数据驱动的块结构复杂非线性工业过程动态模型化研究

The complex industrial process usually has the well-known characteristics of multiple variables, nonlinear and strong coupling, and it is also seriously affected by many uncertain factors,which makes its modeling is very difficult.In view of these problems, the block-oriented nonlinear dynamic model description and identification method for complex nonlinear industrial process is proposed systematically in this project together with the dynamic data-driven technology research.The noise's impact is considered and a new multiple-input multiple-output nonlinear Hammerstein-Wiener model is structured based upon the fusion of the mechanism information, which assures the characteristics of industrial process is fully captured by the newly constructed block-oriented nonlinear dynamic model.A combined multi-source signal is researched and designed then to solve the identification problem of block-oriented nonlinear dynamic model and the separation problem of parameter estimation of the series blocks.A new identification method refraining the noise for block-oriented complex nonlinear industrial process is majorly researched,including the series blocks modelling method based on data-driven and integrated modeling technology, new intelligent separation algorithm and block parameters identification algorithm combining sensitivity analysis of key variables, colored noise modeling of block-oriented nonlinear dynamic system through compensation technique and its performance analysis and comparison by using the stochastic process theory.The research will provide a strong theoritical and practical basis to improve the utilizaiton of advanced optimization and control technology for chemical and the other complex industrial process in China.

针对复杂工业过程具有多变量、变量间非线性和强耦合的特点，并且受多种不确定因素干扰导致难以建模的问题，本项目拟结合动态数据驱动模式，系统地提出复杂工业过程块结构非线性动态模型描述和辨识新方法:考虑过程噪声的影响，研究全新融合机理信息的多输入多输出Hammerstein-Wiener非线性系统模型描述方法，确保块结构非线性动态模型能反映复杂工业过程的特性；研究和设计组合式多源激励信号，以解决块结构非线性动态模型的可辨识性问题和各模块的参数估计分离问题；重点提出能够抑制干扰的块结构复杂非线性工业过程辨识新方法，包括采用动态数据驱动理论和集成建模技术的串联模块建模方法、结合灵敏度分析的全新智能分离算法和模块参数辨识算法、利用补偿技术的块结构非线性动态系统有色噪声建模方法等，并利用随机过程理论分析和比较提出方法的性能。本项目将为提高我国化工等复杂工业过程的先进优化和控制技术水平奠定理论与方法基础。

针对复杂工业过程具有多变量、变量间非线性和强耦合的特点，并且受多种不确定因素干扰导致难以建模的问题，本项目结合动态数据驱动模式，系统地提出了复杂工业过程块结构非线性动态模型描述和辨识新方法。.(1)研究基于二进制-随机复合信号源的Hammerstein-winner模型辨识方法，提出两种基于二进制-随机复合信号源的神经模糊Hammerstein-Wiener模型，通过组合式信号实现Hammerstein-Wiener模型中静态非线性环节和动态线性环节的分离, 同时设计了神经模糊模型参数的非迭代优化算法, 将研究结果拓广到分段非线性系统。该方法适用范围广、计算简单、辨识精度高、对具有分段非线性特性的过程具有较好的拟合精度。.(2)研究含过程噪声的块结构模型，采用二进制-随机复合信号实现块结构模型的可辨识性和参数估计分离问题，提出块结构模型的三种噪声补偿方法：辅助模型递推最小二乘法、偏差补偿递推最小二乘算法、辅助模型多新息随机梯度算法。.(3)研究多输入多输出Hammerstein模型的辨识问题，通过可分离信号来解决多输入多输出Hammerstein模型的可辨识性问题和串联环节的参数估计分离问题。深入分析过程输入为可分离信号时静态非线性环节的不变特性，并采用随机理论给出数学证明，提出一种基于可分离信号源的多输入多输出Hammerstein模型辨识方法。.(4)考虑更广的一类输入激励信号，研究含过程噪声的Hammerstein模型辨识方法。将Bussgang定理推广到含过程噪声的多输入多输出Hammerstein模型中，通过可分离信号实现静态非线性环节和动态线性环节辨识问题的分离；采用相关分析法补偿过程噪声，估计动态线性环节参数，推导出基于最小二乘法的静态非线性环节参数辨识算法。考虑Hammerstein输出误差滑动平均系统，基于辅助模型辨识思想和相关分析法估计Hammerstein输出误差滑动平均系统中的静态非线性环节和动态线性环节参数，有效补偿噪声信号的干扰。.(5)以信号在某种特殊状态下激发出的性质为依据，提出基于可分离信号的Hammerstein-Wiener模型辨识方法，进一步研究噪声干扰下的Hammerstein-Wiener模型，提出基于递推广义增广最小二乘法的Hammerstein-Wiener模型辨识方法，有效补偿噪声的干扰。