低热值煤发电过程污染物的预测与智能控制

In recent years, the power generation technology by circulating fluidized bed (CFB) boilers burning the low calorific value coal has been widely applied due to its outstanding benefits in environment and economy aspects. The implementation of the new ultra-low emission policies compels the single mode of pollutants removal to be a jointly removing mode of CFB units, increases the operation cost and heightens the control requirements. This project will investigate the prediction and intelligent control of combined pollutants removal processes for CFB units. The main research contents are as follows: (i) the predicting problems of pollutants generating processes are studied by using an orthogonal experiment method and multi-layer perceptron networks ; (ii) the system modeling problems of combined pollutants removal processes are dealt with using the modeling methods of support vector regression, etc; (iii) the multi-object optimization problem of the combined pollutants removal process is formulated by finding suitable technical and economical indexes; (iv) the intelligent control algorithms, such as model predictive control based on economical optimization and fuzzy control based on dynamical setting values, are designed by taking the characteristics of multivariable, large delay, and strong coupling into consideration; (v) a set of intelligent control integrated schemes is designed by the obtained research results and is validated in some test units. In the condition of ensuring safe, stable operation and the ultra-low emission requirement, this project brings about economical operation of the power generation unit, improves the automatic and intelligent level of the combined pollutants removal processes, and provides the theoretical and technical support for similar types of power generation units in service and new-built.

近年来，低热值煤循环流化床（CFB）发电技术以其突出的环境和经济效益得到了广泛应用。超低排放政策的实施迫使CFB机组污染物脱除工艺由单一模式变为联合脱除，运行成本随之增加，控制要求相应提高。本项目针对CFB机组开展污染物联合脱除过程的预测与智能控制研究。研究内容包括：采用正交试验方法与多层感知器网络，研究全工况污染物生成的预测问题；采用支持向量回归等建模方法，研究污染物联合脱除过程系统建模问题；基于技术性能和经济性能指标，建立污染物联合脱除多目标优化的数学描述；针对污染物联合脱除过程多变量、大滞后、强耦合等特性，设计基于经济优化的模型预测控制和基于动态设定值的模糊控制等智能控制算法；利用研究成果设计一套智能控制集成方案并进行工业验证。在保证机组安全稳定运行、满足超低排放的基础上，实现系统经济运行，提高污染物脱除过程自动化、智能化水平，为同类在役机组和新建机组提供理论依据和技术支撑。

本项目针对CFB机组开展低热值煤发电过程SO2、NOx联合脱除污染物预测与智能控制问题研究，研究内容包括：污染物生成预测与脱除过程系统建模、污染物联合脱除智能控制与经济优化、智能控制与经济优化的工业验证与集成化应用三个方面。针对试验机组炉内炉外联合脱硫系统、SNCR-SCR耦合脱硝系统开展现场扰动试验与数据采集处理，经过定性分析与算法研究确定了影响CFB机组SO2与NOx生成的主要因素；设计了一种基于变量选择的支持向量机（SVM）算法建立了SO2生成浓度的软测量模型；采用BP神经网络建立了NOx生成浓度的软测量模型；提出一种带有时滞的非线性滑动平均模型（MA）对耦合脱硝系统NOx生成过程进行非线性建模；采用自适应权重粒子群算法（APSO）对系统模型进行参数辨识，建立了耦合脱硝系统在三种典型工况下的动态模型；设计了一种基于罚函数法的改进型粒子群优化算法（SUMT-MPSO）得出污染物联合脱除系统在不同负荷下的最优运行成本；设计了一种基于预测模型的单节点神经网络（SNN）预测控制系统提高了联合脱硫控制系统的稳定性与鲁棒性；提出一种用于变工况运行的多模切换DMC-PID串级预测控制方法解决了污染物联合脱除系统在变工况时的最优控制问题；设计了一种基于多目标遗传算法（GA）优化的动态设定值算法对污染物联合脱除系统的设定值进行动态寻优，申请发明专利1项；研究了带有时滞时间、时变、非线性系统的智能控制算法，以及多率采样系统、具有时滞的非周期采样系统和时变时滞不确定系统的稳定性问题；设计了一种基于专家模糊的脱硝系统分层优化自适应智能控制方法，授权发明专利1项，并成功应用于试验机组脱硝控制系统；设计了一种用于火电机组烟气脱硝的模型预测控制装置，和一种燃煤电厂湿法脱硫效率软测量系统，申请实用新型专利2项。项目研究成果可为CFB机组污染物联合脱除系统的污染物预测与智能控制提供理论依据和技术支撑。