system constitutes a primary factor constraining overall unit flexibility. Under transient load fluctuations, conventional PID control strategies frequently encounter limitations. Such systems often lack the responsiveness required to adapt effectively to rapid variations in outlet parameters. To address these challenges, a data-mechanism hybrid model of the pulverizing system was established. Based on the model, this study attempted to employ the advanced control algorithms to improve pulverizing system performance. Specifically, the effectiveness of a deep reinforcement learning (DRL)-compensated fuel control framework was evaluated, and the utilization of the deep deterministic policy gradient (DDPG) algorithm with additive compensation was proposed. The results showed significant improvements in control accuracy due to the capability of data representation and intelligent decision-making. Compared to the PID control system, the mean absolute error (MAE) of pulverized coal output using DDPG algorithm was reduced by 18%, while the MAE of unit power output decreased by 22%. The performance of the dynamic matrix control (DMC) algorithm was evaluated for the coupled regulation of mill hot and cold air valves. Relative to the PID control system, the multi-parameter DMC strategy significantly reduced the overshoot of the mill outlet temperature by 24% and the response time by 38%. Furthermore, the response time of the primary air flow rate was reduced by 18%, and the response time for suspended coal storage within the suspension zone decreased by 22%. In conclusion, the control strategy significantly strengthens the control performance of the pulverizing system and attains a notable enhancement in the operational flexibility of the coal-fired power plant.
KW - Control strategy
KW - Deep reinforcement learning (DRL)
KW - Dynamic matrix control (DMC)
KW - Operational flexibility
KW - Pulverizing system
UR - https://www.scopus.com/pages/publications/105043076501
U2 - 10.1016/j.applthermaleng.2026.132052
DO - 10.1016/j.applthermaleng.2026.132052
M3 - 文章
AN - SCOPUS:105043076501
SN - 1359-4311
VL - 302
JO - Applied Thermal Engineering
JF - Applied Thermal Engineering
M1 - 132052
ER -