Abstract: The combustion mechanism and flue gas purification mechanism of Municipal Solid Waste Incineration (MSWI) process are complex and difficult to be described by accurate mathematical models.The intelligent control and operation optimization algorithms studied offline is difficulty to be verified.Aiming at these problems,the data-driven MSWI whole process model based on XGBoost serial and parallel ensemble was established.On the basis of describing the current situation of typical grate furnace control in China,the input of combustion process model in furnace was reduced based on empirical cognition.Then,XGBoost was used to construct a serial model of combustion process in furnace.Next,a parallel model of flue gas treatment process was constructed based on input feature selection with mutual information.Finally,the proposed model was debugged by using the step-by-step progressive training strategy.The effectiveness of the model was verified by actual operational data,which could provide support for insight into the internal mechanism of MSWI process and verification of intelligent control and operation optimization algorithms.

Key words: municipal solid waste incineration, data-driven, whole process model, XGBoost, mutual information, progressive training

摘要： 针对城市固废焚烧（MSWI）过程燃烧机理与烟气净化机制复杂难以采用数学模型刻画、离线智能控制与运行优化算法难以验证等问题,构建了能够体现工艺顺序特性的基于XGBoost串并联集成的数据驱动MSWI全流程模型。首先,在描述目前国内典型炉排炉控制现状的基础上,基于经验认知对炉内燃烧过程模型的输入进行约简处理;接着,采用适应工业数据特性的XGBoost构建炉内燃烧过程串行模型;然后,基于互信息选择输入特征构建基于XGBoost的烟气处理过程并行模型;最后,采用逐阶段递进式训练策略对所提出的MSWI全流程模型进行调试。通过实际运行数据仿真验证了模型的有效性,为洞悉MSWI过程的内在机理和验证智能控制与运行优化算法提供了支撑。

关键词: 城市固废焚烧, 数据驱动, 全流程模型, XGBoost, 互信息, 递进式训练