Multi objective dynamic optimization control of combustion process in large-scale power plant boilers based on deep reinforcement learning

In the combustion process control of large power plant boilers, dynamic programming algorithms are usually used to solve the optimal control objectives. However, when the problem scale is large, it will face the problem of dimensional disaster, which can lead to inaccurate control results. Based on this, a multi-objective dynamic optimization control method for the combustion process of large-scale power plant boilers based on deep reinforcement learning is proposed. Taking into account the requirements of combustion efficiency and NOx content in flue gas, define the optimization objective function for the combustion process of large power plant boilers. Combining deep neural networks with reinforcement learning algorithms to solve multi-objective optimization functions and select the optimal combustion control strategy. Finally, by utilizing the PID closed-loop control structure, the optimal control strategy is implemented to achieve optimized control of the boiler combustion process under dynamic operating conditions. The experimental results show that the maximum control error of NOx content in flue gas exhibited by this method is 2.7mg·m 3, and the maximum control error of boiler combustion efficiency is 0.02%, significantly improving the control accuracy of boiler combustion process.