城市固废焚烧过程数据驱动建模与自组织控制  

Data-driven Modeling and Self-organizing Control of Municipal Solid Waste Incineration Process

作　　者：丁海旭 汤健 乔俊飞 DING Hai-Xu;TANG Jian;QIAO Jun-Fei(Faculty of Information Technology,Beijing University of Technology,Beijing 100124;Beijing Laboratory of Smart Environmental Protection,Beijing 100124)

机构地区：[1]北京工业大学信息学部,北京100124 [2]智慧环保北京实验室,北京100124

出　　处：《自动化学报》2023年第3期550-566,共17页Acta Automatica Sinica

基　　金：科技创新2030—“新一代人工智能”重大项目(2021ZD0112300);国家自然科学基金创新群体项目(62021003);国家科技重大专项(61890930);国家自然科学基金(62073006);北京市自然科学基金(4212032,4192009)资助。

摘　　要：城市固废焚烧(Municipal solid waste incineration,MSWI)是处置城市固废(Municipal solid waste,MSW)的主要手段之一.中国MSW来源范围广、组分复杂、热值波动大,其焚烧过程通常依靠人工干预,这导致MSWI过程智能化水平较低且难以满足日益提升的控制需求.MSWI具有多变量耦合、工况漂移等诸多不确定性特征,因而难以建立其被控对象模型并设计在线控制器.针对以上问题,提出了一种面向MSWI过程的数据驱动建模与自组织控制方法.首先,构建了基于多输入多输出Takagi Sugeno模糊神经网络(Multi-input multi-output Takagi Sugeno fuzzy neural network,MIMO-TSFNN)的被控对象模型;然后,设计了基于多任务学习的自组织模糊神经网络控制器(Multi-task learning selforganizing fuzzy neural network controller,MTL-SOFNNC)用于同步控制炉膛温度与烟气含氧量,其通过计算神经元的相似度与多任务学习(Multi-task learning,MTL)能力对控制器结构进行自组织调整;接着,通过Lyapunov定理对MTLSOFNNC稳定性进行了证明;最后,通过北京市某MSWI厂的过程数据验证了模型与控制器的有效性.Municipal solid waste incineration(MSWI)is one of the main means to dispose of municipal solid waste(MSW).MSW in China has a wide range of sources,complex components,and large fluctuations in calorific value.Its incineration process usually relies on manual intervention.This will lead to a low degree of intelligence in the MSWI process and it is difficult to meet the increasing control requirements.MSWI has many uncertain characteristics such as multivariable coupling and working condition drift,so it is difficult to build the model of controlled object and design the on-line controller.To solve the above problems,this paper proposes a data-driven modeling and self-organizing control method for MSWI process.Firstly,the model of controlled object based on multi-input multi-output Takagi Sugeno fuzzy neural network(MIMO-TSFNN)is constructed.Secondly,a multi-task learning self-organizing fuzzy neural network controller(MTL-SOFNNC)is designed to synchronously control the furnace temperature and flue gas oxygen content,which can self-organize the structural parameters of the controller by calculating the similarity of neurons and the ability of multi-task learning(MTL).Meanwhile,the stability of MTLSOFNNC is proved by Lyapunov theorem.Finally,the effectiveness of the model and controller is verified by the process data of an MSWI plant in Beijing.

关 键 词：城市固废焚烧 多任务学习 自组织控制 数据驱动建模 模糊神经网络 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置] X799.3[自动化与计算机技术—控制科学与工程]