Research On Boiler Combustion Optimization Method Based On Deep Reinforcement Learning

Combustion optimization technology can realize efficient and low pollution operation of coal-fired boilers by adjusting the coal and the air distribution,which is of great significance to improve the level of energy conservation and emission reduction of power plants.However,it is difficult to apply traditional modeling and optimization control methods to the boiler combustion system with high-dimensional input and output dynamic characteristics.Therefore,in view of the object characteristics of the coal-fired boiler combustion system with front and rear wall hedging combustion,this paper constructs the optimization objective function and corresponding constraints on the basis of the economic predictive control framework.Then the optimal control quantities could be obtained through combining with the adaptive data-driven model and the self-learning optimization algorithm,so as to improve the boiler efficiency and reduce the NOx emission while maintaining the basic stability of reheat steam temperature.In the modeling of boiler combustion system,this paper constructs a multi-kernel function by linear weighting of multiple kernel functions based on the traditional least squares support vector machine firstly.Combined with the corresponding online update strategy,an online adaptive least squares support vector machine modeling method based on multi-kernel function is proposed.After that,this method is applied to the modeling of boiler combustion system.Aiming at the problem of high data complexity of boiler combustion system,an incremental modeling algorithm is proposed,which is combined with time-delay analysis and variational modal decomposition method.Finally,the time series similarity based on time warping-edit distance is selected as the evaluation index of multi-step prediction accuracy,and the online update strategy is adjusted accordingly.The simulation results show that the modeling algorithm proposed in this paper has higher accuracy in both single-step prediction and multistep prediction.In terms of optimization algorithm,firstly,on the premise of ensuring the safety of engineering application,this paper proposes a boiler combustion optimization strategy based on the deep deterministic policy gradient algorithm.Aiming at the time-delay characteristics of boiler combustion system,a kind of improved deep deterministic policy gradient algorithm based on gated recurrent unit is proposed.The simulation results show that the algorithm proposed in this paper has good dynamic optimization effect in different working conditions.In the application of boiler combustion optimization engineering,a modification scheme of control configuration is designed,and the development of boiler combustion optimization control software is completed based on Python.