基于运行数据和过程机理的电站锅炉精细化协同建模方法研究

With the increasingly severe environment pollution problem, the issues of efficiency and emissions of power station boiler become more highlighted. The optimization operating of the boiler becomes more and more urgent with frequently coal quality and load varying. The problem of efficiency and emissions of the ultra supercritical boiler is discussed in this project. On the basis of the unit operating mechanism and data-driven modeling theory, the precise collaborative modeling theory of optimizing indexes based on large noisy dynamic data and process mechanism is taken as the key problem. The missing values and outliers in operating data are handled by adaptive hierarchical clustering method and local nested probability method respectively. The effect of random noise in measured data on the desired optimization indexes is revealed by exploring its quantitative expression based on the data statistical characteristics. According to the timeliness of the measurement data, an adaptive updating strategy based on remember law is proposed for online modeling. The local model integrating strategy is discussed based on history data and the process mechanism at all conditions. And then a precise collaborative modeling method for operation optimization indexes is proposed. The results will provide sophisticated reference for precise modeling and state reconstruction of complex process systems, and also provide theory foundation for further improving the load suitability, efficiency of the ultra-supercritical boiler at all conditions, and the pollution emissions can be reduced simultaneously.

随着环境问题日益严峻，电站锅炉的效率和排放问题更加突出，开展煤质、负荷多变条件下机组的运行优化研究尤为迫切。本项目以超超临界电站锅炉变负荷运行过程中的效率和排放问题为切入点，在分析机组工艺过程和数据驱动建模理论的基础上，重点研究基于工业含噪大数据与过程机理的电站锅炉精细化协同建模方法。通过研究基于层次化自适应聚类的不完备数据填补方法和基于局部概率嵌套的异常值检测方法，对现场缺失数据和异常值进行处理；研究基于统计特征的工业数据随机噪声定量表达方法，探索测量数据随机噪声对系统优化指标的影响规律；根据系统测量数据的时效性，构建基于记忆模式的模型自适应更新策略；基于全工况运行机理和历史数据，研究局部模型的集成策略。最终形成面向运行优化指标的建模方法。为复杂过程系统精细化建模与状态重构提供一种途径，为进一步提高超超临界电站锅炉负荷适应性、全工况运行效率、降低污染排放提供理论基础。

随着我国新时期绿色发展理念的不断推进，电站锅炉运行的高效率和低排放要求更为迫切。本项目以超超临界电站锅炉变负荷运行过程中的效率和排放问题为切入点，在分析机组工艺机理和数据驱动建模方法的基础上，重点研究了基于工业含噪数据与运行机理的电站锅炉精细化协同建模方法。研究了基于层次化自适应聚类的不完备数据填补方法和基于局部概率嵌套的异常值检测方法，对现场缺失数据和异常值进行处理；研究了基于偏互信息（PMI）的特征选择方法，尝试研究基于支持向量机（SVR）、人工神经网络（ANN）、深度学习（LSTM）等方法的系统关键参数的数据驱动建模方法；根据系统测量数据的时效性，构建模型自适应更新策略，设计了基于支持向量机的在线建模算法；并研究了基于粒子群优化算法的模型参数优化问题，建立NOx、主蒸汽温度、汽温偏差等关键参数的数据驱动模型，为进一步提高电站锅炉负荷适应性、运行效率、降低污染排放奠定基础。