A boiler combustion multi-objective optimization method based on deep fusion of reinforcement learning and NSGA-II

By deeply integrating reinforcement learning with NSGA-II, a combustion characteristic proxy model was trained and boiler operating variables were optimized, solving the problems of slow convergence speed and poor real-time performance in boiler combustion optimization, and achieving high efficiency, stable combustion efficiency and low NOx emissions.

CN122389985APending Publication Date: 2026-07-14SHAANXI GUOHUA JINJIE ENERGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-05-26
Publication Date
2026-07-14

AI Technical Summary

Technical Problem

Existing boiler combustion optimization methods suffer from slow convergence speed, poor real-time performance, and a tendency to get trapped in local optima under strong constraints and dynamic operating conditions, making it difficult to stably balance combustion efficiency and NOx emissions.

Method used

A method based on deep integration of reinforcement learning and NSGA-II is adopted to predict combustion characteristics by training a deep neural network model. The SAC strategy is combined for offline pre-training to construct a policy network and a value network, adaptively adjust the cross-variation parameters, and introduce dominance to guide environmental selection to optimize boiler operating variables.

Benefits of technology

It significantly improves the convergence speed and diversity of boiler combustion optimization, enhances adaptability to dynamic operating conditions, and reduces the duration of NOx exceedance and efficiency loss.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122389985A_ABST
    Figure CN122389985A_ABST
Patent Text Reader

Abstract

The present application relates to a kind of coal-fired boiler combustion multi-objective optimization method based on deep fusion of reinforcement learning and NSGA-II, the method first utilizes historical operation data to train deep neural network as combustion characteristic proxy model, to quickly predict boiler efficiency and NOx emission;Subsequently, strategy network and value network are pre-trained off-line by SAC algorithm, to provide intelligent decision basis for subsequent optimization;Finally, these networks are deeply fused in the framework of NSGA-II, i.e. strategy network is used to generate high-quality initial population, and the value network evaluation value is used to dynamically adaptively adjust the crossover and mutation operator parameters, while the dominance guiding mechanism is introduced in the environment selection, so that in the boiler combustion scene of strong constraint, high dimension and time-varying, a set of Pareto optimal solution set is efficiently output, which takes into account high combustion efficiency and low NO x X emission, significantly improving convergence speed, solution set quality and adaptability to dynamic working conditions.
Need to check novelty before this filing date? Find Prior Art